CN105915919A

CN105915919A - Trans-plane prediction

Info

Publication number: CN105915919A
Application number: CN201610412836.1A
Authority: CN
Inventors: 海纳·基希霍弗尔; 马丁·温肯; 海科·施瓦茨; 德特勒夫·马佩; 托马斯·维甘徳
Original assignee: GE Video Compression LLC
Current assignee: Fraunhofer Gesellschaft zur Forderung der Angewandten Forschung eV; GE Video Compression LLC
Priority date: 2010-04-13
Filing date: 2010-04-13
Publication date: 2016-08-31
Anticipated expiration: 2030-04-13
Also published as: CN105915921B; CN105933715A; CN105915919B; CN105915921A; CN105872563A; CN105915923B; CN105872563B; CN105915924A; CN105915918B; CN105872562A; CN105933715B; CN105872562B; CN105915920A; CN105915918A; CN105915922B; CN105915923A; CN105915920B; CN105933716A; CN105915922A; CN105915924B

Abstract

The application relates to trans-plane prediction. The requirement of notifying a decoder of trans-plane prediction information by a signal can cause an extra overhead, but in order to reduce redundancies, a higher rate-distortion ratio can be obtained by utilizing a mutual relationship among encoding parameters of different planes. Particularly, a plurality of planes can be respectively determined whether to be subjected to trans-plane prediction. Moreover, or optionally, in consideration of one secondary plane, determination can be carried out by using block as unit.

Description

Across planar prediction

The application is divisional application, the Application No. 201080067394.2 of its parent application, filing date in April, 2010 13 days, invention entitled " across planar prediction ".

Technical field

The present invention relates to the encoding scheme for scene image different spaces sample intelligence component planar, Mei Geping Face includes the message sample array of such as video or still image.

Background technology

In image and Video coding, image or the specific sample array set for this image are typically dissected into block, This block and specific coding parameter association.Image is typically to be made up of multiple sample arrays.Additionally, an image also can associate additionally Aid sample array, this sample array (such as) shows transparent information or depth map.The sample array of one image (includes assisting sample This array) also can assemble one or more so-called plane group, each plane group is by one or more sample array herein Composition.The plane group of one image can encode independently, or, if this image associates more than one plane group, then utilize Encode from the prediction of other plane group of same image.Each plane group is typically dissected into multiple block.This block (or The corresponding block of sample array) it is to be predicted by prediction in image prediction or image.Each block can have different size And can be square or rectangle.One image is divided into multiple block can be fixed by grammer, or can (at least in part) in place Stream internal signal notice.The predefined size that the syntactic element signalisation often sent segments for block.This syntactic element Can be shown that whether and how a block is subdivided into smaller block and the coding parameter being associated, such as, be used for predicting purpose.Pin Whole samples to a block (or corresponding block of sample array), the decoding of the coding parameter being associated is in certain mode Show.In this example, the whole samples at a block are to use identical Prediction Parameters ensemble prediction, and this Prediction Parameters is all in this way Benchmark index (mark reference picture in encoded image set), kinematic parameter (show that a reference picture is current with this The measured value of the block motion between image), show the parameter of interpolation filter, intra-prediction mode etc..Kinematic parameter can be by The motion vector with a horizontal component and a vertical component represents, or is represented by order motion parameter, such as includes six The affine motion parameters of component.Be likely to more than one particular prediction parameter sets (such as benchmark index and kinematic parameter) be with Single block is associated.In the case of this kind, for each set of this particular prediction parameter, produce for this block (or sample number The corresponding block of group) single medium range forecast signal, and final prediction signal is by including superposition medium range forecast signal One combination is set up.Corresponding weighting parameters and possibly, also includes that a constant offset (adding to this weighted sum) can be for a figure Picture or a reference picture or a reference picture collection are combined into fixing, or it is included in the Prediction Parameters set for corresponding block In.Difference between original block (or corresponding block of sample array) and its prediction signal, also referred to as residual signals, this is poor Generally it is transformed and quantifies.Often, two-dimensional transform is applied to this residual signals (or corresponding sample number for this residual error block Group).For transition coding, the block (or corresponding block of sample array) of particular prediction parameter sets has been used to execute Divided further before adding conversion.Transform blockiis can be equal to or less than the block for prediction.It is also likely to be, a transform blockiis Including the more than one block for predicting.Different transform blockiis can have different size, and transform blockiis can represent square or square Shape block.After the conversion, gained conversion coefficient is quantified, it is thus achieved that so-called conversion coefficient level.Conversion coefficient level and prediction If parameter and in the presence of, subdivided information is coded by entropy.

In image and video encoding standard, what grammer was provided an image (or a plane group) is subdivided into block Probability be extremely limited.Be typically only capable to define whether (and the most how) have define a block of size in advance can be thin It is divided into smaller block.Lifting an example, maximum resource block size H.264 is 16 × 16.16 × 16 blocks are also referred to as macro zone block, At first step, each image is divided into macro zone block.For each 16 × 16 macro zone block, signal notifies whether it is encoded into 16 × 16 blocks, or two 16 × 8 blocks, or two 8 × 16 blocks, or four 8 × 8 blocks.If 16 × 16 blocks are subdivided into four Individual 8 × 8 blocks, then each 8 × 8 block may be encoded as 8 × 8 blocks, or two 8 × 4 blocks, or two 4 × 8 blocks, Or four 4 × 4 blocks.Show to be divided into the small set probability of block to have in current image and video encoding standard is excellent Putting is that the side information rate for signalisation subdivided information can keep less, but has the drawback that for this encrypted communication pre- Survey the bit rate needed for parameter quite big, describe in detail after a while.The side information rate of signalisation information of forecasting the most generally represent for Notable a large amount of total bit rates of one block.When this side information reduces, code efficiency increases, such as can be relatively large by using Block size realizes side information and reduces.The actual image of video sequence or image are by the arbitrary shape with special properties Object forms.Lifting example, this object or object part are to be its feature with unique texture or unique motion.Usual identical prediction Parameter sets can be applicable to this object or object part.But object bounds generally and misfit large-scale prediction block (such as, according to H.264 16 × 16 macro zone blocks) possible block border.

Encoder generally determines that segmenting (in limited kind of probability set) causes the minimum of particular rate distortion cost measuring Change.For arbitrary shape of object, substantial amounts of block of cells so may be caused.And due to this block of cells it is and needs the one of transmission Prediction Parameters set is associated, therefore side information rate becomes most of total bit rate.But due in block of cells several still Represent same target or the district of an object part, therefore the Prediction Parameters to multiple gained blocks is identical or very much like.

In other words, an image segmentation piece into relatively miniature part together or piece block together or block substantially affect code efficiency and Encoder complexity.Such as outline above, an image is subdivided into multiple smaller block and allows the space of coding parameter more finely to set, mat This allows this coding parameter to be more preferably adapted to image/video material.On the other hand, coding parameter is set to notice with more fine granularity The required side quantity of information of the decoder setting value about needing adds all more high load capacities.Furthermore, notably encoder (further) Spatially segmentation image/video becomes any degree of freedom of block, the coding parameter setting value amount increasing severely possible, and thus generally Make the search for the coding parameter setting value causing optimal ratio/distortion tradeoff more difficult.

Summary of the invention

It is an object of the present invention to provide a kind of for scene image different spaces sample information component planar The encoding scheme of coding, each plane includes message sample array, and the program can obtain more preferable rate distortion ratio.

The potential conception of the present invention is, although the demand across planar prediction information signal notice to decoder will be caused Overhead, but the mutual relation that ought may utilize for the target that redundancy reduces between the coding parameter of Different Plane obtains more Good rate distortion ratio.

Tree root spatially it is placed according to the message sample array of an embodiment, first representation space sample information signal District, then basis is drawn from the multiway tree subdivided information of a data stream, by the subset in this tree root district of recursively repeated segmentation, and A subset to this tree root district of major general is divided into various sizes of smaller simple bonding pad.In order to allow with regard to rate distortion sense On, find out the good compromise between the excessive tiny segmentation with reasonable coding complexity and excessive thick segmentation, message sample array The maximum district size carrying out the tree root district that space is divided into is to include in this data stream and extract from this data stream in decoding end. Accordingly, decoder can include a withdrawal device, and it is configured to from the maximum district's size of data stream extraction and multiway tree subdivided information；One Subdivider, it is configured to would indicate that a message sample array space of spatial sampling information signal is divided into maximum district size Tree root district, and according to this multiway tree subdivided information, by least one subset in this tree root district by recursively this tree of multi-division This subset in root district and be subdivided into smaller simple connection different size district；And a reconstructor, it is configured with this segmentation And the message sample array from this data stream is reconstructed into more small-sized simple bonding pad.

According to an embodiment, data stream also contains up to tree root district subset and experiences the summit of recursively multi-division Formula level.By this way, the signalisation of multiway tree subdivided information becomes easier to and needs less bits of coded.

Additionally, reconstructor can be configured to depend on the granularity of middle segmentation, perform the one or many in following measures Person: at least inner estimation mode and in predictive mode determine be intended to use which predictive mode；From frequency domain transform to spatial domain, Perform and/or set the parameter across prediction；Perform and/or set for the parameter for interior prediction.

Additionally, withdrawal device can be configured to depth-first traversal order from the extraction of data stream and sectorized tree block The syntactic element that is associated of leaf district.By this kind of way, withdrawal device can develop the syntactic element in the most encoded neighbouring leaf district Statistics amount, it has ratio and uses the breadth first traversal higher probability of order.

According to another embodiment, use another subdivider according to another multiway tree subdivided information by this more small-sized list At least one subset of pure bonding pad is subdivided into and smaller simple bonding pad.First order segmentation can be used for performing letter by reconstructor The prediction of breath sample area, and second level segmentation can be used for performing converting again from frequency domain to spatial domain by reconstructor.Definition residual error Segmentation is subdivided into subordinate relative to prediction so that the coding less consumption position of total segmentation；On the other hand, by the residual error of subordinate gained Code efficiency is only had small negative effect by degree of restriction and the degree of freedom of segmentation, and reason is that major part has similar movement and mends The image section repaying parameter is bigger than the part with similar spectral nature.

According to still another embodiment, another maximum district size is to be contained in this data stream, another maximum district size definitions tree Root district size, the sub-district of this tree root is that at least one subset in the sub-district of this tree root is subdivided into according to the most more multiway tree subdivided information Front elder generation for the most smaller simple bonding pad is divided.So transfer to allow the independence of the on the one hand maximum district size of prediction segmentation Set, and on the other hand allow residual error segmentation, so can find out preferable rate/distortion tradeoff.

According to the still another embodiment of the present invention, data stream packets includes and is formed the second grammer unit of this multiway tree subdivided information The first syntactic element subset that sub-prime collection separates, wherein the combiner in this decoding end allows according to the first syntactic element subset And interblock space neighbouring multiway tree segmentation small-sized simple bonding pad obtain this sample array one in the middle of segmentation.Reconstructor Centre can be configured with segment and reconstruction sample array.By this mode, encoder is easier to the optimal ratio/mistake to find The spatial distribution of the very compromise character that effectively segmentation is adapted to message sample array.For example, if maximum district is a size of big, Then multiway tree subdivided information may become big and more complicated because of tree root district.But then, if maximum district is the least, the most more likely Be neighbouring tree root district be the information content about having similar quality so that this tree root district also can process together.Merge before filling up This gap between stating extremely, allows to segment close to optimized granularity by this.From encoder viewpoint, merge syntactic element and permit Permitted the most easily or encoding procedure less complex in computing, if reason is that encoder mistake uses the finest segmentation, then This error can be by encoder with post-compensation, by by setting merging syntactic element subsequently with or without only adjusting adaptation sub-fraction The syntactic element being set before merging syntactic element and setting is reached.

According to still another embodiment, maximum district's size and multiway tree subdivided information are that non-anticipating is thin for residual error segmentation Point.

It is used for processing the simple bonding pad of the multiway tree segmentation of a message sample array of representation space sample information signal One depth-first traversal order rather than breadth first traversal order be based on one embodiment use.By by using this degree of depth excellent First traversal order, each simple bonding pad has higher probability to have the neighbouring simple bonding pad being traversed so that when weight When building the most simple indivedual bonding pad, can be positively utilized about these information adjacent to simple bonding pad.

The tree root district of the hierarchy type size being first separated into zero layer level when message sample array is regularly arranged, the thinnest When at least one subset dividing this tree root district becomes different size of smaller simple bonding pad, reconstructor can use zigzag to sweep Retouch and scan this tree root district, be intended to the tree root district of subregion for each, with this leaf connected merely of depth-first traversal sequential processing District, the most more steps into next tree root district with zigzag scan order.Additionally, according to depth-first traversal order, have identical The simple leaf district connected of hierarchical level can also travel through according to zigzag scan order.So, maintenance has neighbouring simple The probability connecting leaf district increases.

According to an embodiment, although the labelling being associated with the node of multiway tree structure is suitable according to depth-first traversal Sequence arranges in proper order, but the coding in proper order of labelling uses probability estimation context, and it is for the identical stratum in multiway tree structure Being labeled as that formula level inside is associated with multiway tree structure node is identical, but for the different estate formula layer in multiway tree structure What the multiway tree structure node within Ji was associated is labeled as difference, by this allow between the context number to be provided good Compromise, and on the other hand, adjust the actual symbol statistics adapting to labelling.

According to an embodiment, the probability for the predetermined labels used estimates that context is also dependent on according to the degree of depth First traversing order labelling before this predetermined labels, and correspond to the district corresponding with this predetermined labels there is predetermined phase Each district to the tree root district of position relationship.It is similar to the conception that aforementioned aspect is potential, uses depth-first traversal order to ensure height Probability: the most encoded labelling also includes the labelling corresponding to the district that the district corresponding with this predetermined labels is adjacent, this knows available More excellently to adjust context for this predetermined labels.

Can be used for setting that can to correspond to be positioned at this predetermined labels for the labelling of the context of a predetermined labels relative Answer this labelling in Qu Shangqu and/or left district.Additionally, in order to select the labelling of context can be limited to and to belong to predetermined labels and be associated The labelling of the identical hierarchical level of node.

According to an embodiment, coding signalisation include summit formula level instruction and be not equal to summit The labelled sequence that the node of formula level is associated, each labelling shows whether associated nodes is intermediate node or child node, And derive from the labelled sequence of this data stream according to depth-first or the decoding in proper order of breadth first traversal order, skip summit The node of formula level and be automatically directed to identical leaf node, thus reduce encoding rate.

According to another embodiment, the coding signalisation of multiway tree structure can include the instruction of summit formula level. By this mode, the existence of labelling may be limited to the hierarchical level beyond summit formula level, for reason is always Eliminating there is the further subregion of block of summit formula level.

The leaf node of multiway tree segmentation is belonged to and without the subdivision in subregion tree root district in the segmentation of space multiway tree A part in the case of, the context for encoding subdivision labelling may be selected so that this context for equal greatly Being labeled as that community is associated is identical.

According to an embodiment, merging or packet that a simple bonding pad that this message sample array is segmented is favourable are Encode with little data amount.In order to reach this purpose, for simple bonding pad, defining a predetermined relative location relation, it is permitted Permitted to make a reservation for simple bonding pad for one and identify that making a reservation for simple bonding pad in inside, multiple simple bonding pads with this has predetermined phase Simple bonding pad to position relationship.In other words, if this number is zero, then may not there are for this inside this data stream One merging index of predetermined simple bonding pad.If additionally, making a reservation for simple bonding pad with this there is the list of predetermined relative location relation Pure bonding pad number is 1, then can use the coding parameter of this simple bonding pad, or can be used to predict for this predetermined simple connection The coding parameter in district and without any extra syntactic element.Otherwise, i.e. if making a reservation for simple bonding pad with this there is predetermined phase para-position The simple bonding pad number putting relation is more than 1, then can suppress the introducing of an extra syntactic element, even if identified with these The coding parameter that simple bonding pad is associated is each other for identical also multiple such.

According to an embodiment, if this coding parameter adjacent to simple bonding pad is each other, then one with reference to proximity identification Symbol is recognizable makes a reservation for, with this, the suitable subset that simple bonding pad has the simple bonding pad number of predetermined relative location relation, and When using this coding parameter or predicting that this uses this suitable subset when making a reservation for the coding parameter of simple bonding pad.

According to other embodiments, be would indicate that the spatial sampling of this two-dimensional information signal by recursively Multiple Segmentation as Local area and space are subdivided into multiple has one first grammer that different size of simple bonding pad is depending in this data stream Subset of elements and perform, then for the one second grammer unit in this data stream depending on being not connected with this first subset Sub-prime collection and interblock space, adjacent to simple bonding pad, obtain and will be subdivided into the simple connection being not connected with in the middle of this sample array District gathers, and it is combined for the plurality of simple bonding pad.The segmentation of this centre is used in when this sample array of this data stream reconstruction.As This allows for the optimization for should segmenting and does not the most have critical importance, and reason is that the most meticulous segmentation can be led to Merging after subsequently is compensated for.Additionally, segmentation allow with the combination merged to reach separately through recurrence Multiple Segmentation impossible Segmentation in the middle of reaching, therefore performs segmentation and the cascade merged by the syntactic element set that use is not connected with (concatenation) allow effectively or middle segmentation more preferably adjusts the actual content adapting to this two-dimensional information signal.With it Advantage compares, the overhead for indicating the extra syntactic element subset merging details to be caused be insignificant.

Accompanying drawing explanation

Below, for the following drawings, the preferred embodiments of the present invention are described, wherein:

Fig. 1 shows the block chart of the embodiment encoder according to the application；

Fig. 2 shows the block chart of the embodiment decoder according to the application；

Fig. 3 A to Fig. 3 C schematically shows the specific embodiment that quaternary tree is segmented, and wherein Fig. 3 A shows the first hierarchy type layer Level, Fig. 3 b shows the second hierarchical level, and Fig. 3 C shows third class formula level；

Fig. 4 schematically shows foundation one embodiment tree construction for the illustrative quaternary tree segmentation of Fig. 3 A to Fig. 3 C；

Fig. 5 A, Fig. 5 B schematically show the quaternary tree segmentation of Fig. 3 A to Fig. 3 C and have the index indicating indivedual leaf blocks Tree construction；

Fig. 6 A, Fig. 6 B figure schematically shows that the different embodiments of foundation represent the tree construction of Fig. 4 and four forks of Fig. 3 A to Fig. 3 C The binary string of tree segmentation or labelled sequence；

Fig. 7 shows a flow chart, and display foundation one embodiment is by the step performed by data stream withdrawal device；

Fig. 8 shows a flow chart, illustrates the function of the data stream withdrawal device according to another embodiment；

Fig. 9 A, Fig. 9 B display, according to the schematic diagram of an embodiment illustrative quaternary tree segmentation, emphasizes a predetermined block Neighbor candidate block；

Figure 10 shows a flow chart of the function according to another embodiment data stream withdrawal device；

Figure 11 schematically show according to an embodiment from the image in plane and plane group composition and illustrate say The coding that bright use adapts to across plane/predicts；

Figure 12 A and Figure 12 B schematically illustrates the sub-tree structure according to an embodiment and corresponding segmentation describes Succession scheme；

Figure 12 C and 12D schematically illustrate the sub-tree structure according to an embodiment describe use respectively use and The succession scheme of prediction；

Figure 13 shows that a flow chart, display realize the step performed by succession scheme according to an embodiment by encoder；

Figure 14 A and Figure 14 B shows once segmentation and subordinate segmentation, illustrates and implements association according to an embodiment Across-prediction one succession scheme probability；

Figure 15 shows that a block chart illustrates a kind of coding/decoding method associating this succession scheme according to an embodiment；

Figure 16 shows a schematic diagram, illustrates according to an embodiment in the scanning sequency in multiway tree segmentation subinterval, is somebody's turn to do Sub-district is to experience interior-prediction；

Figure 17 shows the block diagram of the decoder according to embodiment；

Figure 18 A to Figure 18 C shows a schematic diagram, illustrates the segmentation probability different according to other embodiments；

Specific embodiment

Later in the detailed description of accompanying drawing, occurring in the assembly between several pieces of drawings is to keep away with the instruction of common element numbers Exempt from these assemblies of repeat specification.The explanation of the assembly about presenting inside an accompanying drawing is also applied for wherein occurring indivedual groups on the contrary Other accompanying drawing of part, as long as deviation therein is pointed out in the explanation presented at this other accompanying drawing.

The encoder and decoder embodiments that Fig. 1 to Figure 11 is explained is started from additionally, be explained later.This accompanying drawing is presented Embodiment combines the many aspects of the application, if but individually the most excellent to encoding scheme internal implementation, so, the most subsequently attached Figure, embodiment is by aforementioned for short discussion indivedual aspects, and this embodiment is the enforcement described with regard to Fig. 1 and Figure 11 with different meaning representations The summary of example.

Fig. 1 shows the encoder according to embodiments of the invention.The encoder 1010 of Fig. 1 includes that a predictor 12, is residual Difference precoder 14, residual error reconstructor 16, data stream inserter 18 and a block dispenser 20.Encoder 10 be in order to One space-time sample information Signal coding is become a data stream 22.Temporal and spatial sampling information signal can be such as video, that is an image Sequence.Each graphical representation one image sample array.Other example of space time information signal such as includes by (time-during such as light Of-light) degree of depth image of camera shooting.The most notably a spatial sampling information signal can include that each frame or timestamp are many In an array, such as in the case of color video, color video such as includes that each frame one luma samples array is together with two Individual chroma sample array.The Temporal sampling being likely to the different components to information signal (that is brightness and colourity) may not With.In like manner, it is adaptable to spatial resolution.Video also can be attended by exceptional space sample information, the such as degree of depth or transparence information. But the focus of attention being hereinafter described will focus on the process of the one in this array carrys out the purport of the clearest understanding present invention, Then turn to the process of more than one plane.

The encoder 10 of Fig. 1 is configured to form data stream 22 so that the syntactic element in data stream 22 describes granularity and exists Image between full images and indivedual image sample.In order to reach this purpose, dispenser 20 is configured to be subdivided into each image 24 Different size of simple bonding pad 26.Hereinafter, this district will be simply referred to as block or sub-district 26.

As describing in detail after a while, dispenser 20 uses multiway tree segmentation that image 24 is subdivided into various sizes of block 26. In more detail, hereinafter with regard to the specific embodiment major part use quaternary tree segmentation of Fig. 1 to Figure 11 institute outline.As describing in detail after a while, Dispenser 20 is internal can include the cascade of subdivider 28 for image 24 being subdivided into aforementioned block 26, be then combiner 30 its Allow that this block 26 is combined into group to obtain between the segmentation defined without segmentation and subdivider 28 of image 24 Effectively segmentation or granularity.

Dotted line such as Fig. 1 illustrates, it was predicted that device 12, residual error precoder 14, residual error reconstructor 16 and data stream insert Device 18 is to operate in the image segmentation defined by dispenser 20.For example, as describing in detail after a while, it was predicted that device 12 use by For the individual small pin for the case district of prediction segmentation, the prediction segmentation that dispenser 20 is defined decides whether that this small pin for the case district should experience Have according to selected predictive mode in the image of the setting value of the corresponding Prediction Parameters in this small pin for the case district prediction or Across image prediction.

Residual error precoder 14 again then uses the sub-district of residual error of image 24 to encode the image provided by predictor 12 The prediction residual of 24.The syntactic element that residual error reconstructor 16 is exported from residual error precoder 14 rebuilds residual error, residual error reconstructor 16 also operate in the segmentation of aforementioned residual error.Data stream inserter 18 may utilize previous segmentation, that is prediction and residual error are segmented, and come sharp Determine that the insertion sequence between syntactic element and proximity relations are for by residual error precoder 14 and predictor 12 with such as entropy code The syntactic element of output inserts data stream 22.

As it is shown in figure 1, encoder 10 includes an input 32, this original information signal enters encoder 10 herein.One subtracts Musical instruments used in a Buddhist or Taoist mass 34, residual error precoder 14 and data stream inserter 18 and are compiled at the input 32 of data stream inserter 18 with described order Connect between the outfan of code data stream 22 output.Subtractor 34 and residual error precoder 14 are a part for prediction loop, and this is pre- Survey time road is to be surrounded by residual error reconstructor 16, adder and predictor 12, and these assemblies are to prelist in residual error with described order The outfan of code device 14 is connected with between the inverting input of subtractor 34.The outfan of predictor 12 is also coupled to adder 36 Another input.Additionally, predictor 12 includes the input being connected directly to input 32 and can include and another input End, it is also to be connected to the outfan of adder 36 via wave filter 38 in optional loop.Additionally, predictor 12 is in operation Period produces side information, and therefore the outfan of predictor 12 is also coupled to data stream inserter 18.In like manner, dispenser 20 includes One outfan its be coupled to another input of data stream inserter 18.

Have been described above the structure of encoder 10, after the further detail below of its operator scheme describes in detail such as.

It has been observed that dispenser 20 determines how to be subdivided into by image community 26 for each image 24.According to being intended to in advance The segmentation of the image 24 surveyed, it was predicted that device 12 determines how to predict individual cells for each community corresponding to this kind of segmentation. Predictor 12 exports the prediction of community to the inverting input of subtractor 34, and output is to the another input of adder 36, and The information of forecasting of the mode how from the previous coding part of video, reflection predictor 12 are obtained this prediction exports to data Stream inserter 18.

At the outfan of subtractor 34, so obtaining prediction residual, wherein residual error precoder 14 is based on also by splitting The residual error segmentation of device 20 defined processes this kind of prediction residual.As further described for Fig. 3 to Figure 10 below, The residual error segmentation of the image 24 used by residual error precoder 14 can be relevant to the prediction segmentation that predictor 12 is used, and makes each Predict that sub-district uses and become the less sub-district of residual error as the sub-district of residual error or further segmentation.It is possible that it is completely self-contained Prediction and residual error segmentation.

Sub-for each residual error district is experienced from space to the conversion of frequency domain by residual error precoder 14 by two-dimensional transform, is then Or peculiarly relate to the gained quantization of transform coefficients of gained transform blockiis, therefore distorted result comes from quantizing noise.Such as, Data stream inserter 18 can use (such as) entropy code that the syntactic element describing foregoing transformation coefficient is nondestructively encoded into number According to stream 22.

Residual error reconstructor 16 uses again re-quantization then for convert again, and conversion coefficient is re-converted into residual signals, its In this residual signals be at adder 36 internal combination by the prediction of subtractor 34 gained to obtain prediction residual, by this obtain The reconstruction part of the current image of outfan one or sub-district in adder 36.Predictor 12 can be used directly this reconstruction image subsection For interior-prediction, in other words, it is used for by being used for predicting that certain is pre-in the neighbouring prediction sub-district extrapolation rebuild by from previously Ce Zi district.But by by directly carrying out inside frequency domain from neighbouring spectrum prediction current sub-district frequency spectrum-prediction theory on also Belonging to may.

For interaction prediction, it was predicted that device 12 can use image version that is the most encoded and that rebuild, has passed through In selectivity loop, wave filter 38 filters.Wave filter 38 such as can include solution blocking filtering device and/or an adaptive filter, There is the transfer function being suitable for excellently forming aforementioned quantizing noise.

Predictor 12 selects Prediction Parameters, shows and predicts certain by being compared with the original sample within image 24 by use The mode in the sub-district of individual prediction.As describing in detail after a while, it was predicted that each is predicted that sub-district can include the instruction of predictive mode, such as by parameter In image-predict and across image prediction.In image-prediction in the case of, it was predicted that parameter also include be intended in-prediction prediction son The angle instruction that inner edge, district mainly extends；And in the case of image prediction, motion vector, moving image index and the highest Power motion transform parameter；And in image inside and/or in the case of both image predictions, be used for filtering reconstructed image sample Selective filter information, based on this measurable sub-district of current prediction.

As describing in detail after a while, the aforementioned segmentation defined by dispenser 20 substantially affect by residual error precoder 14, The maximum rate that predictor 12 and data stream inserter 18 can be reached/distortion ratio.In the case of segmentation is too thin, by predictor 12 The Prediction Parameters 40 being exported data stream 22 to be inserted needs too high code rate, but may by the prediction of predictor 12 gained Preferably, and by residual error precoder 14 residual signals to be encoded less may make it can be encoded by less bits.In segmentation In the case of the thickest, then it is suitable for reverse situation.Segment additionally, aforementioned thinking is applicable to residual error the most in a similar manner: image Using the finer grain conversion of individual transform block, result causes reducing for the complexity of operation transform and the conversion of result gained Spatial resolution increase.In other words, the sub-district of less residual error allows the frequency spectrum in the content within the sub-district of indivedual residual errors to distribute more For unanimously.But, spectral resolution lowers, and the ratio of notable coefficient and not notable coefficient (that is being quantified as zero) is deteriorated.In other words, Conversion granularity must adjust and adapt to local image content.Additionally, independent of the positive effect of more fine particle size, more fine particle size rule Ground increases the side quantity of information needed and indicates the segmentation selected by this decoder.As describing in detail after a while, aftermentioned embodiment pair Encoder 10 provides extremely effective adjustment to adapt to segment to content of information signals to be encoded, and it is by instruction data stream inserter Subdivided information is inserted data stream 22 by 18 carrys out the segmentation of signalisation decoding end to be used in.Details shows as follows.

But before the segmentation with further detail below definition dispenser 20, put up with according to the decoder of embodiments of the invention Fig. 2 describes in detail with further detail below.

The decoder of Fig. 2 is to indicate and include withdrawal device 102, dispenser 104, residual error weight with reference number 100 Build wave filter 112 and a selectivity postfilter 114 in device 106, adder 108, predictor 110, selectivity loop. Withdrawal device 102 receives encoded data stream at the input 116 of decoder 100, and extracts subdivided information from this encoded data stream 118, Prediction Parameters 120 and residual error data 122, these information are exported to image segmentating device 104, prediction by withdrawal device 102 respectively Device 110 and residual error reconstructor 106.Residual error reconstructor 106 has an outfan and is connected to the first input end of adder 108.Add Another input of musical instruments used in a Buddhist or Taoist mass 108 and outfan thereof are coupled to a prediction loop, filter in this prediction loop in selectivity loop Ripple device 112 and predictor 110 are from the bypass path of the outfan of adder 108 to predictor 110 with described sequential series, directly Connect and be similar to connection between adder described in Fig. 1 36 and predictor 12 above, in other words, one in image-prediction and Another one is for across image prediction.In the outfan of adder 108 or selectivity loop, the outfan of wave filter 112 is connectable to The outfan 124 of decoder 100, reconstruction information signal e.g. exports to transcriber herein.Selectivity postfilter 114 can It is connected to guide the visual quality that the eye impression of reconstruction signal at outfan 124 is improved in the path of outfan 124.

It sayed in outline, the assembly 16,36 of similar Fig. 1 of effect of residual error reconstructor 106, adder 108 and predictor 110 and 12.In other words, the same operation emulating earlier figures 1 assembly.In order to reach this purpose, residual error reconstructor 106 and predictor 110 By Prediction Parameters 120, and the segmentation indicated according to the subdivided information 118 deriving from withdrawal device 102 by image segmentating device 104 is carried out Control, carried out with predictor 12 or determine that the same way carried out predicts the sub-district of this prediction, and such as residual error precoder The mode of 14, remaps received conversion coefficient with same particle sizes.Image segmentating device 104 be dependent on again subdivided information 118 with The method of synchronization is rebuild by the segmentation selected by dispenser 20.Withdrawal device can use subdivided information to control data pick-up, the most just Context selects, adjacent to decision, probability estimation, the anatomy etc. of data stream grammer.

Previous embodiment can be carried out some deviations.Some deviation will describe in detail later with regard to performed by subdivider 28 Merging performed by segmentation and combiner 30 is addressed, and other deviation is put up with Figure 12 to Figure 16 subsequently and explained.Without any In the presence of obstacle, all these deviations all individually or can apply the detailed description part to Fig. 1 and Fig. 2 above in subset. For example, dispenser 20 and 104 determined prediction are segmented and are only determined the residual error segmentation of each image.It is likely on the contrary In being respectively directed to selectivity loop, wave filter 38 and 112 determines filtering segmentation.Other prediction segmentation or the segmentation of other residual coding Independence is the most unrelated or has dependence.Additionally, determine that segmentation may not be carried out based on frame by frame by these assemblies.The most right The segmentation that certain frame is carried out can reuse or be used in the following frame of certain number, the new segmentation of transfer the most subsequently.

Thering is provided relevant image to be partitioned in the further detail below in sub-district, first be explained later is that focus is concentrated The subdivided portions being responsible for is estimated at subdivider 28 and 104a.Then describe combiner 30 and combiner 104b is responsible for carrying out Merging treatment program.Finally, describe across plane adaptation/prediction.

It is different size of many that the mode of subdivider 28 and 104a segmentation image makes an image may be partitioned into as being likely to be of Individual block is used for image or the prediction of video data and residual coding.It has been observed that image 24 can be used as one or more image Sample value array.In the case of YUV/YCbCr color space, the such as first array can represent luminance channel, and another two numbers Group represents chrominance channe.These arrays can have different dimensions.All array can be grouped into one or more plane group, and each is put down Face group is made up of one or more continuous levels so that each plane is included in one and only one plane group.After Literary composition is applicable to each plane group.First array of one specific plane group can be referred to as an array of this plane group. Possible array subsequently is subordinate array.The block segmentation of array can be carried out based on quaternary tree way, as describing in detail after a while. Subordinate array block segmentation can segmentation based on an array and lead and calculate.

According to aftermentioned embodiment, subdivider 28 and 104a is configured to an array is partitioned into multiple equal sizes Square block, hereinafter referred to as tree block.When use quaternary tree time, tree block the length of side be typically 2 multiple, such as 16, 32 or 64.But for completeness, notably use other type of tree and binary tree or have any number of sheets purpose tree all to belong to possibility. Additionally, the filial generation number of tree can be depending on the level of tree and depends on which kind of signal of this tree representation.

The most as before, sample array also can represent the out of Memory beyond video sequence, such as depth map respectively Or light field.Simple and clear for asking, the focus being explained later is the typical example focusing on quaternary tree as multiway tree.Quaternary tree is The tree of four filial generations is just had at each internal node.Each tree block one quaternary tree of composition is together with this quaternary tree The subordinate quaternary tree of each leaf.Quaternary tree determine the segmentation of this given tree block for predicting, and subordinate quaternary tree Determine that one gives the segmentation of pre-assize block in order to residual coding.

The root node of quaternary tree is with completely to set block corresponding.For example, Fig. 3 A shows a tree block 150. Must remember, each image is divided into row and the regular grid of row, thus the seamlessly Covering samples number of this kind of tree block 150 Group.But notably for the whole blocks segmentation hereinafter shown, do not have critical importance without overlapping seamless segmentation.Adjacent on the contrary Near region block can overlap each other, as long as there is no the suitable subdivision that leaf block is neighbouring leaf block.

Together with the quad-tree structure of tree block 150, each node can be further divided into being four child nodes, at one time four In the case of fork is set, represent that tree block 150 can split into four sub-block, there is half-breadth and half height of tree block 150.At Fig. 3 A In, these sub-block are to indicate with reference number 152a to 152d.In the same manner, these sub-block are split the most again Become four less sub-block and there is half-breadth and half height of original sub-block.In Fig. 3 d, it is to illustrate for sub-block 152c Display, it is four small-sized sub-block 154a to 154d that sub-block 152c is subdivided into.To so far, Fig. 3 A to Fig. 3 C shows tree How block 150 is first separated into being four sub-block 152a to 152d, and then lower-left sub-block 152c is divided into again four Individual small-sized sub-block 154a to 154d；And the most as shown in Figure 3 C, the upper right block 154b of these small-sized sub-block is divided once again Being slit into is four blocks, each has 1/8th width and 1/8th height of raw element tree block 150, and these are the least Sub-block indicates with 156a to 156d.

Fig. 4 shows potential tree construction based on Quadtree Partition example as shown in Fig. 3 A to Fig. 3 d.The numeral that tree node is other For so-called subdivided mark value, will further describe when quad-tree structure signalisation discussed hereinafter.The root of quaternary tree Node is shown in this figure top (being denoted as level " 0 ").This root node in four branches of level 1 is and four sons shown in Fig. 3 A Block is corresponding.Because the third party in these sub-block is subdivided into its four sub-block the most in fig 3b, Fig. 4 is in level 1 3rd node also has four branches.Once again, the segmentation with the second of Fig. 3 C (upper right) child node is corresponding, has four sub-branches It is connected to the secondary nodal point of quaternary tree stratum level 2.Node in level 3 the most further segments.

Each leaf of quaternary tree be with item forecast parameter (that is internal or across, predictive mode, kinematic parameter etc.) can The variable-sized block being specified is corresponding.Hereinafter, these blocks are referred to as prediction block.Especially, these leaf blocks are Fig. 3 C Shown block.Briefly refer back to the explanation of Fig. 1 and Fig. 2, dispenser 20 or subdivider 28 and determine the quaternary tree as explained orally above Segmentation.In subdivider 152a-d execution tree block 150, sub-block 152a-d, small-sized sub-block 154a-d etc., which is by the thinnest The decision-making divided or further split, target is to obtain as indicated the most tiny prediction segmentation and the thickest prediction segmentation the most above Between optimal compromise.Predictor 12 is transferred the prediction segmentation specified by use and is predicted the granularity of segmentation or for such as with foundation Each sub-district of prediction that block shown in Fig. 3 C represents is to determine aforementioned Prediction Parameters.

Prediction block shown in Fig. 3 C can be further divided into smaller block in order to residual coding.For each prediction block, That is for each leaf node of a quaternary tree, determine corresponding by one or more subordinate quaternary trees for residual coding Segmentation.Such as, when allowing the maximum residul difference block size of 16 × 16, one gives 32 × 32 prediction block will be divided into four 16 × 16 blocks, are individually and are determined by the subordinate quaternary tree for residual coding.In this example each 16 × 16 block be with from The root node belonging to quaternary tree is corresponding.

Being subdivided into just like a given tree block described in the situation of prediction block, each prediction block can use subordinate quaternary tree to divide Solution is divided into multiple residual error block.Each leaf of one subordinate quaternary tree corresponds to a residual error block, can to this residual error block Showing indivedual residual coding parameter (that is pattern conversion, conversion coefficient etc.) by residual error precoder 14, such residual error is compiled Code parameter again then controls residual error reconstructor 16 and 106 respectively.

In other words, subdivider 28 can be configured to for each image or for each groups of pictures determine one prediction segmentation and one from Belong to prediction segmentation, can first divide the image into into the regularly arranged of tree block 150, segmented by quaternary tree and recursively subregion this One subset of a little tree blocks obtains prediction and is subdivided into prediction block, if not carrying out subregion, then this Target area at indivedual tree blocks Block can be tree block, and the probable rear subset segmenting these prediction block further, then for the leaf block of quaternary tree segmentation；With Reason, if a prediction block is greater than the full-size of subordinate residual error segmentation, via first item forecast block being divided into subtree Block regularly arranged, then segments program according to quaternary tree, and a subset segmentation of these subtree blocks is obtained residual error district Block, if not carrying out being divided into subtree block at item forecast block, then this residual error block can be prediction block, if or this small pin for the case Tree block does not carries out being divided into and smaller area, then this residual error block is subtree block, or the leaf block for the segmentation of residual error quaternary tree.

Such as outline above, subordinate array can be mapped to for the segmentation selected by an array.When considering and an array During the subordinate array of identical dimensional, this point is relatively easy.But must adopt when subordinate array dimension is different from an array dimension Use special measure.It sayed in outline, and in the case of different size, an array segmentation is mapped to subordinate array and can be mapped by space Carry out, that is map to subordinate array via by the block border space of an array segmentation.Especially, for each subordinate number Group, in the horizontal direction and vertical direction can have scaling factor, it determines array dimension ratio to subordinate array.Subordinate array It is divided into the sub-block for prediction and residual coding can pass through a quaternary tree, the common locating tree block of an array respectively Respective subordinate quaternary tree determines, subordinate array gained tree block is to be calibrated by the relative calibration factor.When horizontal direction and hang down Nogata to scaling factor different (such as in 4:2:2 colourity time sampling) time, the prediction block of gained subordinate array and residual error Block will be no longer square.In such cases, can predetermine or adaptability selects (for whole sequence, in this sequence Individual image or for each Individual forecast block or residual error block) whether non-square residual error block should split into square block.Example As, in the first case, encoder and decoder will agree to, when mapping block and being the most square every time, the side of being subdivided into when segmenting Shape block.In a second situation, subdivider 28 will lead to subdivider 104a signal via data stream inserter 18 and data stream 22 Know this selection.Such as, in the case of 4:2:2 colourity time sampling, subordinate array has the half-breadth of an array but contour, residual error The height of block is the twice of width.By by this block longitudinal splitting, two square blocks can be obtained once again.

It has been observed that subdivider 28 or dispenser 20 are pitched based on four to subdivider 104a signalisation via data stream 22 respectively The segmentation of tree.In order to reach this purpose, subdivider 28 notifies thin about for selected by image 24 of data stream inserter 18 Point.Data stream inserter transmission primaries quaternary tree again and the structure of secondary quaternary tree, therefore, transmission picture number group is partitioned into can Become size block at the prediction block within data stream or bit stream 22 or residual error block to decoding end.

Minimum and maximum admissible block size transmission as side information and can change according to different images.Or, Minimum and maximum allows that the big I of block is fixed at encoder and decoder.These minimum and maximum big I of block are for prediction Block and residual error block and have difference.For the signalisation of quad-tree structure, quaternary tree must be traversed, must for each node Must show that whether this specific node is a leaf node (that is corresponding block the most further segments) of quaternary tree, or this is specific Whether node is branched off into its four child nodes (that is corresponding block becomes four sub-block with double sized divisions).

Signalisation within one image is to carry out, the most from left to right and by upper by tree block with raster scan order Under to, if Fig. 5 A is in 140 displays.This kind of scanning sequency also can be different, such as, carry out from bottom right to upper left in chessboard mode.Relatively In good embodiment, respectively set block and thus each quaternary tree is to travel through for this subdivided information of signalisation with depth-first fashion.

In the preferred embodiment, not only subdivided information (that is tree construction), simultaneously prediction data etc. (that is with the leaf of this tree The payload that node is associated) it is with depth-first level transmission/process.The reason so carried out is that depth-first traversal has There is the advantage being better than breadth-first.In figure 5b, quad-tree structure be denoted as a with leaf node, b ..., j present.Fig. 5 A shows Show that gained block is split.If block/leaf node is with breadth-first order traversal, then obtain following order: abjchidefg.But According to depth-first order, this order is abc...ij.As knowable to Fig. 5 A, according to depth-first order, left adjacent block and Top adjacent block was always transmitted before current block/is processed.So, motion vector prediction and context modeling can always use Parameter specified by a left side and top adjacent block is reached improvement coding efficiency.For breadth-first order, not this kind of situation, Reason is that block j e.g. transmitted before block e, g and i.

As a result, the signalisation for each tree block is to carry out along the quad-tree structure recurrence of a quaternary tree so that For each node-node transmission one labelling, show whether corresponding block splits into four sub-block.If this labelling has value " 1 " (for "true"), then whole four sub-Node price are repeated by this signalisation program, that is sub-block is suitable with raster scanning Sequence (upper left, upper right, lower-left, bottom right) is until reaching the leaf node of a quaternary tree.Notice that leaf node is characterized by segmentation The value of labelling is " 0 ".It is to reside in the lowest-order laminar level of a quaternary tree and so allow pre-corresponding to minimum for node Survey the situation of block size, subdivided mark need not be transmitted.For the example of Fig. 3 A to Fig. 3 C, as shown in the 190 of Fig. 6 A, first will Transmission " 1 " shows that setting block 150 is split into as its four sub-block 152a-d.Then, with raster scan order 200 recursively Encode the subdivided information of whole four sub-block 152a-d.For first two sub-block 152a, b, " 0 " will be transmitted, show that it is not Through segmentation (in reference to Fig. 6 A 202).For the 3rd sub-block 152c (lower-left), " 1 " will be transmitted, show that this block is through segmentation (with reference in Fig. 6 A 204).Now according to recurrence way, four sub-block 154a-d of this block will be processed.Herein will be for First sub-block (206) is transmitted " 0 " and transmits " 1 " for second (upper right) sub-block (208).The minimum sub-block of Fig. 3 C now Four block 156a-d of size will be processed.If the minimum having reached this example allows block size, then need not transmit again Data, reason is impossible segment further.Otherwise, show that " 0000 " that these blocks the most further segment will transmit, If Fig. 6 A is in 210 instructions.Subsequently, by two, the lower section encrypted communication " 00 " (with reference in Fig. 6 A 212) to Fig. 3 b and last to figure The bottom right encrypted communication " 0 " (with reference to 214) of 3A.Therefore represent that the complete binary string of quad-tree structure will be for shown in Fig. 6 A.

This kind of binary string of Fig. 6 A represents that the different background shade of kenel corresponds to stratum based on quaternary tree segmentation Different levels in relation.Shade 216 represents level 0 (corresponding to block size equal to raw element tree block size), shade 218 table Showing level 1 (the most half as large equal to raw element tree block corresponding to block size), shade 220 represents that level 2 is (corresponding to block size Equal to raw element tree block size 1/4th), shade 222 represents that level 3 is (big equal to raw element tree block corresponding to block size Little 1/8th).Identical hierarchical level (corresponding to example binary string represent the same block size in kenel and Same hue) whole subdivided mark such as can use one by inserter 18 and same probability model does entropy code.

Noting the situation for breadth first traversal, subdivided information will transmit with different order, be shown in Fig. 6 B.

Being similar to each segmentation setting block for prediction, each gained prediction block is divided into the residual error block must be in position Streaming.Maximum and minimum block can be had big for the residual coding transmitted as side information and may change according to image again Little.Or the maximum and smallest region block size for residual coding can fix at encoder and decoder.Each a quaternary tree Individual leaf node, as shown in Figure 3 C, corresponding prediction block may be partitioned into the residual error block of maximum allowable size.These blocks be from Belong to the quad-tree structure composition root node for residual coding.For example, if the maximum residul difference block size of image be 64 × 64 and prediction block size be 32 × 32, the most whole prediction block would correspond to subordinate (residual error) four fork of size 32 × 32 Root vertex.On the other hand, if the maximum residul difference block for image is 16 × 16, then 32 × 32 prediction block will be residual by four Difference quaternary tree root node is formed, and each has the size of 16 × 16.Inside each prediction block, subordinate quad-tree structure Signalisation is to carry out by root node with raster scan order (left-to-right, up under).It is similar to once (prediction) quaternary tree knot The situation of structure, for each node, encodes a labelling, shows whether this specific node divides and becomes four child nodes.If then This labelling has value " 1 ", then (left with raster scan order for whole four corresponding child nodes and corresponding sub-block thereof Above, upper right, lower-left, bottom right) recursively repeat until reaching the leaf node of subordinate quaternary tree.Such as the situation of a quaternary tree, For the node in subordinate quaternary tree lowest-order laminar level without signalisation, reason is that these nodes correspond to Little may the block of residual error block size and cannot further split.

For entropy code, the residual error block subdivided mark of the residual error block belonging to same block size can use one and with One probability model coding.

So, according to the example presented with regard to Fig. 3 A to Fig. 6 A above, subdivider 28 defines the once segmentation for prediction And the sub-subordinate segmentation with different size block once segmented for residual coding purpose.Data stream inserter 18 is logical Crossing to encode with zigzag scan order signalisation for each tree block and once segment, bit sequence is based on Fig. 6 A and sets up, even The block size of maximum once segmented with coding and maximum hierarchical level.The prediction block so defined for each, The Prediction Parameters being associated has included at bit stream.Additionally, similar information (that is according to the full-size size of Fig. 6 A, maximum Hierarchical level and bit sequence) coding can carry out for each prediction block, the size of this prediction block is equal to or less than residual The full-size size of difference segmentation；And carry out for each residual error tree root block, wherein prediction block is split in advance Become to exceed the full-size size that residual error block is defined.The residual error block so defined for each, residual error data is slotting Enter this data stream.

Withdrawal device 102 extracts indivedual bit sequences and notice dispenser 104 about such institute at input 116 from this data stream The subdivided information obtained.Additionally, data stream inserter 18 and withdrawal device 102 can use aforementioned sequence to be used in prediction block and residual error district Extra syntactic element is transmitted, the residual error data that such as exported and by predictor 12 institute by residual error precoder 14 between block The Prediction Parameters of output.The advantage using this kind of order is the syntactic element by utilizing adjacent block coding/decoding, The optional suitable context for certain block coding individual grammar element.Additionally, in like manner, residual error precoder 14 and prediction Device 12 and residual error reconstructor 106 and precoder 110 can the sequential processing item forecast block of outline and residual error districts above Block.

The flow chart of Fig. 7 step display, this step can perform, by withdrawal device 102, the side that digest is stated in the past when encoding Formula extracts subdivided information from data stream 22.In the first step, image 24 is divided into tree root block 150 by withdrawal device 102.This Step is to indicate with step 300 at Fig. 7.Step 300 relates to withdrawal device 102 and extracts maximum predicted block size from data stream 22. Further additionally or alternatively, step 300 can relate to withdrawal device 102 and extracts maximum hierarchical level from data stream 22.

It follows that in step 302, withdrawal device 102 is from this data stream one labelling or one.Carry out very first time step Rapid 302, it is suitable that the position being labeled as individually belonging to the first tree root block 150 according to tree root Reginal-block scanning order 140 known by withdrawal device 102 First labelling of sequence.Therefore this labelling being labeled as there is hierarchical level 0, in step 302, withdrawal device 102 can use with The context modeling that this hierarchical level 0 is associated determines a context.Each context have indivedual probability estimation for The entropy code of its labelling being associated.The probability estimation of context can individually adapt to context and add up in individual contexts symbol Numeral.Such as, for the suitable context determining to be used for decoding the labelling of hierarchical level 0 in step 302, withdrawal device 102 can Selecting a context in a set of context, it is to be associated with hierarchical level 0, depends on the stratum of neighbouring tree block Formula level 0 labelling, is more dependent upon again defining the current neighbouring tree block (such as pushing up and left neighbouring tree block) processing tree block Information contained in the bit string of quaternary tree segmentation.

In next step, that is step 304, withdrawal device 102 checks that current decoder marks whether to point out subregion.If belonging to this kind Situation, then withdrawal device 102 is by current block (at present for tree block) subregion, or indicates this kind of subregion to segmentation in step 306 Device 104a, in step 308, it checks whether current hierarchical level subtracts 1 equal to maximum hierarchical level.For example, withdrawal device 102 the most also have the maximum hierarchical level extracted in step 300 from data stream.If hierarchical level is not equal to the most at present Big hierarchical level subtracts 1, then in step 310 withdrawal device 102, current hierarchical level be incremented by 1 and return step 302 from this Data stream next one labelling.Now, the labelling to be decoded in step 302 belongs to another hierarchical level, therefore foundation One embodiment, the one in the optional different set of context of withdrawal device 102, this set is belonging to current hierarchical level.Should Select the segmentation bit sequence that may be based on the most decoded neighbouring tree block according to Fig. 6 A.

If decoding a labelling, and the inspection of step 304 discloses this labelling and does not point out the subregion of current block, then withdrawal device 102 advance steps 312 check whether current hierarchical level is 0.If belonging to this kind of situation, withdrawal device 102 with regard to step 314 according to The next tree root block of scanning sequency 140 processes, if or not leaving any tree root block to be processed, then stopping process Extraction subdivided information.

It should be noted that the description focus of Fig. 7 is the decoding of the segmentation cue mark focusing on only prediction segmentation, historical facts or anecdotes On border, step 314 relates to other storehouse (bin) or the decoding of syntactic element that the relevant block of tree the most at present is associated.This kind of feelings Under condition, if there are another or next tree root block, then withdrawal device 102 is advanced to step 302 by step 314, from segmentation The next labelling of information decoding, that is the first labelling of the labelled sequence about new tree block.

In step 312, if hierarchical level is not equal to 0, then operation advances to step 316, has checked for Close other child node of current node.In other words, when withdrawal device 102 checks in step 316, examined in step 312 Looking into current hierarchical level is the hierarchical level beyond 0 hierarchical level.Transfer the most again expression and there are parent node, its It is belonging to tree root block 150 or small-sized block 152a-d or the one in smaller area block 152a-d etc. again.Current decoder labelling institute The tree construction node belonged to has a parent node, and the other three node that this parent node is this current tree construction is shared.Tool There is the scanning sequency between these child nodes sharing parent node to be illustrated in Fig. 3 A, for hierarchical level 0, there is ginseng Examine label 200.So, in step 316, withdrawal device 102 checks whether that whole four child nodes are the most in the process journey of Fig. 7 Sequence is accessed.If not belonging to this kind of situation, that is parent node has extra child node at present, then the processing routine of Fig. 7 advances to Step 318, it is accessed that this is in the internal next child node according to zigzag scan order 200 of current hierarchical level, because of This its corresponding sub-block represents the current block of Fig. 7 now, and subsequently, or saves at present from relevant current block in step 302 Data stream one labelling of point.But, in step 316, if current parent node be there is no extra child node, the then side of Fig. 7 Method advances to step 320, and at present hierarchical level successively decreases 1 herein, and the method is to carry out with step 312 the most subsequently.

By performing step shown in Fig. 7, withdrawal device 102 and subdivider 104a pull together to cooperate to come in encoder-side from data stream Fetch selected segmentation.The method focus of Fig. 7 concentrates the situation in aforementioned prediction segmentation.The flow chart of constitutional diagram 7, figure How 8 display withdrawal devices 102 and subdivider 104a pull together to cooperate to fetch residual error segmentation from data stream.

In specific words, Fig. 8 shows for from prediction segmentation each prediction block of gained, by withdrawal device 102 and subdivider The step that 104a is carried out respectively.It has been observed that these prediction block are according to zigzag scanning between the tree block 150 of prediction segmentation Sequentially 140 traversals, and use shown in such as Fig. 3 C to come by tree in the internal depth-first traversal accessed at present of each tree block 150 Block.According to depth-first traversal order, the leaf block once setting block through subregion is to visit with depth-first traversal order Ask, access the sub-block of certain hierarchical level with shared current node with zigzag scan order 200, and advancing So far the respective segmentation of these sub-block is mainly scanned before planting the next sub-block of zigzag scan order 200.

It is to show with reference number 350 for gained scanning sequency between the example of Fig. 3 C, the leaf node of tree block 150.

For the prediction block accessed at present, the processing routine of Fig. 8 starts from step 400.In step 400, indicate current district The inner parameter of the current size of block is set equal to the size of the hierarchical level 0 of residual error segmentation, that is residual error is segmented Big block size.Must remember that maximum residul difference block size is smaller than the smallest region block size of prediction segmentation, maybe can equal to or more than The latter.In other words, according to an embodiment, encoder can unrestricted choice any one possibility aforementioned.

At next step, that is step 402, perform to check whether the prediction block size about accessing block at present is more than It is denoted as the inner parameter of current size.If belonging to this kind of situation, then it is probably a leaf block of prediction segmentation or predicts segmentation One tree block and be greater than maximum residul difference block size without the prediction block that accesses at present of any further subregion, this kind of situation Under, the processing routine of Fig. 8 advances to the step 300 of Fig. 7.In other words, the prediction block accessed at present is divided into residual error tree root Block, the first labelling of the labelled sequence of the first residual tree block within this kind of current access prediction block is in step 302 Decoding etc..

If but access at present prediction block and there is size equal to or less than the inner parameter indicating current size, then Fig. 8 Processing routine advances to step 404, checks that prediction block size determines that whether it is equal to the inside indicating current size herein Parameter.If it has, then segmentation step 300 can skip, processing routine directly continues to the step 302 of Fig. 7.

If but the prediction block size accessing prediction block at present is less than indicating the inner parameter of current size, then Fig. 8 Processing routine advance to step 406, hierarchical level is incremented by 1 herein, and current size is set as the size of new hierarchical level, Such as with 2 segmentations (downloading two direction of principal axis in quaternary tree segmentation situation).Subsequently, carry out the inspection of step 404 once again, pass through step The loop effect that 404 and 406 are formed is that hierarchical level is regularly corresponding with the corresponding block size being intended to subregion, and And have less than or equal to/independent more than the item forecast block of maximum residul difference block size the most unrelated.So, when in step 302 During coding symbols, the context modeling carried out is hierarchical level and the block size simultaneously depending on this labelling indication.Pin The advantage of different context is used to be that probability is estimated the most applicable respectively the labelling of different estate formula level or block size The actual probability distribution that mark value occurs, on the other hand has the moderate context number to be managed, thus reduces context pipe Manage expense, and increase context adjusts and is adapted to actual symbol statistics.

As the most already described, having more than one sample array, these sample arrays can be grouped into one or more plane group. Such as enter input 32 is intended to the image that coded input signal can be video sequence or static images.So this image is In one or more sample array form.In a picture coding context of video sequence or static images, sample array is Refer to three color planes, the reddest, green and blue, or refer to that luminance plane and colorimetric plane are such as in the colored expression of YUV or YCbCr Kenel.In addition, it is possible to present the sample array of the depth information representing α (that is transparency) and/or 3-D video data.Multiple These sample arrays can be grouped into so-called plane group together.Such as, brightness (Y) can be the one of only one of which sample array Individual plane group, and colourity (such as YCbCr) can be to have another plane group of two sample arrays；Or at another example In, UV can be that to have a plane group of three matrixes and the depth information of 3-D video data can be only one of which sample number The Different Plane group of group.For each plane group, a quad-tree structure can be used in data stream 22 in-line coding Represent and be divided into prediction block；And for each prediction block, secondary quad-tree structure represents and is divided into residual error block.So, According to aforementioned first example, luminance component is a plane group, and chromatic component forms another plane group herein, one four Fork tree construction is the prediction block for luminance plane, and a quad-tree structure is the residual error block for luminance plane, one Quad-tree structure is the prediction block for colorimetric plane, and a quad-tree structure is the residual error block for colorimetric plane. But in aforementioned second example, may there is a quad-tree structure for brightness and colourity prediction block (YUV) together, one Quad-tree structure for brightness and colourity residual error block (YUV) together, deep for 3-D video data of quad-tree structure The prediction block of degree information, and quad-tree structure is for the residual error block of the depth information of 3-D video data.

Additionally, in described above, input signal is to use a quad-tree structure to be divided into multiple prediction block, now Describe how these prediction block use subordinate quad-tree structure to be further subdivided into residual error block.According to another embodiment, Segmentation not terminates in subordinate quaternary tree level.In other words, subordinate quad-tree structure is used may to use from the block of segmentation gained Ternary quad-tree structure is segmented further.This kind of segmentation again then is used in the purpose using extra coding tools, and it may be assisted The coding of residual signals.

Focus described above is concentrated and is being finely divided by subdivider 28 and subdivider 104a respectively.It has been observed that point The process granularity of the module that can control afore-mentioned code device 10 and decoder 100 it is not finely divided by subdivider 28 and 104a.But According to embodiment described later, subdivider 228 and 104a is then combiner 30 and combiner 104b respectively.But notably close And device 30 and 104b is selectivity and can exempt.

But actually and as describing in detail after a while, if encoder is provided with by prediction block or residual error block by combiner Dry person is combined into the chance of group or group variety so that in other module or other module at least partially can be by these blocks group Group processes together.For example, it was predicted that device 12 can sacrifice measured part by the segmentation optimization using subdivider 28 Deviation between the Prediction Parameters of prediction block, and use the Prediction Parameters to all these prediction block are shared to replace, as long as Prediction block packet together with the signalisation of the shared parameter transmission of the whole blocks belonging to this group with regard to rate/distortion ratio in meaning Speech, has more prospect property than the Prediction Parameters individually signalisation of all these prediction block.Share pre-based on these Survey parameter, fetch the processing routine of prediction itself at predictor 12 and 110 and remain and carry out one by one with prediction block.It is also possible to Predictor 12 and 110 is even once predicted program to whole prediction block group.

As describing in detail after a while, it is also possible to prediction block group not only uses for the identical of one group of prediction block or shares Prediction Parameters, the most additionally or alternatively, it is allowed to encoder 10 sends a Prediction Parameters for this group together with right Belong to the prediction residual of the prediction block of this group, thus the letter of the Prediction Parameters for this group of signalisation can be reduced Number notification overhead.In the case of aftermentioned, consolidation procedure only affect data stream inserter 18 rather than impact by residual error precoder 14 and The decision-making that predictor 12 is done.But further detail below is as describing in detail after a while.But, for completeness, the most aforementioned aspect is also fitted Segment for other, the segmentation of the most aforementioned residual error or filtering segmentation.

First, the merging of sample set (the most aforementioned prediction block and residual error block) is to encourage with more typically property meaning, That is it is not limited to the segmentation of aforementioned multiway tree.But explanation focus subsequently will focus on previous embodiment and segmented gained by multiway tree The merging of block.

It sayed in outline, merges, for transmitting the purpose of the coding parameter being associated, the grammer being associated with specific sample set Element, it is allowed to apply upper minimizing side information rate in image and Video coding.For example, the sample array of the signal to be encoded Being typically to divide specific sample set or sample set into, it can represent rectangle block or square block, or sample any its Its set, including arbitrary shape district, triangle or other shape.In the aforementioned embodiment, simple bonding pad is thin from multiway tree Divide prediction block and the residual error block of gained.The segmentation of sample array can be fixed by grammer；Or it has been observed that segmentation also can be at least Part notifies at bit stream internal signal.In order to the side information rate being used for signalisation subdivided information is maintained little, grammer leads to Often only allow a limited number of selection to cause simple subregion, such as block segmentation is become smaller block.Sample set Being to be associated with specific coding parameter, it can be shown that information of forecasting or residual coding pattern etc..About the details of this subject under discussion be as Described above.For each sample set, can transmit and such as encode individually ginseng for show predictive coding and/or residual coding Number.In order to reach improvement code efficiency, the merging aspect being hereinafter described, also will be merged into what is called by two or more sample sets Sample set group, it is allowed to reach some advantages, as describing in detail after a while.For example, sample set can be through merging so that this Whole sample sets of one group share identical coding parameter, and it can transmit together with the one in the sample set in group.Logical Crossing this mode, coding parameter individually need not transmit for each sample set in sample set group, replaces on the contrary, Coding parameter is to whole sample set group transmission the most once.As a result, the side information being used for transmitting coding parameter reduces, and always Code efficiency can improve.As for instead road, the additional purification of one or more coding parameters can be for a sample set group In one or more sample sets transmission.The refined whole sample sets that can apply to a group, or only apply to for This sample set of its transmission.

Encoder also is provided to form relatively high-freedom degree during bit stream 22, reason by the merging aspect hereinafter further described It is that merging way dramatically increases the probability number for selecting subregion one image pattern array.Because encoder can be at relatively multiselect Select between Xiang, be such as used for reducing particular rate/distortion measurement, therefore code efficiency can be improved.Operation encoder has several possible. In simple approach, first encoder can determine the optimal segmentation of sample array.Brief with reference to Fig. 1, subdivider 28 will be in the first order Determine optimal segmentation.Subsequently, for each sample set, check whether and merge with another sample set or another sample set group Reduce particular rate/distortion cost measure.In this connection, merge, with one, the Prediction Parameters that sample set group is associated Can re-evaluate, such as by performing new motion search and estimation；Or have been directed towards sharing sample set and candidate samples collection Close or sample set group Prediction Parameters the most after measured for merging can be assessed for the sample set group considered. In more comprehensive way, particular rate/distortion cost measure can be to the assessment of additional candidate sample set group.

The merging way being notably hereinafter described does not changes the processing sequence of sample set.In other words, merging conception can It is embodied as in one way so that postpone not to be further added by, that is each sample set maintains and can decode in moment at the same time And do not use merging way.

For example, if the bit rate saved by reducing coding Prediction Parameters number is greater than additionally consuming closing at coding And information is used to refer to merge to the bit rate of decoding end, merges way (as describing in detail after a while) and cause code efficiency to increase.Enter one Step must mention that the described grammer for merging extends provides extra discretion to select image or plane group subregion encoder Become multiple block.In other words, encoder is not limited to first be finely divided and whether then check the some persons in gained block There is Prediction Parameters identity set or similar set.As for a simple instead road, according to rate-distortion cost measure, First encoder determines segmentation, and then can check adjacent block for each block coder or be associated the most after measured Block group in one merge whether attenuating rate-distortion cost measure.So, can re-evaluate and this new block group The Prediction Parameters being associated, such as by by search of newly moving；Or to current block and adjacent block or block group This Prediction Parameters that group determines can be assessed for new block group.Pooling information carries out signalisation in units of block.Effectively Ground, merges the result that also can be interpreted as the Prediction Parameters inference for current block, and wherein the Prediction Parameters of inference is set to Prediction Parameters equal to the one in adjacent block.It addition, residual error can be transmitted for the block in a block group.

So, the potential basic conception below conception that merges being described later on is by adjacent block merging is become a district The bit rate needed for communicating predicted parameter or other coding parameter is lowered in block group, and the most each block group is all with coding parameter A unique set such as Prediction Parameters or residual coding parameter is associated.In addition to subdivided information (if present), pooling information Also notify at bit stream internal signal.The advantage merging conception is to reduce from the side information of coding parameter to cause code efficiency to increase High.Merging method the most described herein also may extend to other dimension beyond Spatial Dimension.For example, in several differences One sample of the inside of video image or block sets group can be merged into a block group.Merge and be equally applicable to 4-compression And light code field.

So, briefly refer back to the explanation of Fig. 1 to Fig. 8 above, notice that the consolidation procedure after segmentation has superiority, And it is independent with the ad hoc fashion of subdivider 28 and 104a subdivision graph picture unrelated.More clearly saying it, H.264 the latter can also be similar to that Mode subdivision graph picture, in other words, each image is subdivided into and there are preliminary dimension such as 16 × 16 luma samples or at data stream Internal signal notifies the rectangle of size or the regularly arranged of square gathering block, and each macro zone block has some volumes associated there Code parameter, is used as including such as defining the regular sublattice dividing 1,2,4 or some other number of partitions into for each macro zone block Corresponding Prediction Parameters in prediction granularity and bit stream and being used for define residual error and corresponding real transform granularity point The partitioned parameters in district.

Sum it up, merge the advantage providing short discussion above, such as reduce in image and Video coding are applied Side information rate position.Represent rectangle or square block or arbitrary shape district or the most any simple connection of other sample set any The specific sample set of district or sample is commonly connected specific coding parameter sets；For each sample set, coding parameter is to contain Including at bit stream, coding parameter such as represents Prediction Parameters, and it specifies corresponding sample set is how to use encoded sample in addition Prediction.One image pattern array is divided into sample set and can be fixed by grammer, maybe can corresponding by within this bit stream Subdivided information signalisation.Coding parameter for this sample set can pass with predefined procedure (that is the given order of grammer) Defeated.According to pooling function, combiner 30 can share sample set or a current block (such as with one or more other sample for one The prediction block of this set merging or residual error block), combiner 30 signal notifies into a sample set group.One group's sample The coding parameter of set therefore need only transmission primaries.In a particular embodiment, if sample set is and has transmitted coding at present One sample set of parameter or existing sample set group merge, then the coding parameter of sample set does not transmits at present.On the contrary, mesh The coding parameter of front sample set is set to this sample set or this sample set merged equal to current sample set with it The coding parameter of group.As for instead road, one or more the additional purification in coding parameter can be to current sample set Transmission.The refined whole sample sets that can be applicable to a group or only applying are to this sample set transmitted for it.

According to an embodiment, for each sample set (the most aforementioned prediction block, aforementioned residual error block or aforementioned multiway tree The leaf block of segmentation), the set of whole previous coding/decoding sample set is referred to as " set of cause and effect sample set ".Example As with reference to Fig. 3 C.Whole blocks that this figure shows are all certain segmentation result, such as prediction segmentation or residual error segmentation or any many Unit's tree segmentations etc., the coding/decoding order defined between these blocks is to define with arrow 350.Consider certain between these blocks Block is current sample set or current simple bonding pad, the set of its cause and effect sample set be by along order 350 at present Whole blocks in block front are formed.As long as but must remember to consider hereinafter about the discussion of Unite principle, then do not use polynary Other segmentation of tree segmentation also falls within possibility.

This sample set that can be used to merge with current sample set is in hereinafter referred to as the " collection of candidate samples set Close ", regular is the subset of " set of cause and effect sample set ".The mode how forming this subset is that decoder is known, Or can show inside data stream or bit stream to decoder from encoder.If specific current sample set is encoded/decoding, The then set non-NULL of this candidate samples set, it is to notify at data stream internal signal at encoder, or at decoder from this number Calculate whether this shared sample set merges with a sample set in the set of this candidate samples set according to conductance, and if so, It is with any one in this sample set to merge.Otherwise, merging and be not used to this block, reason is the collection of candidate samples set It is empty for closing regular.

There is different modes so that the set measure cause and effect sample set would indicate that this subset of the set of candidate samples set. For example, the mensuration of candidate samples set can be based on the sample within current sample set, and it has the geometry of uniqueness Definition, the upper left image sample of such as rectangle block or square block.Start from this kind of unique geometry definition sample, determine spy Determine non-zero number sample, represent that the straight space of this kind of unique geometry definition sample is adjacent to sample.For example, this kind of spy Determine the neighbouring sample of upper neighbouring sample and a left side that non-zero number sample includes unique geometric definition sample of current sample set, thus adjacent The non-zero number of nearly sample is at most 2, if the one on or or in left neighbouring sample cannot obtain or be positioned at outside this image Side, then non-zero number is 1；Or, if disappearance two is in the case of sample, then non-zero number is 0.

The set of candidate samples set can be through at least in the non-zero number that decision is contained containing aforementioned neighbouring sample Those sample sets of person.Such as with reference to Fig. 9 A.The sample set being presently considered is combining objects, must be block X, and its geometry Shape distinct definition sample must illustrate as upper left sample, with 400 instructions.Top and the left neighbouring sample of sample 400 refer to respectively It is shown as 402 and 404.The set of cause and effect sample geometry or the set of cause and effect block are to add shade mode to emphasize.Therefore these districts In block, block A and B includes the one in neighbouring sample 402 and 404, and these blocks form candidate block set or candidate samples The set of set.

According to another embodiment, can extraly or exclusively for merging the set of the candidate samples set that purpose is determined Including the sample set containing a specific non-zero number sample, this number can be 1 or 2, and the two has same spatial location, but It is contained in different images, that is previous coding/decoding image.For example, in addition to block A and B of Fig. 9 A, can use previously The block of coded image, it is included in the sample of same position of sample 400.By this mode, note upper neighbouring sample 404 or the most left neighbouring samples 402 can be used to define the non-zero number of aforementioned neighbouring sample.Generally, candidate samples set Set can be led calculate from the internal previous treated data of current image or other image.Lead calculation and can include direction in space Information, the conversion coefficient being such as associated with specific direction and the image gradient of current image；Maybe can include time orientation information, Such as adjacent to movement representation kenel.By these in the available data of receiver/decoder and at other number within data stream According to and side information (if present), the set calculating candidate samples set can be led.

Notably leading of candidate samples set passes through the combiner 30 in encoder-side and the merging in decoder end at last Device 104b performs side by side.Just it has been observed that the two can determine the most unrelated based on the mode defined in advance known to the two The set of candidate samples set；Or encoder can imply clue in bit stream internal signal notice, it is to be carried by combiner 104b , with the same way of the combiner 30 in the set determining candidate samples set in encoder-side, to perform these to a position Candidate samples set lead calculation.

As describing in detail after a while, combiner 30 and the cooperation of data stream inserter 18 transmit one or more for each sample set Syntactic element, it shows whether this sample set merges with another sample set, and this another sample set can be again to have merged The part of sample set group, and which one in this set of candidate samples set is for merging.Withdrawal device 102 Then extract these syntactic elements and notice combiner 104b accordingly.Especially, according to the specific embodiment being hereinafter described, for one Specific sample set one or two syntactic element of transmission shows pooling information.First syntactic element shows that current sample set is No merge with another sample set.If the first syntactic element shows that this current sample set is to merge, only with another sample set The second syntactic element having this kind of situation just to transmit shows which one in the set of candidate samples set is for merging.If leading calculation The collection going out candidate samples set is combined into sky, then can suppress the transmission of the first syntactic element.In other words, if leading the candidate samples calculated The set of set non-NULL, then have this kind of situation only and just transmit the first syntactic element.Have only and lead the candidate samples set that calculates Set is containing just transmitting the second syntactic element during more than one sample set, if reason is the set in this candidate samples set In comprise only a sample set, then must not do further selection.Furthermore, if the set of candidate samples set includes being more than One sample set, then the transmission of the second syntactic element can be suppressed；If but the whole samples in the set of candidate samples set Set is the most no when being to be associated with one and same coding parameter.In other words, the second syntactic element has only and leads, one, the candidate's sample calculated Just transmission when at least two sample set in the set of this set is to be associated with different coding parameter.

Inside this bit stream, the pooling information of a sample set can the Prediction Parameters being associated with this sample set or its Encode before its specific coding parameter.Prediction Parameters or coding parameter have only and merge at the current sample set of pooling information signalisation Just transmission when not merging with other sample set any.

Such as, the pooling information of certain sample set (that is, one block) can encode after suitable Prediction Parameters subset；Or More typically property definition, has transmitted the coding parameter being associated with these indivedual sample sets.Prediction/coding parameter subset can be by one Individual or multiple reference picture index or one or more components of kinematic parameter vector or benchmark index and kinematic parameter vector One or more components etc. formed.The Prediction Parameters transmitted or coding parameter subset can be used to from just like described above The set calculating a candidate samples set is led in the interim set of the bigger candidate samples set having led calculation.Lift an example, Encoded Prediction Parameters and coding parameter prediction corresponding with the previous candidate sample set ginseng of current sample set can be calculated Difference measurement between number or coding parameter or the distance of foundation preset distance measured value.Then, the difference only calculated is surveyed Value or distance are less than or equal to predetermined critical or lead those sample sets of the marginal value calculated and included at final collection Close (that is the set of the candidate samples set reduced).Such as with reference to Fig. 9 A.Sample set must be block X at present.Relevant local area One subset of the coding parameter of block must be already inserted into bit stream 22.For example, it is assumed that block X is prediction block, in the case of this kind, coding The suitable subset of parameter can be the Prediction Parameters subset of this block X, such as includes that image reference index and motion map information are (all Such as motion vector) one set in a subset.If block X is residual error block, then the subset of coding parameter is residual information Collection, such as conversion coefficient or the instruction mapping table in the notable conversion coefficient position within block X.Based on this information, data Both stream inserter 18 and withdrawal device 102 can use this information to the subset determining in block A and B, and this subset is at Ben Te Determine embodiment is constituted the preliminary set of aforementioned candidates sample set.In specific words, cause and effect sample set is belonged to because of block A and B Set, its coding parameter when the coding parameter of block X is current encoder/decoding by both encoder and decoder can profit With.Therefore, the aforementioned any number compared in the preliminary set that can be used to get rid of candidate samples set A and B of different way is used Mesh block.Then, the set of the candidate samples set that gained reduces can use as before, in other words, is used for determining to merge Whether indicator indicates transmission from this data stream to merge or extraction merges from this data stream, depends on reducing candidate's sample at this Depending on whether the sample set number within set of this set and the second syntactic element must transmit wherein；Or from This data stream extracts, has the second syntactic element instruction reduces candidate samples set which sample set palpus internal at this For merging companion's block.

Afore-mentioned distance relative to its aforementioned marginal value compared can for fixing and for both encoder and decoder Knowing, or can lead calculate based on the distance calculated, middle number or some other of such as different value take middle tendency etc..This kind of situation Under, inevitably, the set reducing candidate samples set must be for the suitable subset of the preliminary set of candidate samples set.Separately Outward, have those sample sets according to distance measure is minimum range only just to select from the preliminary set of this candidate samples set Go out.It addition, use afore-mentioned distance measured value, from the preliminary set of this candidate samples set, only select just what a sample set Close.In the case of aftermentioned, which current sample set pooling information need only show be intended to single candidate samples set to merge ?.

So, candidate block set can be as hereinafter calculated with regard to being formed or lead shown in Fig. 9 A.Start from the current block X of Fig. 9 A Upper left sample position 400, lead in encoder-side and decoder end and calculate the neighbouring sample in its left neighbouring sample 402 position and top thereof 404 positions.So, candidate block set at most only two elements, that is the cause and effect set adding picture shade of Fig. 9 A contain The block in one (belonging to the situation of Fig. 9 A) in two sample position is block B and A.So, candidate block set only has There are two direct neighbor blocks of upper left sample position of current block as its element.According to another embodiment, candidate block Set can be given by whole blocks the most encoded before current block, and containing representing the direct of any sample of current block One or more samples of spatial neighbor sample.The most left neighbouring sample of the neighbouring any sample being limited to current block of straight space This and/or directly the neighbouring sample in top and/or the rightest neighbouring sample and/or the direct end adjacent to sample.Such as show with reference to Fig. 9 B Another block segments.In such cases, candidate block includes four blocks, that is block A, B, C and D.

It addition, candidate block set can include that (it is that position is at present containing one or more samples extraly or exclusively Any sample same position of block, but be contained in different images, that is encoded/decoding image) block.

Again it addition, a subset of candidate block set expression aforementioned zones set of blocks, it is by direction in space or time side To proximity relations determine.Candidate block subset can be through fixing, signalisation or lead calculation.Candidate block subset lead calculation it is contemplated that The decision-making in this image or other image, other block done.Lift example, or extremely phase identical with other candidate block As the block that is associated of coding parameter can not include in candidate block set.

Hereinafter the description to embodiment is applicable to the neighbouring sample in a left side and top of the upper left sample containing current block and only has two Individual block is considered as the situation of the most possible candidate.

If candidate block set non-NULL, a labelling of the most referred to as merge_flag, by signalisation, shows current district Whether block merges with any candidate block.If merge_flag is equal to 0 (for "false"), then this block will not be with its candidate regions One in block merges, and generally transmits whole coding parameters.If merge_flag is equal to 1 (for "true"), then it is suitable for aftermentioned person. If candidate block set contains one and only one of which block, then this candidate block is used for merging.Otherwise, candidate block set is proper Containing two blocks.If the Prediction Parameters of this two block is identical, then these Prediction Parameters are used for current block.Otherwise (this two block There is different Prediction Parameters), signalisation is referred to as the labelling of merge_left_flag.If merge_left_flag is equal to 1 (for "true"), then a left side for the upper left sample position containing current block is from this candidate regions adjacent to this block of sample position Set of blocks is selected.If merge_left_flag is equal to 0 (for "false" "), then select another from this candidate block set One (that is pushing up neighbouring) block.The Prediction Parameters of selected block be for current block.

Just merge the several persons in outline previous embodiment, show with reference to Figure 10 and perform from entering defeated by withdrawal device 102 Enter and the data stream 22 of end 116 extracts the step that pooling information is carried out.

Process starts from 450, identifies for current sample set or the candidate block of block or sample set.Must remember, district The coding parameter of block is in data stream 22 internal transmission with certain one-dimensional order, and accordingly, Figure 10 refers to for accessing at present Sample set or the block method of fetching pooling information.

It has been observed that identify and step 450 includes based on neighbouring aspect previously decoded blocks (that is cause and effect block collection Close) in identification.Such as, those adjacent block may point to candidate, candidate contain in space or on the time at current block X Neighbouring certain of one or more geometry predetermined sample adjacent to sample.Additionally, identification step can include two levels, that is The first order relates to causing a preliminary candidate block sets based on neighbouring just like aforementioned identification；And the most simple such block in the second level For before step 450 from the sensing block of data stream, the coding parameter that this block has transmitted meets current district Certain relation of the suitable subset of the coding parameter of block X.

Secondly, method advances to step 452, determines that whether candidate block number is more than zero at this.If belonging to this kind of situation, Then from data stream, extract merge_flag in step 454.Extraction step 454 can relate to entropy decoding.In step 454 for entropy solution The context of code merge_flag can be based on belonging to such as candidate block set or the syntactic element of preliminary candidate block sets, its In the dependence of syntactic element be can be limited to following information: belong to and pay close attention to the block of set and whether experience merging.Selected context Probability estimation can adjusted adapt to.

But, if candidate block number is determined as 0 452, Figure 10 method advances to step 456, herein the volume of current block Code parameter is to extract from bit stream, or in the case of said second identification instead road, wherein sweeps with block at withdrawal device 102 Retouch order (all as shown in Figure 3 C order 350) and process after next block carries out, remaining coding parameter.

With reference to step 454, the method advance step 458 after the extraction of step 454, check the merge_flag extracted Whether point out appearance that current block merges or do not exist.If not merging, then method advances to abovementioned steps 456.Otherwise, Method is advanced with step 460, including checking that whether candidate block number is equal to 1.If belonging to this kind of situation, in candidate block, certain is waited Not necessarily, therefore Figure 10 method advances to step 462 in the transmission of constituency block instruction, and the merging companion of block sets at present accordingly For unique candidate block, the coding parameter merging companion's block the most after step 464 is used to adjust coding parameter or mesh The adjusting or predicting of remaining coding parameter of front block.As a example by adjusting, the coding parameter that current block is omitted is merely to replicate From merging companion's block.In another case, that is prediction in the case of, step 464 can relate to take out further from data stream Take residual error data, about the residual error data of prediction residual omitting coding parameter of current block and derive from and merge companion's block The combination of the prediction of these residual error data and these omission coding parameters.

But, if candidate block number is determined as more than 1 in step 460, Figure 10 method advances to step 466, herein Carry out checking coding parameter or the concern part of coding parameter, that is the part not yet transferred inside the data stream of current block The subdivision being associated is consistent with each other.If belonging to this kind of situation, these shared coding parameters are set to merge reference, or candidate Block is to be set as merging companion in step 468, or indivedual coding parameter of paying close attention to is used in adjusting or predicting of step 464.

It should be noted that merging companion itself can be the block that signalisation merges.In this example, the warp of companion is merged Adjust or predicted gained coding parameter is for step 464.

But otherwise, in the case of coding parameter difference, Figure 10 method advances to step 470, extra syntactic element herein It is to be drawn from data stream, that is this merge_left_flag.The separately set of context can be used for entropy and decodes this labelling.For The set of context of entropy decoding merge_left_flag may also comprise a simple context.After step 470, merge_ The candidate block of left_flag instruction is set as merging companion in step 472, and is used for adjusting or predicting in step 464.In step After rapid 464, withdrawal device 102 is with block sequential processing next one block.

It is of course possible to have other instead road.Such as, combination syntactic element can be in data stream internal transmission, rather than as the most front State separately syntactic element merge_flag and merge_left_flag, combine syntactic element signalisation merging treatment program.This Outward, whether aforementioned merge_left_flag in data stream internal transmission, and can have identical Prediction Parameters with two candidate block Unrelated, the computing overhead performing Figure 10 processing routine is lowered by this.

As already described in the most such as Fig. 9 B, can include in candidate block set more than two blocks.Additionally, pooling information, That is the information whether signalisation one block merges；If so, the candidate block to be merged can pass through one or more grammers Elemental signals notifies.One syntactic element can be shown that this block is (the most aforementioned with any one in aforementioned candidates block Merge_flag) merge.When being all in the set non-NULL of candidate block, just transmission labelling.Second syntactic element signal notice where One candidate block is used in merging, the most aforementioned merge_left_flag, but is indicated generally at two or more than two candidate regions Selection between block.Just can be transmitted when having the one in the first syntactic element signalisation current block candidate block to be merged only Two syntactic elements.Second syntactic element the most more has only when candidate block set contains more than one candidate block, and/or candidate Just transmission when any one in block has the Prediction Parameters different from other person any of candidate block.Grammer can be depending on to Give how many candidate block and/or how different Prediction Parameters is associated with candidate block.

The grammer of which block in signalisation candidate block to be used can encoder-side and decoder end simultaneously and/ Or set side by side.For example, if identifying that three candidate block select in step 450, grammer is to select as only three selections It is available, such as, considers for entropy code in step 470.In other words, syntactic element is to be selected so that its symbols alphabet Only there are multiple elements as the selection of existing candidate block.All other selects probability it is contemplated that be zero, and entropy is compiled Code/decoding can adjust at encoder and decoder simultaneously.

Additionally, as remembered with regard to step 464 above, the Prediction Parameters being referred to as merging methods and results can represent and current block The Prediction Parameters full set being associated, maybe can represent a subset of these Prediction Parameters, and such as pin is used for multiple hypothesis The Prediction Parameters of one hypothesis of the block of prediction.

It has been observed that the syntactic element about pooling information can use context modeling to carry out entropy code.Syntactic element can be by Aforementioned merge_flag and merge_left_flag forms (or similar syntactic element).In an instantiation, three contexts One in model or context can be in step 454 for coding/decoding merge_flag.The context model index used Merge_flag_ctx can lead as follows and calculate: if candidate block set contains two elements, then the value of merge_flag_ctx is Value sum equal to the merge_flag of two candidate block.But, if candidate block set contains element, then a merge_ The value of flag_ctx is equal to the twice of the merge_flag value of this candidate block.Each merge_ because of neighbor candidate block Flag can be 1 or 0, has three contexts to use for merge_flag.Merge_left_flag can only use single probability mould Type encodes.

But according to alternate embodiment, different context model can be used.Such as, nonbinary syntactic element can map to two Hex notation sequence (so-called storehouse).The context model of the some syntactic elements or syntactic element storehouse that define pooling information can base Lead calculate in the syntactic element of the adjacent block transmitted or candidate block number or other measured value, other grammer simultaneously Element or syntactic element storehouse can encode with fixing context model.

The description merged about block above, notably candidate block set can also be to described in aforementioned any embodiment Same way is led and is calculated and have following correction: candidate block is limited to use motion compensated prediction or the block of interpretation.Have those only Element can be the element of candidate block set.The signalisation of pooling information and context modeling can be carried out in the foregoing manner.

Turn to the combination segmenting embodiment with reference to aforementioned multiway tree and the merging aspect presently described, if this image is to pass through Use the square block being divided into the size such as not based on quaternary tree sub-structure, such as merge_flag and merge_left_ Flag or other show merge syntactic element can with the Prediction Parameters transmitted for each leaf node of quad-tree structure interleave. Consider such as Fig. 9 A once again.Fig. 9 A shows that an image is subdivided into the example of variable-size prediction block based on quaternary tree.Maximum chi Very little upper two blocks are so-called tree block, that is the prediction block that it is maximum possible size.Other block in this figure It is to obtain the segmentation for its corresponding tree block.Block is denoted as " X " at present.All shade block is before current block Coding/decoding, therefore it forms cause and effect block sets.As described, only in the calculation of leading for the one candidate block set in embodiment Direct (that is top or left) the neighbouring sample having the upper left sample position containing current block just can become candidate block set Member.So, current block merges block " A " or block " B ".If merge_flag is equal to zero (for "false"), current district Block " X " does not merge any one in two blocks.If block " A " and " B " have identical Prediction Parameters, then need not distinguish, Reason is to merge with any one in two blocks to cause identical result.Therefore, in such cases, merge_ is not transmitted left_flag.Otherwise, if block " A " and " B " have different Prediction Parameters, merge_left_flag=1 (for "true") will Merge block " X " and " B ", and merge_left_flag will merge block " X " and " A " equal to 0 (for "false").At another relatively In good embodiment, extra neighbouring (transmission) block represents merging candidate.

Fig. 9 B shows another example.Block " X " and left adjacent block " B " are tree block at present herein, that is it has Allow greatly block size.The size on top adjacent block " A " is 1/4th of tree block size.Belong to the unit of cause and effect block sets The block of element adds shade.Noting according to the one in preferred embodiment, current block " X " closes with two blocks " A " or " B " only And, and do not merge with other top adjacent block any.In a further preferred embodiment, the most adjacent (transmission) block table Show merging candidate.

Before this aspect of different sample arrays about how processing image according to the embodiment of the present application, palpus Notice that relevant multiway tree discussed above segments, and on the one hand signalisation and another aspect merging aspect obviously these aspects can carry For the advantage inquired into independently of one another.In other words, as noted above, multiway tree segmentation has specific advantages with the combination merged, but Advantage also comes from alternative, merges feature herein and e.g. implements to be finely divided by subdivider 30 and 104a, and Be not based on quaternary tree or multiway tree segmentation, be on the contrary with these macro zone block rule subregions become smaller subregion macro zone block segmentation Corresponding.On the other hand, the combination that multiway tree segmentation is transmitted together with the maximal tree block size within bit stream, and multiway tree is thin The use of point corresponding coding parameter sequentially transferring block together with depth-first traversal has and merges spy with use the most simultaneously Levy independent unrelated advantage.Generally, intuitively consider when sample array Encoding syntax is to not only allow for segmenting a block, with Time allow also to merge two or more segmentations after the mode of block that obtained when extending, code efficiency improves, it may be appreciated that merge Advantage.Result, it is thus achieved that one group of block its be to encode with identical Prediction Parameters.The Prediction Parameters of this group block need only encode one Secondary.Additionally, about the merging of sample set, once again it is understood that the sample set considered can be rectangle block or square block, In the case of this kind, merge sample set and represent rectangle block and/or the set of square block.It addition, the sample set considered For arbitrarily shaped image district, and merge sample set and represent the set in arbitrarily shaped image district.

The focus being hereinafter described is the different samples of one image when each image has more than one sample array The process of array, in the most secondary description, some aspects of outline are unrelated advantage independent with the segmentation kind used, that is Whether independent the most unrelated based on multiway tree segmentation with segmentation and with whether use merging independent the most unrelated.Start from the relevant image of description not Before specific embodiment with the process of sample array, the main themes of the present embodiment is A brief introduction each image difference sample number The process field of group.

Focus discussed hereinafter concentrates in image or Video coding application purpose, in the different sample arrays of an image Coding parameter between block, between the different sample arrays of a special image, the mode of adaptive forecasting coding parameter is applied to such as scheme The encoder of 1 and Fig. 2 and decoder or other image or video coding environment.It has been observed that sample array representation and different colours The sample array that component is associated, or the image associated with extraneous information (such as transparence information or depth map image).With The sample array that the chrominance component of image is relevant is also referred to as color plane.The technology being hereinafter described is also referred to as adopting across plane With/prediction, can be used on image based on block and video encoder and decoder, by this for the sample array district of an image The processing sequence of block is random order.

Image and video encoder typical case are to be designed for coding colour image (rest image or video sequence image).Color Color image includes multiple color plane, and it represents the sample array of different chrominance component.Often, to be encoded as one bright for coloured image Degree plane and the sample array set of two colorimetric plane compositions, the latter shows color difference components herein.In some applications, also Common coded samples array set is made up of three color planes of the sample array representing primary colors red, green and indigo plant.This Outward, in order to improve colored expression kenel, coloured image can be made up of more than three color plane.Additionally, an image can with show The aid sample array of the extraneous information of this image is associated.Such as, these aid sample arrays can be to show correlated color The sample array (being suitable for showing to show purpose) of the transparency of sample, or be that the sample array showing depth map (is suitable for using Present multiple sight line, such as, show for 3D).

In conventional image and video encoding standard (the most H.264), color plane encodes typically together, thus specific coding Parameter (such as macro zone block and secondary macro zone block predictive mode, benchmark index and motion vector) is for whole colored point of a block Amount.Luminance plane, it is contemplated that be a color plane, shows specific coding parameter in bit stream；And colorimetric plane can be considered secondary Plane, corresponding coding parameter is from a luminance plane presumption.Each luma blocks and same district in this image of expression Two colourity blocks are associated.According to the chroma used, chroma sample array is than the brightness for a block Sample array is less.For each macro zone block being made up of a luminance component and two chromatic components, use divide into less Type block (if macro zone block is through segmentation).For each block being made up of a luma samples block and two chroma sample blocks (can be macro zone block itself or the sub-block for macro zone block), uses identical Prediction Parameters set, such as benchmark index, motion ginseng Number and once in a while interior-predictive mode.(such as at 4:4:4 profile H.264) in the contoured of convention video coding standard, can The different color planes of absolute coding one image.In the configuration, can separate for the chrominance component of a macro zone block or sub-block Select macro zone block subregion, predictive mode, benchmark index and kinematic parameter.According to conventional coding standard, whole color planes are to make Encode together with identical specific coding parameter (such as subdivided information and Prediction Parameters) set, or be respective for whole color planes It is completely independent coding.

If color plane is to encode together, a set of segmentation and Prediction Parameters must be used for whole colored point of a block Amount.So guaranteeing that side information maintains in a small amount, but compared to absolute coding, may cause the reduction of code efficiency, reason is Use different blocks to decompose and Prediction Parameters different color components, the attenuating of rate-distortion cost may be caused.For example, Use different motion vector or reference frame for chromatic component, the residual signals energy of chromatic component can be substantially reduced, and increase it Overall coding efficiency.If color plane is absolute coding, then coding parameter (such as block partition, benchmark index and kinematic parameter) can Separately select for each chrominance component optimization code efficiency for each chrominance component.But chrominance component can not be used Between redundancy.The polynary transmission of specific coding parameter causes the increase (compared with assembly coding) of side information rate, this kind really Improve side information rate may overall coding efficiency be adversely affected.Additionally, in existing video encoding standard (such as H.264), in, supporting that aid sample array is limited to aid sample array is to use the coding parameter sets coding of itself.

So, to so far, in described whole embodiments, the plane of delineation can be such as aforementioned processing, but as previously discussed, many The overall coding efficiency (may be relevant from different color planes and/or aid sample array) of the coding of individual sample array may increase Height, now can determine the most all to compile with identical coding parameter for whole sample arrays of a block based on block benchmark (such as) Code, or whether use different coding parameter.Hereinafter the basic conception across planar prediction such as allows to make this based on block Plant adaptive decision.Such as based on rate distortion criterion, the optional all or part of sample number for a particular block of encoder Whether group uses identical coding parameter to encode, or whether uses different coding parameter coding for different sample arrays.Pass through pin To a specific sample array block, signalisation is never with the encoded common location block whether specific volume of inference of sample array Code parameter, it is possible to reach this and select.May be for the image configurations difference sample array in group, it is also referred to as sample Array group or plane group.Each plane group can be containing one or more sample arrays of an image.Then, in a plane The sample array block of group internal shares identical selected coding parameter, such as subdivided information, predictive mode and residual coding mould Formula；And other coding parameter (such as conversion coefficient level) is the separately biography of each sample number group for this plane group internal Defeated.One plane group is to be encoded to a secondary flat group, that is does not estimate or predictive coding parameter from other plane group. For each block of two secondary flat groups, adaptability selects whether a new set of selected coding parameter is transmitted, or selected coding Whether parameter estimates or prediction from a plane set or another secondary plane set.For coding parameter selected by particular block it is No be presumption or prediction decision-making be to include in this bit stream.The folding between side information rate and forecast quality is allowed across planar prediction Inner feelings, has more high-freedom degree compared to the present image coding being made up of multiple sample arrays.Advantage is relative to by multiple samples The normal image coding that this array is formed, code efficiency improves.

Use/predict expansible image or video encoder, the image of such as previous embodiment or Video coding in plane Device so that can be for a color catalog array or a block of aid sample array or color catalog array and/or aid sample One set of array, whether the coding parameter sets that adaptability is selected is from other sample array in same image The most encoded common location block estimates or prediction, or whether the coding parameter sets selected for this block is encoded separately And the common location block of not other sample array with reference within same image.The coding parameter sets selected whether pin One sample array block or multiple sample array block are estimated or the decision-making of prediction can include in this bit stream.Relevant to an image The different sample arrays of connection need not have formed objects.

It has been observed that the sample array being associated with an image (sample array can represent chrominance component and/or aid sample number Group) two or multiple so-called plane groups can be arranged in, the most each plane group is made up of one or more sample array.It is contained in The sample array need not have equal sizes of specific plane group.Notice that this kind is arranged in plane group and includes each sample array quilt Situation about being encoded separately.

More clearly saying it, according to an embodiment, for each block of a plane group, whether adaptability selects encoded block Show how a block estimates or prediction from the most encoded common location block of the Different Plane group for same image, Or whether these coding parameters are encoded separately for this block.Show that the coding parameter how a block is predicted includes following coding One or more in parameter: Block predictions pattern shows to use any prediction (interior prediction, to use single fortune for this block Moving vector and reference picture across-predict, use two motion vectors and reference picture across-predict, use high-order across-pre- Survey, that is non-translational motion model and single reference picture, use multiple motion model and reference picture across-prediction), interior- In predictive mode indicates how to produce-prediction signal, an identifier shows that the combination of how many prediction signal produces for this block Final prediction signal, benchmark index show which (which) reference picture for motion compensated prediction, kinematic parameter is (such as Motion vector or affine motion parameters) show how prediction signal uses reference picture to produce, an identifier shows with reference to figure Motion compensated prediction signal is produced as how to be filtered.Noting generally, a block can be only sub with the one of described coding parameter Collection is associated.For example, if Block predictions pattern shows that a block is interior-prediction, then the coding parameter of a block can be extra In ground includes-predictive mode, but do not show coding parameter, such as indicate how to produce the benchmark index across-prediction signal and fortune Dynamic parameter；If or Block predictions pattern shows that the coding parameter being then associated can include benchmark index and fortune extraly across-prediction Dynamic parameter, but in not showing-predictive mode.

One in two or more plane groups can this bit stream in-line coding or instruction as a secondary flat group.Pin Whole blocks to this secondary flat group, show coding parameter that how prediction signal to produce through transmission not with reference to same figure Other plane group of picture.Remaining plane group is encoded as two secondary flat groups.For each block of two secondary flat groups, pass One or more syntactic element defeated, this syntactic element signalisation shows whether the coding parameter how this block is predicted puts down from other The common location block presumption of face group or prediction, or whether transmit the one of these coding parameters for this block and newly gather.One One in individual or multiple syntactic element can be referred to as across planar prediction labelling or across planar prediction parameter.If syntactic element signal Notice does not estimates or predicts corresponding coding parameter, then the corresponding coding parameter of this block be newly integrated into this bit stream in pass Defeated.If the corresponding coding parameter of syntactic element signalisation is through presumption or prediction, then determine in so-called reference plane group In jointly position block.Appointment for the reference plane group of this block can assemble in many ways.In one embodiment, It is to specify for each two secondary flats group with particular reference to group；This kind of appointment can be fixing, or can be (all in high-order grammatical structure Such as parameter sets, access unit header, image header or sheet header) in signalisation.

In a second embodiment, the appointment of reference plane group is in bit stream in-line coding, and by compiling for a block One or more syntactic element signalisations of code show that whether selected coding parameter is through estimating or predicting or whether divide Begin the compilation of code.

In order to be readily understood by associating across planar prediction and the aforementioned possibility of embodiment shown in detail below, with reference to Figure 11, show The image 500 that the display of meaning ground is made up of three sample arrays 502,504 and 506.Being easier to understand for asking, Figure 11 only shows The sub-portion of sample array 502-506.Sample array be shown as walking back and forth Buddhist its be in alignment with each other in space so that sample array 502- 506 is overlapping along direction 508 each other, and the sample of sample array 502-506 causes whole sample array along direction 508 projection result The sample of 502-506 is the most spatially to be properly positioned.In other words, plane 502 and 506 is in the horizontal direction and vertical direction Launch adjust its spatial resolution of adaptation each other and be in alignment with each other.

According to an embodiment, whole sample arrays of an image belong to a same part for space scene, wherein along Vertical Square To and the resolution of horizontal direction can be different between independent sample array 502-506.Additionally, in order to for illustrative purposes, Sample array 502 and 504 is considered to belong to a plane group 510, and sample array 506 is considered to belong to another plane group Group 512.Additionally, Figure 11 shows example case, the spatial resolution along the trunnion axis of sample array 504 is in sample array herein The twice of the resolution of the horizontal direction of 502.Additionally, sample array 504 is considered to form a number relative to sample array 502 Group, sample array 502 forms the subordinate array being relevant to an array 504.As before, in such cases, as passed through Fig. 1 Subdivider 30 determines, it is to be used by subordinate array 502 that sample array 504 is subdivided into multiple block, wherein according to Figure 11's Example, because of the vertical direction resolution half that vertical resolution is an array 504 of sample array 502, each block is Half-and-half be divided equally into two horizontal Tile blocks, in units of the sample position in sample array 502 measure time, each block due to Half-and-half thus become square block once again.

As Figure 11 illustrate, to the segmentation selected by sample array 506 be the segmentation with another sample group 510 not With.It has been observed that subdivider 30 can select the thin of array of pixels 506 dividually or independently with the segmentation of plane group 510 Point.Certainly, the resolution of sample array 506 also can be different from the resolution of the plane 502 and 504 of plane group 510.

Now, when encoding indivedual sample array 502-506, encoder 10 starts code plane group the most in the foregoing manner Array 504 of group 510.Block shown in Figure 11 can be such as aforementioned prediction block.It addition, block can be define granularity for Define residual error block or other block of some coding parameter.It is not limited by quaternary tree or multiway tree segmentation across planar prediction, but Quaternary tree or multiway tree segmentation are illustrated in Figure 11.

After the syntactic element of an array 504 transmits, encoder 10 can determine to announce that an array 504 is subordinate plane The reference plane of 502.Encoder 10 and withdrawal device 30 can carry out signalisation via bit stream 22 respectively, and this determines, simultaneously from sample Array 504 forms the brightest wherein relatedness of the fact that an array of plane group 510, and this information transfers again alternatively position A part for stream 22.Generally speaking for each block within sample array 502, inserter 18 or encoder 10 any its Its module can decide whether together with inserter 18 coding parameter suppressing this block in the transfer within bit stream, and in bit stream Portion's signalisation and replace for this block bit stream internal signal notice be used within an array 504 common The coding parameter of location block substitutes；Or determine that the coding parameter at the common location block within an array 504 whether will It is used as the prediction of the coding parameter of the current block of sample array 502, and only transfers at bit stream internal needle this sample array Its residual error data of the current block of 502.In the case of negative decision-making, coding parameter is as usual to be transferred inside data stream.Pin It is signalisation in data stream 22 to each block decision-making.In decoder end, withdrawal device 102 use this kind for each block across Planar prediction information obtains the coding parameter of the individual block of sample array 502 accordingly, in other words, if across plane use/pre- Measurement information prompting uses/prediction across plane, then by the common coding parameter positioning block of array 504 of inference, or separately Outward from this data stream extract this block residual error data and by this residual error data with derive from an array 504 jointly position block Coding parameter prediction combination；Or as usual independent unrelated with an array 504, the current block of extraction sample array 502 Coding parameter.

Also it has been observed that reference plane are not limited to reside in this block place identical bits paid close attention at present across planar prediction puts down Face.The most as before, plane group 510 represents a secondary flat group or the reference plane group of two secondary flat groups 512. In such cases, bit stream may contain a syntactic element, and this syntactic element indicates aforementioned one for each block of sample array 506 Adopting of the coding parameter of the common location macro zone block of any plane 502 and 504 of secondary flat group or reference plane group 510 With/predict whether to carry out, in the case of aftermentioned, the coding parameter of the current block of sample array 506 is transmission as usual.

Notably segmentation and/or Prediction Parameters for the multiple planes at a plane group internal can be identical, Yi Jiyou In it, plane group is only encoded once (whole two secondary flats of a plane group are from approximately the same plane group internal Secondary flat presumption subdivided information and/or Prediction Parameters), subdivided information and/or the adaptive forecasting of Prediction Parameters or interference be Carry out between multiple plane groups.

Notably reference plane group can be a secondary flat group or two secondary flat groups.

Common location between the Different Plane block of a plane group internal is readily understood by the thin of a sample array 504 Dividing is to be used in space by subordinate sample array 502, but the segmentation of aforementioned block makes used leaf block become square block Except.In the case of Different Plane group span plane uses/predicts, common location can define in one way, thus permits Permitted the more high-freedom degree between the segmentation of these plane groups.This reference plane group given, determines in this reference plane group Portion positions block jointly.Common location block and leading of reference plane group are calculated and can be entered by the similar method being explained later OK.One in the sample array 506 of selected two secondary flat groups 512 is in the specific sample 514 within current block 516.Mesh The upper left sample of front block 516 is also same, as Figure 11 is shown in 514 for illustrative purposes；Or in the sample of current block 516 Close to other sample any within current block 516 central authorities or current block, its geometry is through distinct definition.Calculate in ginseng Examine the position of this kind within sample array 502 and 504 selected sample 515 of plane group 510.In sample array 502 and 504 Sample 514 position in portion indicates respectively in 518 and 520 in Figure 11.Which plane is actually used in reference plane group 510 502 and 504 through predetermining or can notify at bit stream internal signal.Determine the corresponding sample in reference plane group 510 The internal sample closest to position 518 and 520 of array 502 or 504, the block indivedual sample number of selected conduct containing this sample Group 502 and 504 within jointly position block.In case of fig. 11, respectively block 522 and 524.It is used in other plane Determine that the possible alternative of common location block is as describing in detail after a while.

In one embodiment, the coding parameter showing the prediction of current block 516 is the Different Plane in identical image 500 The internal corresponding Prediction Parameters using common location block 522/524 of group 510 estimates completely and does not transmits extra side letter Breath.Presumption can include replicating merely corresponding coding parameter, or the adjustment of coding parameter adapts to current plane group 512 and ginseng Examine the difference between plane group 510 taken into consideration.Lifting an example, this kind adjusts adaptation and can include adding motion parameters correction (example As motion vector corrects) for considering the phase contrast between brightness and chroma sample array；Or adjustment adaptation can include amendment motion The precision (such as revising the precision of motion vector) of parameter considers brightness and the different resolution of chroma sample array.Additionally In embodiment, it is used for showing that one or more the estimated coding parameter that prediction signal produces is not directly used in current block 516, it is used as the prediction of corresponding coding parameter into current block 516 on the contrary, and these coding parameters of block 516 at present Refined be to transmit in bit stream 22.Lifting example, the most directly use the kinematic parameter estimated, show between kinematic parameter on the contrary is inclined The kinematic parameter poor (such as motion vector is poor) of difference is for current block 516, and the kinematic parameter of presumption is to be encoded in bit stream；? Decoder end, obtains actually used kinematic parameter via the kinematic parameter of combination presumption and the kinematic parameter difference of transmission.

In another embodiment, the segmentation of a block, the most aforementioned prediction is subdivided into the tree block of prediction block (even if also Sample block with identical Prediction Parameters set) it is from the Different Plane group according to Fig. 6 A or 6B identical image that is bit sequence The most encoded jointly positions block and estimates adaptively or predict.In one embodiment, in two or more plane groups One be to be encoded to a secondary flat group.For whole blocks of this secondary flat group, transmit subdivision parameter and do not estimate Other plane group in same image.Remaining plane group is to be encoded to two secondary flat groups.For two secondary flat groups Block, transmits one or more syntactic element, and whether signalisation subdivided information positions block jointly from other plane group Presumption or prediction, or whether subdivided information is in this bit stream.One in one or more syntactic element can be referred to as across plane Predictive marker or across planar prediction parameter.If syntactic element signalisation does not estimates or predicts subdivided information, then this block is thin Point information is not address other plane group of same image at this bit stream.If this segmentation of syntactic element signalisation Information is through presumption or prediction, then determine jointly to position block in so-called reference plane group.The reference of this block is put down The configuration of face group can assemble in many ways.In one embodiment, one it is assigned to each secondary with particular reference to plane group Plane group；This specifies can be fixing, or can be in high-order grammatical structure signalisation as parameter sets, access unit report Head, image header or sheet header.In the second embodiment, the appointment of reference plane is in bit stream in-line coding, and by one Or multiple syntactic element signalisation, these syntactic elements be for a block coding to show subdivided information whether through presumption or Predict or be encoded separately.Reference plane group can be a secondary flat group or other two secondary flats group.Given reference plane group Group, determines jointly to position block at this reference plane group internal.Common location block is identical corresponding to current block The reference plane group of image area, or represent the reference plane group internal sharing this image area with the largest portion of current block Block.Common location block can be partitioned into smaller prediction block.

In Additional examples of composition, the subdivided information of current block, such as according to the segmentation based on quaternary tree of Fig. 6 A or 6B Information is used in the subdivided information presumption of the common location block of the Different Plane group of same image, and does not transmits extra Side information.Lifting a particular instance, if location block is partitioned into 2 or 4 prediction block jointly, then this current block also divides District becomes 2 or 4 sub-block in order to predict purpose.As for another particular instance, if location block is partitioned into four Ge Zi districts jointly One in block, and these sub-block is further partitioned into four smaller sub-block, then block is also partitioned into four at present One (corresponding to the common further decomposer of this sub-block positioning block) in sub-block and these sub-block is also partitioned Become four smaller sub-block.In another embodiment, the subdivided information of presumption is not directly used in current block, on the contrary Being used as the prediction of the actual subdivided information for current block, corresponding refined information is in bit stream.Lift an example, by altogether The subdivided information estimated with location block can refine further.It is not partitioned into smaller district in common positioning area block Each sub-block that one sub-block of block is corresponding, syntactic element can encode in bit stream, and it shows that whether sub-block is the most flat Face group decomposes further.The transmission of this kind of syntactic element can be using the size of sub-block as condition.Or can in bit stream signal Notify that the sub-block at the further subregion of reference plane group is not further partitioned into smaller block in current plane group.

In another embodiment, a block to the segmentation of prediction block and shows the coding parameter two how sub-block is predicted Person is the most encoded common location block adaptability presumption from the Different Plane group for same image or prediction.At this In the preferred embodiment of invention, the one in two or more plane groups is that coding is as a secondary flat group.For this kind Whole blocks of one secondary flat group, subdivided information and Prediction Parameters are not pass with reference to other plane group of same image Defeated.Residue plane group is encoded to two secondary flat groups.For the block of two secondary flat groups, transmit one or more grammer unit Whether element, its signalisation subdivided information and Prediction Parameters estimate or prediction from the common location block of other plane group；Or Whether subdivided information and Prediction Parameters be at bit stream.One in one or more syntactic elements can be referred to as across planar prediction Labelling or across planar prediction parameter.If syntactic element signalisation subdivided information and Prediction Parameters are without presumption or prediction, then The subdivided information of this block and the Prediction Parameters of result resulting bottle block are not address its of identical image in this bit stream Its plane group.If syntactic element signalisation is through presumption or prediction for subdivided information and the Prediction Parameters of this sub-block, Then determine this so-called reference plane group positions block jointly.Appointment for the reference plane group of this block is permissible Various ways assembles.In one embodiment, one it is assigned to each two secondary flats group with particular reference to plane group；This kind is assigned Can be to fix or can lead in high-order grammatical structure (such as parameter sets, access unit header, image header or sheet header) signal Know.In a second embodiment, the appointment of reference plane group is in bit stream in-line coding, and by for the one of a block coding Individual or multiple syntactic element signalisations show that whether subdivided information and Prediction Parameters are through estimating or predicting or be encoded separately.Ginseng Examining plane group can be a secondary flat group or other two secondary flats group.Given reference plane group, determines to put down in this reference Face group internal jointly position block.This positions block jointly can be identical with current block in reference plane group The block that image area is corresponding, or represent and share largest portion image area at this reference plane group internal and this current block This block of block.Common location block can be divided into smaller prediction block.In the preferred embodiment, for this current block Subdivided information and the Prediction Parameters of resulting bottle block be used in the Different Plane group of identical image common positioning area The subdivided information of block and the Prediction Parameters of corresponding sub-block, and do not transmit extra side information.As particular instance, if jointly Location block is divided into 2 or 4 prediction block, then at present block be also divided into 2 or 4 sub-block for predicting purpose, and Prediction Parameters for the sub-block of current block is to lead calculation as aforementioned.Lift another particular instance, if location block quilt jointly It is divided into four sub-block, and the one in these sub-block is further partitioned into four smaller sub-block, then current district Block is also divided into four sub-block, and one (this sub-district being further divided with common positioning area block in these sub-block The corresponding person of block) be also divided into four smaller sub-block, but further the Prediction Parameters of whole sub-block of subregion be as Presumption described above.In another embodiment, the common location that subdivided information is based entirely in reference plane group The subdivided information presumption of block, but the Prediction Parameters of this sub-block presumption only serves as actual prediction parameter pre-of sub-block Survey.Deviation between actual prediction parameter and presumption Prediction Parameters is stream encryption in place.In another embodiment, the segmentation letter of presumption Breath is used as the prediction of the actual subdivided information into current block, and difference is to transmit in bit stream (described above), but Prediction Parameters Completely through presumption.In another embodiment, the subdivided information of presumption and both Prediction Parameters of presumption are used as predicting, and Difference and presumed value thereof between actual subdivided information and Prediction Parameters are at bit stream.

In another embodiment, for a block of a plane group, adaptability selects residual coding pattern (such as to convert Type) whether or predict from the most encoded common location block presumption of Different Plane group for identical image, or residual error is compiled Whether pattern is encoded separately for this block.This embodiment be analogous to the aforementioned adaptability for Prediction Parameters estimate/ The embodiment of prediction.

In another embodiment, a block (such as one prediction block) be subdivided into transform blockiis (that is application two dimension become The sample block changed) it is the most encoded common location block adaptability presumption from the Different Plane group for same image Or prediction.The present embodiment is the embodiment of the similar aforementioned adaptability presumption/prediction being subdivided into prediction block.

In another embodiment, a block is subdivided into the residual coding pattern of transform blockiis and gained transform blockiis (such as Alternative types) it is the most encoded common location block presumption from the Different Plane group for same image or prediction.This Embodiment is the adaptability presumption/prediction and the Prediction Parameters for gained prediction block being similar to and being subdivided into prediction block above Embodiment.

In another embodiment, a block be subdivided into prediction block, the segmentation of the Prediction Parameters that is associated, prediction block letter Breath and be from the most encoded common of the Different Plane group for same image for the residual coding pattern of this transform blockiis Location block adaptability presumption or prediction.The present embodiment represents the combination of previous embodiment.It is likely to only to estimate or predict described A part in coding parameter.

So, afore-mentioned code efficiency can be improved across plane employing/prediction.But by using/predict gained to encode across plane Efficiency gain also can obtain based on multiway tree other block segmentation of being used of segmentation, and merges nothing with whether implementing block Close.

The previous embodiment just adapting to across plane/predicting can be applicable to image and video encoder and decoder, and it is by one The aid sample array that the color plane of image and (if present) are associated with this image be divided into block and by these blocks with Coding parameter is correlated with.For each block, a coding parameter sets can include at bit stream.Such as, these coding parameters can be to describe In the parameter how decoder end one block is predicted and decoded.As particular instance, coding parameter can represent macro zone block or block Predictive mode, subdivided information, interior-predictive mode, for the benchmark index of motion compensated prediction, kinematic parameter such as displacement to Amount, residual coding pattern, conversion coefficient etc..The different sample arrays being associated from an image can have different size.

Strengthen for coding parameter inside splitting scheme based on tree referring to figs. 1 to described in Fig. 8 above it follows that describe One scheme of signalisation.As for other scheme, that is merge and use/prediction across plane, strengthening signalisation scheme (hereinafter In commonly referred to as inheriting) effect and advantage be and previous embodiment independent description, but aftermentioned scheme can with in previous embodiment Any one or alone or in combination formula combination.

Generally, the inside at a splitting scheme based on tree is used for encoding the improvement encoding scheme of side information (referred to as For inheriting, it is described as follows) allow to process relative to conventional coding parameter to obtain following advantages.

In conventional image and Video coding, image or the specific sample array set for image are typically dissected into multiple Block, these blocks are to be associated with specific coding parameter.Image is typically to be made up of multiple sample arrays.Additionally, image is also Can associate extra aid sample array, it such as can be shown that transparent information or depth map.The sample array of one image (includes auxiliary Sample array) one or more so-called plane group can be grouped into, each plane group is by one or more samples herein Array forms.The plane group of one image can absolute coding, if or this image be to be associated with more than one plane group, then one The plane group of image can predict from other plane group of same image.Each plane group is typically dissected into multiple block. This block (or corresponding block of sample array) be by image prediction or image-prediction and predict.Block can have Different size and can be square or rectangle.One image is partitioned into multiple block and can be fixed by grammer, or can (at least partly) Notify at bit stream internal signal.Often the syntactic element signalisation of transmission has the segmentation of predefined size block.These grammers Element can be shown that whether a block segments and how to be subdivided into smaller block, and associates with coding parameter for such as predicting mesh 's.For whole samples (or block of corresponding sample array) of a block, the decoding of the coding parameter that is associated is with predetermined Mode shows.In this example, whole samples of a block are the identity set predictions using Prediction Parameters, such as benchmark index (identifying the reference picture in coded image set), kinematic parameter (show between a reference picture and current image The measurement of one block motion), show the parameter of interpolation filter, interior-predictive mode etc..Kinematic parameter can have horizontal component And the motion vector of vertical component represents, or with higher order kinematic parameter (affine motion parameters that such as six compositions are formed) Represent.It is relevant to single block for may have more than a particular prediction parameter sets (such as benchmark index and kinematic parameter) Connection.In the case of this kind, for each set of these particular prediction parameters, produce for this block that (or sample array is corresponding Block) single medium range forecast signal, final prediction signal is by including that the combination of overlapping medium range forecast signal is set up.Relatively Answer weighting parameters and may also constant offset (adding to this weighted sum) can be for an image or a reference picture or a reference picture Gather and fix；Maybe can include in the Prediction Parameters set of corresponding block.Original block (or corresponding sample array Block) and its prediction signal between difference be also referred to as the most transformed as residual signals and quantify.Often two-dimensional transform is applied extremely Residual signals (or corresponding sample array of residual error block).For transition coding, use particular prediction parameter sets Block (or corresponding sample array block) can be in the segmentation that takes a step forward applying this conversion.Transform blockiis can be equal to or less than using Block in prediction.Be likely to a transform blockiis include for prediction block in more than one.Different transform blockiis can have Having different size, transform blockiis can represent square block or rectangle block.After the conversion, gained conversion coefficient is quantified, and obtains Obtain so-called conversion coefficient level.Conversion coefficient level and Prediction Parameters and (if present) subdivided information are to be entropy encoded.

According to some images and video encoding standard, provided by grammer and an image (or a plane group) is subdivided into many The probability of individual block is extremely limited.Generally only indicate whether that (and the most how) has the block of predefined size and can segment Become smaller block.For example, it is 16 × 16 according to largest block H.264.This 16 × 16 block is also referred to as Hong Qu Block, becomes macro zone block at each picture portion of first step.For each 16 × 16 macro zone block, whether signal notice is encoded into 16 × 16 blocks, or it is encoded into two 16 × 8 blocks, or two 8 × 16 blocks, or four 8 × 8 blocks.If 16 × 16 blocks are thin Be divided into four 8 × 8 blocks, then these 8 × 8 blocks each can be encoded into 8 × 8 blocks, or two 8 × 4 blocks, or two 4 × 8 blocks or four 4 × 4 blocks.Current image and video encoding standard show segment the probability becoming multiple block Minimal set have and be maintained little advantage with the side information rate signaling to subdivided information, but there is transmission for district Bit rate required by the Prediction Parameters of block becomes big shortcoming, as describing in detail after a while.Believe with the side signaling to Prediction Parameters The fixed rate of interest represents the total bit rate of significant quantity of a block the most really.And when this side information reduces, such as, can use relatively large district When block size is reached, code efficiency can be improved.The actual image of one video sequence or image are by the arbitrary shape of tool special properties Shape object forms.Lifting example, these objects or object part are to be its feature with unique texture or unique motion.The most identical Prediction Parameters set can be applicable to this kind of object or object part.But object bounds does not the most overlap, large-scale prediction block may Block border (such as in 16 × 16 macro zone blocks H.264).Encoder generally determines to cause smallest particular rate-distortion cost The segmentation (in limited probability set) of measured value.For arbitrary shape of object, so may result in great quantity of small block.And because of These small-sized blocks are to be associated with the Prediction Parameters set that must transmit, therefore side information rate becomes the notable portion of total bit rate Point.But because some small-sized blocks still represent that the district of same object or object part, the Prediction Parameters of multiple gained blocks are phase Same or very much like.Instinctively, when grammer is to expand in one way, it not only allows for segmenting a block, the most also in segmentation Code efficiency can be improved when sharing coding parameter between the multiple block of rear gained.In segmentation based on tree, by by with based on tree Hierarchy type relation prescribed coding parameter or its part give one or more parent nodes, can reach for a given area set of blocks Coding parameter share.As a result, share parameter or its part can be used to reduce for gained block signature notice volume after segmentation Side information needed for code parameter actual selection.Minimizing can be reached by deleting the signalisation of the parameter of block subsequently, or Can be reached by the shared parameter for the forecast model of the parameter of block subsequently and/or context model.

The basic conception of aftermentioned succession scheme is by being shared information by the social strata relation based on tree along such block, coming Reduce the bit rate needed for transmission coding information.Shared information is to notify (in addition to subdivided information) at bit stream internal signal.Inherit The advantage of scheme is for be caused code efficiency to increase for coding parameter by lowering side information rate result.

In order to reduce side information rate, according to aftermentioned embodiment, for the individual encoding parameters of specific sample set, that is Simple bonding pad, it can represent the rectangle block or square block or arbitrary shape district or other sample set any that multiway tree segments Conjunction is to notify at data stream internal signal in an efficient way.Aftermentioned succession scheme permission coding parameter need not be in sample set Each sample set include clearly at bit stream.Coding parameter can represent Prediction Parameters, and it shows that corresponding sample set is to make Use encoded sample predictions.Multinomial possible and example is also applied for herein the most really.As the most already described, and just As described in detail after a while, relevant following several schemes, the sample array of an image is divided into multiple sample set can lead to based on tree-shaped Cross grammer to fix, maybe can be by the corresponding subdivided information signalisation within bit stream.It has been observed that for the volume of sample set The sequential delivery that code parameter can define in advance, this order is to be given by grammer.

According to the scheme of succession, the withdrawal device 102 of decoder or decoder is configured to lead in a specific way and calculates relevant The information of the coding parameter of not simple bonding pad or sample set.In specific words, coding parameter or one part are (such as pre- Survey the parameter of purpose) it is to share between each block, along the shared group of this tree along this given splitting scheme based on tree Group is to be determined by encoder or inserter 18 respectively.In a specific embodiment, the one of cut tree gives the whole of internal node The shared of the coding parameter of child node is to use specific binary values to share labelling instruction.As for instead road, for each node The refined of coding parameter can be transmitted so that along the hierarchy type relation of block based on tree, the accumulation of parameter is refined can apply to Whole sample sets of one this block giving leaf node.In another embodiment, pass along this block social strata relation based on tree A defeated part for the coding parameter of internal node can be used for one give leaf node for the coding parameter of this block or its The context adaptability entropy code of a part and decoding.

Figure 12 A and Figure 12 B display uses the basic conception of the succession of special case based on Quadtree Partition.But as the most for several times Instruction, other multiway tree subdivision scheme is used as.This tree is displayed at Figure 12 A, and with the tree phase of Figure 12 A Corresponding space segmentation is displayed at Figure 12 B.Segmentation shown in it is similar with regard to shown in Fig. 3 A to 3C.It sayed in outline, succession side Case is assigned to the node at the different n omicronn-leaf layers within this tree construction by allowing side information.It is assigned at this according to side information The node of the different layers of tree, the internal node of the such as tree of Figure 12 A or its root node, at the tree hierarchy type of block shown in Figure 12 B Relation can be reached shared side information in various degree.Such as, if determining the whole leaf nodes at layer 4, in the situation of Figure 12 A The most all there is identical parent node, share side information virtually, it means that Figure 12 B indicates with 156a to 156d Small-sized block shares this side information, and is no longer necessary to for all these small-sized block 156a to 156d complete transmission sides letters Breath, that is transmit four times, but so it is maintained the option of encoder.However, it is also possible to determine the hierarchy type level 1 (layer of Figure 12 A 2) the whole district, that is tree block 150 upper right corner four/part include sub-block 154a, 154b and 154d and aforementioned again Smaller sub-block 156a to 156d, is used as the district for wherein sharing coding parameter.So, shared side information is increased District.The next level that increases is the whole sub-block adding up layer 1, that is sub-block 152a, 152c and 152d and aforementioned smaller Block.In other words, in such cases, whole tree block has the side information being assigned to this block, and this sets the whole of block 150 Sub-block shares side information.

Later inherit explanation in, following annotation be used to describe embodiment:

A. the reconstruction sample of current leaf node: r

The reconstruction sample of the most adjacent leaf: r '

C. the predictor of current leaf node: p

D. the residual error of current leaf node: Res

E. the reconstruction residual error of current leaf node: RecRes

F. calibration and inverse transformation: SIT

G. labelling: f is shared

As the first example inherited, the interior-prediction signal notice of internal node can be described in.More accurately say it, describe How signalisation in internal node based on tree block divides-predictive mode is in order to predict purpose.By saving from root Point travels through tree to leaf node, internal node (including root node) can transmitting portions side information, this information will be corresponding by it Child node utilize.More clearly saying it, sharing labelling f is to send for internal node and have following meaning:

If f has numerical value 1 ("true"), then whole child nodes of this given internal node share identical interior-predictive mode. In addition to sharing the labelling f with numerical value 1, in internal node also signalisation-the whole child node of predictive mode parameters cause Use.As a result, child node does not carry any prediction mode information and any shared labelling the most subsequently.In order to rebuild whole phase Close leaf node, decoder from corresponding internal node apply in-predictive mode.

If f has numerical value 0 ("false"), the child node of the most corresponding internal node does not share identical interior-predictive mode, Belong to each child node of internal node to carry one point and open shared labelling.

Figure 12 C show aforementioned in internal node-prediction signal notice.Internal node at layer 1 transmits by interior-prediction The given shared labelling of pattern information and side information, and child node do not carries any side information.

Inherit example as second, can describe across-prediction refined.More clearly say it, how describe at block based on tree Segmentation internal model, signalisation across the side information of-predictive mode for such as by motion vector given motion ginseng The refined purpose of number.By from root node by traveling through tree to leaf node, internal node (including root node) can transmitting portions Side information, this information will be refined by its corresponding child node.More clearly saying it, sharing labelling f is to send out for internal node Send and there is following meaning:

If f has numerical value 1 ("true"), then whole child nodes of this given internal node share same motion vector ginseng Examine.In addition to sharing the labelling f with numerical value 1, internal node also signalisation motion vector and benchmark index.As a result, all Child node does not carries extra labelling of sharing subsequently, and this motion vector references inherited of portability is refined on the contrary.For all The reconstruction of relevant leaf node, decoder adds motion vector at this given leaf node and refines to belonging to its corresponding internal parent node There is the motion vector references value of the succession of the numerical value 1 of shared labelling f.So represent in a motion vector essence giving leaf node It is made as being intended to applying to leaf node since then for the actual motion vector internal parent node corresponding thereto of motion compensated prediction Difference between motion vector references value.

If f has numerical value 0 ("false"), the child node of the most corresponding internal node is the most necessarily shared identical across-prediction Pattern, and be not through using in this child node and derive from the kinematic parameter of corresponding internal node and carry out the essence of kinematic parameter System, belongs to each child node of internal node and carries one point and open shared labelling.

Figure 12 D display aforementioned movement parameter refines.The internal node of layer 1 is that labelling and side information are shared in transmission.Belong to The child node of leaf node is only carried kinematic parameter and is refined, and the inside child node of such as layer 2 does not carries side information.

With reference now to Figure 13,.Figure 13 flow for displaying figure, illustrates decoder (decoder of such as Fig. 2) for from data Stream gravity builds a message sample array of representation space example information signal, and (it is segmented by multiway tree and is subdivided into different size of Leaf district) operator scheme.It has been observed that Ge Ye district has relative a series of hierarchical level selected from multiway tree segmentation In a hierarchical level.Such as, shown in Figure 12 B, whole blocks are all leaf district.Leaf district 156c e.g. with hierarchical level 4 (or level 3) is associated.Ge Ye district has coding parameter associated there.The example of these coding parameters have been described above as Before.For Ge Ye district, coding parameter is to represent with an individual grammar element set.Each syntactic element is selected from a grammer unit Individual syntax element type in element type set.Each syntax element type for example, predictive mode, motion vector component, The instruction etc. of interior-predictive mode.According to Figure 13, decoder carries out the following step.

In step 550, inherited information is to be drawn from data stream.In the case of figure 2, withdrawal device 102 is responsible for step 550.Whether inherited information instruction inherits for current message sample array.Be hereinafter described display inherited information is had some can Can, such as share labelling f and multiway tree structure is divided into a second part and the signalisation of two second part.

Message sample array has constituted a subdivision of an image, such as sets block, the tree block 150 of such as Figure 12 B. Whether so inherited information instruction uses succession for certain tree block 150.This kind of inherited information such as can be for all predictions Segmentation tree block and insert data stream.

If additionally, indicate and use succession, then inherited information instruction is gathered that formed by a leaf district and corresponds to multiway tree At least one of this message sample array of one hierarchical level of this hierarchical level sequence of segmentation is inherited district and is less than this Leaf district gathers each hierarchical level being associated.In other words, inherited information instruction (such as sets block for current sample number group 150) whether succession is used.If it is, represent that this at least one setting block 150 is inherited the leaf district within district or sub-district and shared volume Code parameter.So, inherit district and be not likely to be leaf district.In the example of Figure 12 B, inheriting district (such as) can be by sub-block 156a extremely The district that 156b is formed.Can be bigger it addition, inherit district, the most additionally contain sub-block 154a, b and d, and even it addition, inherit district Can be tree block 150 itself, its whole leaf blocks share the coding parameter being associated with this succession district.

But it should be noted that more than one succession district can be defined inside a sample array or tree block 150.For example, it is assumed that Lower-left sub-block 152c is also divided into smaller block.In such cases, sub-block 152c can form a succession district.

In step 552, check inherited information, if use and inherit.If it is, the processing routine of Figure 13 advances to step 554, herein in relation to each across inheriting district, the succession subset including at least one syntactic element of predetermined syntax element type is Extract from data stream.In later step 556, then this succession subset is copied within this syntactic element set relative Answering syntactic element to inherit subset, or be used as the prediction of this succession subset, this succession subset represents that at least one inherits the leaf in district District's collection is combined into the coding parameter being associated.In other words, for each succession district of instruction inside this inherited information, data stream Succession subset including syntactic element.The most in other words, succession is relevant at least some syntax element type that can be used for and inheriting Or syntactic element classification.For example, it was predicted that pattern or succession can be experienced across-predictive mode or interior-predictive mode syntactic element. Such as can include across-predictive mode syntactic element in the succession subset contained by this data stream inside for succession district.Inherit subset Also including extra syntactic element, its syntax element type is depending on the aforementioned fixed grammer unit being associated with this succession scheme The value of element type.For example, in the case of being, across-predictive mode, the fixed component inheriting subset, motion compensation is defined Syntactic element (such as motion vector component) can be included by grammer or can not include in this succession subset.Such as, false If the upper right 1/4th (that is sub-block 152b) of tree block 150 is for inheriting district, then individually across-predictive mode may indicate that for This succession district, or together with motion vector and motion vector index for across-predictive mode.

Being contained in and inheriting whole syntactic elements of subset is to be copied into leaf block (that is the leaf block within this succession district 154a, b, d and 156a to 156d) corresponding coding parameter, or be used as its prediction.In the case of using prediction, for Indivedual leaf encrypted communication residual errors.

The transmission that a probability is aforementioned shared labelling f of inherited information is transmitted for leaf block 150.In step 550, The extraction of inherited information includes aftermentioned in this example.More clearly saying it, decoder can be configured with from lower-order laminar layer Level is to the hierarchical level order of higher-order laminar level, appointing at least one hierarchical level segmented with this multiway tree What inherits the n omicronn-leaf district that set is corresponding, extracts and check the shared labelling f deriving from this data stream, about whether inherit individually mark Note or shared labelling show whether to inherit.For example, the set of inheriting of hierarchical level can be by the hierarchy type layer 1 of Figure 12 A Formed to layer 3.So, for and nonleaf node and be position any node of being the sub-tree structure of any layer 1 to layer 3, can have Have and share labelling at this data stream internal associated there one.Decoder is (such as with the degree of depth with the order from layer 1 to layer 3 Preferential or breadth first traversal order) extract these shared labellings.Once share the one in labelling and be equal to 1, then decoder is known It is contained in the leaf block in corresponding succession district dawn and shares this succession subset, followed by the extraction in step 554.For current node Child node, it is no longer necessary to inherit labelling inspection.In other words, the succession labelling of these child nodes does not pass inside data stream Defeated, reason is that the succession subset that these node area obvious already belong to wherein syntactic element is this shared succession district.

Share labelling f to intersect with the position of aforementioned signal notice quaternary tree segmentation.Such as, including subdivided mark and shared mark The intersection bit sequence remembering the two can be:

10001101(0000)000,

It is the identical subdivided information shown in Fig. 6 A, has the shared labelling of two interspersions, and this labelling is strong by lower section setting-out Transfer to indicate in Fig. 3 C and share coding parameter in the sub-block setting block 150 lower-left 1/4th.

The another way of the inherited information that district is inherited in definition instruction is to use two segmentations defined each other with slave mode, As distinguished reference prediction segmentation and the explanation of residual error segmentation above.It sayed in outline, and the leaf block once segmented can form this succession District, the succession subset of this succession regional boundary wherein syntactic element is shared district；And inside these succession districts of subordinate segment definition Block, be to be replicated or be used as prediction for the succession subset of this block syntactic element.

For example, it is contemplated that residual tree is as the extension of pre-assize.Consider that prediction block can be further divided into less further Type block is used for residual coding purpose.For each prediction block corresponding to the leaf node predicting relevant quaternary tree, for residual The corresponding segmentation of difference coding is to be determined by one or more subordinate quaternary trees.

In such cases, substituting and use any Prediction Parameters at internal node, inventor considers that residual tree is with following side Formula interprets, and residual tree also indicates that the refined expression of pre-assize uses constant predictive mode (by predicting the corresponding leaf of association tree Node signal notifies) but have through refined sample for reference.Aftermentioned example illustrates this kind of situation.

Such as, Figure 14 A and 14B shows the Quadtree Partition for interior-prediction, and neighbouring sample for reference is for once segmenting A particular leaf node emphasize, and Figure 14 B shows that the residual error quaternary tree segmentation of identical prediction leaf node is with refined ginseng Examine sample.Whole sub-block shown in Figure 14 B share be contained within data stream identical for the indivedual leaf blocks emphasized at Figure 14 A Across-Prediction Parameters.So, Figure 14 A shows the Quadtree Partition example being conventionally used for interior-prediction, shown here as a particular leaf The sample for reference of node.But in the preferred embodiment of inventor, via using leaf node the most reconstructed in residual tree Neighbouring sample (the gray shade lines instruction of such as 4 (b)), calculates an interior-prediction for each leaf node in residual tree Signal.Then, by quantization residual coding signal is added so far prediction signal and obtain with unusual manner and one give residual error leaf segment The reconstruction signal of point.Then this reconstruction signal is used as predicting subsequently the reference signal of program.It is to be noted that the decoding of prediction is suitable Sequence is identical with residual error decoding order.

As shown in figure 15, in decoding program, for each residual error leaf node, via using sample for reference r ' according to actual In-predictive mode (being indicated by the relevant quaternary tree leaf node of prediction), calculate prediction signal p.

After SIT processing routine,

RecRes=SIT (Res)

Calculate the signal r rebuild and store for next one prediction calculation procedure:

R=RecRes+p

Decoding program for prediction is identical with residual error decoding order shown in Figure 16.

Each residual error leaf node is to decode as in the previous paragraph.Reconstruction signal r is stored in buffer, as shown in figure 16.Should In buffer, sample for reference r ' will take in prediction next time and decoding program.

Before with regard to Fig. 1 to Figure 16, the combination type of each side that digest is stated is after separately subset describes specific embodiment, To describe the Additional examples of composition of the application, focus is to concentrate on some aspect aforementioned, but embodiment represents aforementioned some realities Execute the universalness of example.

In specific words, about the many aspects of previous embodiment main combination the application of framework of Fig. 1 and Fig. 2, it is possible to Excellently use in other application purpose or other coding field.As the most often addressed, such as multiway tree segmentation can non-ECDC And and/or without using across plane/predict and/or using without succession.For example, the maximum transmission of block size, the degree of depth The use of first traversing order, context according to the hierarchical level of indivedual subdivided mark adapt to and internal maximum at bit stream Side information bit rate is saved in the transmission of hierarchical level, and all these aspects are all excellent but independent of one another.When considering across plane Also it is such during Utilization plan.The advantage utilized across plane is to be subdivided into simple bonding pad butt formula really independently with an image Unrelated, and advantage is independent with the use of Merge Scenarios and/or succession the most unrelated.Be equally applicable to relate to merge and inherit is excellent Point.

Therefore, the embodiment outline below summarises about using across plane/prediction in terms of aforesaid embodiment.By Representing summary to above-described embodiment in the following examples, many above-mentioned details can be considered to can be combined in described below Embodiment.

Figure 17 shows and represents the data stream of different spaces sample intelligence component in the plane of scene image for decoding The module of decoder, each plane includes message sample array.Decoder can the decoder shown in corresponding diagram 2.Specifically, mould Block 700 is responsible for by processing such as residual error data or the such payload of spectral decomposition data, carries out the every of message sample The reconstruction of individual array 502 to 506, wherein said residual data or spectral decomposition data are relevant to each message sample array 502 The simple connection being subdivided in the way of the coding parameter regulation of the such as Prediction Parameters relevant to simple join domain to 506 Region.Such as, in the decoder situation of Fig. 2, this module can be existing by all frame blocks including frame block 102.But, figure The decoder of 17 needs not to be hybrid decoder.Across and/or interior prediction may be not used.This is equally applicable to transform coding, i.e. Residual error can be at spatial domain coding rather than by spectral decomposition two-dimensional transformations.

Another module 702 is responsible for will be with the list of the first array (array 506 of such as message sample array) from data stream The coding parameter that pure join domain is relevant derives.Therefore, module 702 defines the task for module 700 tasks carrying.? In the case of Fig. 2, withdrawal device 102 is considered to be responsible for the task of module 702.It should be noted that array 506 itself can be the second number Group, can obtain relative coding parameter by the way of using across plane/predict.

Next module 704 will be for will be used for the simple connection of the second array 504 of message sample array from data stream Deriving across plane interchange information of region is next.In the case of Fig. 2, withdrawal device 102 is considered to be responsible for the task of module 702.

Next module 706, for based on the simple join domain for the second array across plane interchange information, is the Which the suitable subset of each simple bonding pad of two arrays or simple bonding pad determines in ensuing module 708 and 710 Activate.In the case of Fig. 2, withdrawal device 102 cooperates with subdivider 104 to carry out the task of module 706.Enter at withdrawal device 102 Row reality extraction while, subdivider control travel through simple bonding pad order, i.e. across plane interchange information which part about Any part of simple bonding pad.In embodiment in more detail above, individually define across plane mutual for each simple bonding pad Change information, about whether carry out using/prediction across plane.But, this is not problem.If the suitable son in simple bonding pad Determining in the unit of collection, that is favourable.Such as, across one or more bigger the connecting merely of plane interchange information definable Meeting district, each in this bonding pad includes one or more adjacent simple bonding pad, every in the district that these are bigger One, perform once to use/prediction across plane.

Module 708 is used for: in the case of Fig. 2, by the responsible derivation co-location relation that cooperates with subdivider 104 Withdrawal device, will be used for the coding parameter of each simple bonding pad of the second array 540 or the suitable subset of simple bonding pad at least Partly derive from the coding parameter of simple bonding pad corresponding to the local of the first array 506 being performed task；And The suitable of each the simple bonding pad with the second array or simple bonding pad is decoded in the way of the coding parameter regulation so derived When the load data that subset is relevant, this task transfers to be carried out by other modules in Fig. 2, i.e. 106 to 114.

For module 708 alternatively, module 710 is used for: in the simple bonding pad that the local ignoring the first array 506 is corresponding Coding parameter while, from data stream, leading-out needle is to each simple bonding pad of the second array 504 or simple bonding pad The suitably coding parameter of subset, this task is considered to be responsible for by the withdrawal device 102 in Fig. 2；And with the phase gone out from data conductance The suitable subset that the mode that pass coding parameter specifies decodes each the simple bonding pad to the second array or simple bonding pad is relevant Payload data, relatively, this task by other modules in Fig. 2, i.e. 106 to 114, connect simple being responsible for always Carry out under the control of the subdivider 104 managing adjacent and co-location relation in district.

As described above for described in Fig. 1 to Figure 16, the array of message sample be not required to represent video image or still image or Its color component.Sample components also can represent the depth map of such as some scene or other two sampling physics numbers of transparent print According to.

As discussed above, the payload data in each district in relevant multiple simple bonding pads can include such as Residual error data in the spatial domain of conversion coefficient or Transformation Domain conversion coefficient very big with in the conversion block identifying corresponding residual block The very big figure of position.On the whole, such as, load data can be directly or as its certain in spatial domain or spectrum domain The data describing its simple bonding pad being correlated with spatially of the residual error of type prediction.In turn, coding parameter is not limited It is made as Prediction Parameters.Coding parameter may indicate that for change payload data conversion maybe can define for working as reconstruction information The wave filter that individual simple bonding pad is used is rebuild during sample array.

It has been observed that the simple bonding pad that this message sample array is subdivided into can be planted comes from a multiway tree segmentation, and can be Square or rectangular shape.Additionally, be particularly described is only specific embodiment for segmenting the embodiment of a sample array, it is possible to make Segment with other.Some probabilities are displayed at Figure 18 A to Figure 18 C.Such as Figure 18 A shows that a sample array 606 is subdivided into that The regular two-dimensional arrangement of this non-overlapped tree block 608 adjoined, wherein part tree block is subdivided into according to multiway tree structure There is different size of sub-block 610.Although it has been observed that quaternary tree segmentation illustrates in Figure 18 A, but what its number in office Each parent node segmentation of child node also falls within possibility.Figure 18 b shows an embodiment, accordingly, is segmented by directly application multiway tree To both full-pixel array 606, a sample array 606 is subdivided into has different size of sub-block.In other words, both full-pixel array 606 are regarded as setting block processes.Figure 18 C shows another kind of embodiment.According to this embodiment, sample array is configured to that The regular two-dimensional configuration of the macro zone block of this square adjoined or rectangle, and each block in these macro zone blocks 612 is independent Ground is relevant to partition information, and according to this partition information, macro zone block 612 is left not to be partitioned or be partitioned and is referred to by partition information Show the regular two-dimensional configuration of the block of size.So understanding, whole segmentations of Figure 18 A to 18C cause this sample array 606 to be subdivided into Simple bonding pad, according to the embodiment of Figure 18 A to 18C, each simple bonding pad is with non-overlapped display.But instead road is also for several Belonging to may.For example, each block can overlap each other.But overlap can be limited to some underlapped any adjacent block of each block Degree, or to make each block sample be at most in the adjacent block being arranged side by side along a predetermined direction and current block Overlapping block.In other words, the latter represent left and right adjacent block can overlapping block at present, thus this current block is completely covered, But do not overlap each other, be in like manner applicable to the vertical and adjacent block of diagonal.Such as Figure 17 still optionally further, module 606 In judge and thus across plane use/granularity that carried out of prediction can be plane.Therefore, according to further implementing , there are the plane of more than two, a principal plane and two possible secondary planes, for each possible secondary plane, module in example 606 judge respectively, and indicating respectively across plane interchange information in data stream, use/predict whether should be applied to across plane Each plane.If it is, further can be manipulated in above-mentioned simple bonding pad mode, but, wherein, mutual across plane Change information only by those graphic memories indicated across plane interchange information and be processed.

Although with regard to the several aspect of device contextual declaration, it is apparent that these aspects also illustrate that saying of corresponding method Bright, a block or a device correspond to a method step or a method step feature herein.In like manner, at method step context Described in aspect also illustrate that the description of feature of corresponding block or project or corresponding intrument.Partly or entirely method step can Performed by (or use) hardware unit, such as microprocessor, programmable calculator or electronic circuit.In several embodiments, Certain one in most important method step or certain multiple performed by this kind of device.

Coding/the compressed signal of the present invention can be stored on digital storage mediums, or can the transmission in such as the Internet be situated between Matter (such as wireless transmission medium or wired transmissions medium) is transmitted.

Implementing requirement according to some, embodiments of the invention can implement in hardware or software.Implementing can Use digital storage mediums performs, and such as floppy disk, DVD, Blu-ray disc, CD, ROM, PROM, EPROM, EEPROM or flash memory, on it Storage can the control signal that reads of electronic type, it is to pull together cooperation (maybe can pull together cooperation) with programmable computer system thus hold Row individual method.Therefore, digital storage mediums can be embodied on computer readable.

Include having electronic type can read the data medium of control signal according to some embodiment of the present invention, its can with can Computer system is pulled together cooperation thus is performed the one in method described herein.

It is said that in general, embodiments of the invention can be embodied as a kind of computer program with program code, this journey The operable one being used for performing in these methods when computer program runs on computers of sequence code.Program code Such as can be stored on machine-readable carrier.

Other embodiments include being stored on machine-readable carrier for performing one in method described herein Computer program.

In other words, therefore, the embodiment of the inventive method is a kind of computer program, and this computer program has program generation Code is used for when this computer program runs on computers the one performing in method described herein.

Therefore another embodiment of the present invention is that (or digital storage mediums or embodied on computer readable are situated between a kind of data medium Matter) include the computer program that records thereon for performing the one in method described herein.

Therefore the another embodiment of the inventive method is to represent for the computer performing the one in method described herein The data stream of program or signal sequence.This data stream or signal sequence are such as configured to connect via data communication, such as warp By the Internet transmission.

Another embodiment such as includes a kind of process means, such as computer or can program logic device, it is configured to Or adjust the one performing in method described herein.

Another embodiment includes the meter being provided with to perform the computer program of the one in method described herein on it Calculation machine.

In several embodiments, programmable logic device (such as field programmable gate array) can be used to perform described herein The part or all of function of method.In several embodiments, field programmable gate array can be held with microprocessor cooperation of pulling together One in row method described herein.It sayed in outline, and the method is preferably performed by any one hardware unit.

Previous embodiment is only used for illustrating the principle of the present invention.It is understood that the amendment of the details of configuration described herein and It is changed to skilled artisan be obviously apparent from.Therefore it is intended to only be limited by the scope of appended claims rather than by for illustrating Illustrate that the specific detail of embodiments herein is limited.

Claims

1. in decoding plane, represent a decoder for the data stream of the different spaces sample intelligence component of the image of scene, Each plane includes that message sample array, described decoder are configured to:

Each message sample array, wherein, each letter is rebuild by processing the payload data relevant to simple bonding pad Breath sample array is to be subdivided into described simple bonding pad to described stating in the way of the relevant coding parameter in simple bonding pad specifies；

The coding relevant to the described simple bonding pad of the first array in described message sample array is derived from described data stream Parameter；

From described data stream leading-out needle in described message sample array the described simple bonding pad of the second array across plane Interchange information；

According to described in the described simple bonding pad for described second array across plane interchange information, for described second array Each simple bonding pad or the suitable subset of described simple bonding pad, it is determined that:

At least in part from the coding parameter of corresponding simple bonding pad, the local of described first array leading-out needle to described second Each simple bonding pad of array or the coding parameter of the suitable subset of described simple bonding pad, and with the described volume so derived The code mode that specifies of parameter decodes the suitable of each the simple bonding pad described with described second array or described simple bonding pad The relevant described payload data of subset；Or

Ignore the coding parameter of corresponding simple bonding pad, the described local for described first array, derive from described data stream For each simple bonding pad of described second array or the coding parameter of the suitable subset of simple bonding pad, and with from described number The mode that the relevant described coding parameter gone out according to conductance specifies decodes each simple bonding pad or the list with described second array The payload data that the suitable subset of pure bonding pad is relevant.

Decoder the most according to claim 1, wherein, described decoder is configured to go out segmentation letter from described data conductance Breath, and according to described subdivided information, described message sample array is subdivided into described simple bonding pad.

Decoder the most according to claim 1 and 2, wherein, described decoder is configured to described message sample array It is subdivided into described simple bonding pad so that described simple bonding pad is of different sizes.

Decoder the most according to claim 1 and 2, wherein, described decoder is configured to described message sample array Be subdivided into described simple bonding pad so that for described first array and the second array to described simple bonding pad described carefully Divide difference.

Decoder the most according to claim 1 and 2, wherein, described decoder is configured to connect merely with described in process Connecing in the payload data that district is relevant, the mode specified according to the described coding parameter relevant to described simple bonding pad is predicted Message sample in described simple bonding pad.

6. for decoding a decoder for the data stream of the different spaces sample intelligence component of the image representing scene in plane, Each plane includes message sample array, and wherein, at least one in described message sample array is the main array of message sample, and In described message sample array at least two other is message sample pair array, described decoder is configured to:

The mode specified according to the coding parameter relevant to each message sample array, by processing and each described message sample The payload data that array is relevant rebuilds each message sample array；

The coding parameter relevant to described main array is derived from described data stream；

From described data stream leading-out needle to each secondary array across plane interchange information；

According to for described in described secondary array across plane interchange information, each secondary array is individually judged:

The leading-out needle coding parameter to array secondary each described from the coding parameter of described main array at least in part, and with such as The mode of this described coding parameter defined derived decodes the payload data that array secondary to each described is relevant；Or

Ignore the coding parameter of described main array, the leading-out needle coding parameter to array secondary each described from described data stream, And independent of the described coding parameter ignored, to decode in the way of the coding parameter defined of pair array for each described The payload data that array secondary to each described is relevant.

Decoder the most according to claim 6, wherein, described message sample array represents different luminance component, color Component and/or the degree of depth of the most described image or the non-brightness/not color component of transparency component.

Decoder the most according to claim 6, wherein, described decoder is configured to from described data stream derive plane Appointment information, and specify information that described message sample array is appointed as secondary array of advocating peace according to described plane.

Decoder the most according to claim 8, wherein, described scene is time change, and each plane includes information The sequence of sample array, the sequence of described message sample array is with every with in the image sequence of described scene of time sampling mode At least one array of the message sample of width image represents correspondence one, described decoder quilt in different spaces sample intelligence component It is configured in the time granularity of correspondence image or image sets, perform described plane and specifies derivation and the appointment of information, or only carry out Once derive and specify to initialize.

Decoder the most according to claim 6, wherein, described scene is time change, and each plane includes information The sequence of sample array, the sequence of described message sample array is with every with in the image sequence of described scene of time sampling mode At least one array of the message sample of width image represents correspondence one, described decoder quilt in different spaces sample intelligence component It is configured to the time granularity of sub-district, image or image sets being subdivided at corresponding described image performs described mutual across plane Change derivation and the judgement of information.