GB2492778A

GB2492778A - Motion compensated image coding by combining motion information predictors

Info

Publication number: GB2492778A
Application number: GB1111867.6A
Authority: GB
Inventors: Guillaume Laroche; Christophe Gisquet
Original assignee: Canon Inc
Current assignee: Canon Inc
Priority date: 2011-07-11
Filing date: 2011-07-11
Publication date: 2013-01-16
Also published as: GB201111867D0

Abstract

A sequence of digital images is encoded into a bitstream, at least one portion of an image being encoded by motion compensation (MC) with respect to a reference image portion. The encoding method comprises, for at least one image portion to encode: obtaining - S1002, S1004 - a first set of motion information predictors usable for encoding; and employing - S1008-1014 - the first set of motion information predictors and one or more predetermined motion information predictors (e.g. predictors from a fixed set S1008, added to form a combined predictor set) to obtain a second set of motion information predictors having a target number of motion information predictors or fewer (e.g., by predictor removal based on priority). An analogous decoding method is also claimed. The first set of predictors (possibly those used in Advanced Motion Vector Prediction, AMVP) may be associated with an image portion having a predetermined spatial and/or temporal relationship with the image region being encoded. Preset, predetermined predictors can be obtained from a fixed set of predictors, greater than or equal to the target; all predictors may be different in the fixed set.

Description

Video encoding and decoding with low complexity

Field of the invention

The invention relates to a method and device for encoding a sequence of digital images and a method and device for decoding a corresponding bitstream.

The invention belongs to the field of digital signal processing, and in particular to the field of video compression using motion compensation to reduce spatial and temporal redundancies in video streams.

Description of the prior-art

Many video compression formats, for example H.263, H.264, MPEG-i, MPEG-2, MPEG-4, SVC, use block-based discrete cosine transform (DCT) and motion compensation to remove spatial and temporal redundancies. They can be referred to as predictive video formats. Each frame or image of the video signal is divided into slices which are encoded and can be decoded independently. A slice is typically a rectangular portion of the frame, or more generally, a portion of a frame or an entire frame. Further, each slice is divided into macroblocks (MB5), and each macroblock is further divided into blocks, typically blocks of 8x8 pixels. The encoded frames are of two types: temporal predicted frames (either predicted from one reference frame called P-frames or predicted from two reference frames called B- frames) and non temporal predicted frames (called Intra frames or I-frames).

Temporal prediction consists in finding in a reference frame, either a previous or a future frame of the video sequence, an image portion or reference area which is the closest to the block to encode.

This step is known as motion estimation. Next, the difference between the block to encode and the reference portion is encoded (motion compensation), along with an item of motion information relative to the motion vector which indicates the reference area to use for motion compensation.

In order to further reduce the cost of encoding motion information, it has been proposed to encode a motion vector by difference from a motion vector predictor, typically computed from the motion vectors of the blocks surrounding the block to encode.

In H.264, motion vectors are encoded with respect to a median predictor computed from the motion vectors situated in a causal neighbourhood of the block to encode, for example from the blocks situated above and to the left of the block to encode. Only the difference, also called residual motion vector, between the median predictor and the current block motion vector is encoded.

The encoding using residual motion vectors saves some bitrate, but necessitates that the decoder performs the same computation of the motion vector predictor in order to decode the value of the motion vector of a block to decode.

Recently, further improvements have been proposed, such as using a plurality of possible motion vector predictors. This method, called motion vector competition, consists in determining between several motion vector predictors or candidates which motion vector predictor minimizes the encoding cost, typically a rate-distortion cost, of the residual motion information. The residual motion information comprises the residual motion vector, i.e. the difference between the actual motion vector of the block to encode and the selected motion vector predictor, and an item of information indicating the selected motion vector predictor, such as for example an encoded value of the index of the selected motion vector predictor.

In the High Efficiency Video Coding (HEVC) currently in the course of standardization, it has been proposed to use a plurality of motion vector predictors as schematically illustrated in figure 1: 3 so-called spatial motion vector predictors V1, V2 and V3 taken from blocks situated in the neighbourhood of the block to encode, a median motion vector predictor computed based on the components of the three spatial motion vector predictors V1, V2 and V3 and a temporal motion vector predictor V0 which is the motion vector of the co-located block in a previous image of the sequence (e. g. block of image N-i located at the same spatial position as block Being coded' of image N). Currently in HEVC the 3 spatial motion vector predictors are taken from the block situated to the left of the block to encode (V3), the block situated above (V2) and from one of the blocks situated at the respective corners of the block to encode, according to a predetermined rule of availability. This motion vector predictor selection scheme is called Advanced Motion Vector Prediction (AMVP). In the example of figure 1, the vector V1 of the block situated above left is selected. Note that a scaling of the motion vectors could be necessary in function of the difference of POC (Picture Order Count: image number) between the image containing the motion vector and the reference image pointed by the motion vector.

The difference of POC represents a distance between the image containing the motion vector and the image pointed by the vector. For instance if the difference of POC for Vi, V2 and VO is 1 and the difference of POC for V3 is 2, then the coordinates of V3 have to be scaled and divided by 2.

Finally, a set of 5 motion vector predictor candidates mixing spatial predictors and temporal predictors is obtained. In order to reduce the overhead of signaling the motion vector predictor in the bitstream, the set of motion vector predictors is reduced by eliminating the duplicated motion vectors, i.e. the motion vectors which have the same value. For example, in the illustration of figure 1, V1 and V2 are equal, and V0 and V3 are also equal, so only two of them should be kept as motion vector prediction candidates, for example V0 and V1. In this case, only one bit is necessary to indicate the index of the motion vector predictor to the decoder.

A further reduction of the set of motion vector predictors, based on the values of the predictors, is possible. Once the best motion vector predictor is selected and the motion vector residual is computed, it is possible to further eliminate from the prediction set the candidates which would have not been selected, knowing the motion vector residual and the cost optimization criterion of the encoder. A sufficient reduction of the set of predictors leads to a gain in the signaling overhead, since the indication of the selected motion vector predictor can be encoded using fewer bits. At the limit, the set of candidates can be reduced to 1, for example if all motion vector predictors are equal, and therefore it is not necessary to insert any information relative to the selected motion vector predictor in the bitstream.

To summarize, the encoding of motion vectors by difference with a motion vector predictor, along with the reduction of the number of motion vector predictor candidates leads to a compression gain.

However, as explained above, for a given block to encode, the reduction of the number of motion vector predictor candidates is based on the values taken by the motion vector predictors of the set, in particular the values of the motion vectors of the neighbouring blocks and of the motion vector of the co-located block. Also, the decoder needs to be able to apply the same analysis of the set of possible motion vector predictors as the encoder, in order to deduce the amount of bits used for indicating the selected motion vector predictor and to be able to decode the index of the motion vector predictor and finally to decode the motion vector using the motion vector residual received.

Referring to the example of figure 1, the set of motion vector predictors of the block being coded' is reduced by the encoder to V0 and V1, so the index is encoded on one single bit. If the block Co-located' of image N-i is lost during transmission, the decoder cannot obtain the value of V0, and therefore cannot find out that V0 and V3 are equal. Therefore, the decoder cannot find how many bits were used for encoding the index of the motion vector predictor for the block Being coded', and consequently the decoder cannot correctly parse the data for the slice because it cannot find where the index encoding stops and the encoding of video data starts.

Therefore, the fact that the number of bits used for signaling the motion vector predictors depends of the values taken by the motion vector predictors makes the method very vulnerable to transmission errors, when the bitstream is transmitted to a decoder on a lossy communication network. Indeed, the method requires the knowledge of the values of the motion vector predictors to parse the bitstream correctly at the decoder. In case of packet losses, when some motion vector residual values are lost, it is impossible for the decoder to determine how many bits were used to encode index representing the motion vector predictor, and so it is impossible to parse the bitstream correctly. Such an error may propagate causing the decoder's de-synchronization until a following synchronization image, encoded without prediction, is received by the decoder.

It would be desirable to at least be able to parse an encoded bitstream at a decoder even in case of packet losses, so that some re-synchronization or error concealment can be subsequently applied.

It was proposed, in the document JCTVC-C166r1, TEll: Study on motion vector coding (experiment 3.3a and 3.3c)' by K. Sato, published at the 3rd meeting of the Joint Collaborative Team on Video Coding (JTC-VC) of Guangzhou, 7-15 of October 2010, to use only the spatial motion vector predictors coming from the same slice in the predictor set. This solution solves the problem of parsing at the decoder in case of slice losses. However, the coding efficiency is significantly decreased, since the temporal motion vector predictor is no longer used. Therefore, this solution is not satisfactory in terms of compression performance.

SUMMARY OF THE INVENTION

It is desirable to address one or more of the prior art drawbacks. It is also desirable to provide a method allowing correct parsing at the decoder even in the case of a bitstream corrupted by transmission losses while keeping good compression efficiency. It is also desirable to generate a set of motion vector predictors that is useful without incurring a significant processing penalty or overhead, whether or not parsing problems are eliminated.

According to a first aspect of the present invention there is provided a method of encoding a sequence of digital images into a bitstream, at least one portion of an image being encoded by motion compensation with respect to a reference image portion, wherein, for at least one image portion to encode, the method comprises: obtaining a first set of motion information predictors usable for said image portion to encode; and employing the first set of motion information predictors and one or more predetermined motion information predictors to obtain a second set of motion information predictors having a target number of motion information predictors or fewer.

Such a method enables further motion vector predictors to be added to the second set without incurring any significant overhead.

Preferably, the first set of motion information predictors are or include motion information predictors each associated with an image portion having a predetermined spatial and/or temporal relationship with the image portion being encoded.

For example, the first set of motion information predictors may be the motion information predictors used in Advanced Motion Vector Prediction.

Preferably, the one or more predetermined motion information predictors are obtained from a fixed set of predetermined motion information predictors. Such a fixed set can be obtained in advance, for example offline, and stored for use when carrying out the encoding, thereby saving processing time.

Preferably, the number of predetermined motion information predictors in said fixed set is greater than or equal to said target number.

Preferably, the predetermined motion information predictors in said fixed set are all different.

These measures can ensure that it will always be possible to generate a second set of motion information predictors of size equal to the target number.

The fixed set of motion information predictors may be the same for at least two different image portions to be encoded. In contrast, the first set of motion information predictors may differ from one said image portion to the next. This enables a good combination of motion vector predictors to be obtained, at a reasonable processing cost. The first set is specific to the block to be encoded, whereas the fixed set can be common to two or more blocks, or even to a whole image or sequence of images.

The predetermined motion information predictors may be added one by one, or in subsets, e.g. of two or more, to form a combined set with the first set. However, in one embodiment, the entire fixed set of motion information predictors is added to said first set to form a combined set of motion information predictors.

If the number of motion information predictors in the combined set is greater than target number, removing one or more motion information predictors of said combined set to obtain said second set having said target number of motion information predictors.

Preferably, the removing is carried out in dependence upon priorities of the motion information predictors. This can enable the most useful motion information predictors to be obtained in the second set.

For example, the second set may be obtained by selecting, from said combined set, Nmax motion information predictors having higher priority and discarding one or more remaining motion information predictors having lower priority.

In one embodiment, the motion information predictors in said combined set are ordered, and the second set is obtained by selecting from said combined set the first Nmax motion information predictors in the order thereof, where Nmax is said target number. This can enable seclection even when priorities are not formally assigned. For example, in the case of the fixed set motion information predictors, the order can simply be the order of generation of the motion information predictors.

Of course, preferably this order of generation should still reflect the expected importance of the motion information predictors concerned.

Generally, the first-set motion information predictors should be before the fixed-set motion information predictors in any such order.

When the first set of motion information predictors are the motion information predictors used in Advanced Motion Vector Prediction (AMVP), those motion vector predictors are preferably ordered in conformity with AMVP. This avoids any further overhead associated with assigning priorities.

The fixed set may be the same for any sequence of digital images. Alternatively, the fixed set may be the same for a set of consecutive image portions of digital images of said sequence of digital images.

According to a second aspect of the present invention there is provided a method of decoding a bitstream comprising an encoded sequence of digital images, at least one portion of an image being encoded by motion compensation with respect to a reference image portion, wherein, for at least one image portion to decode, the method comprises: obtaining a first set of motion information predictors usable for said image portion to decode; and employing the first set of motion information predictors and one or more predetermined motion information predictors to obtain a second set of motion information predictors having a target number of motion information predictors or fewer.

Such a decoding method enables further motion vector predictors to be added to the second set without incurring any significant overhead.

The optional features of the decoding method are the same as those of the encoding method, and corresponding advantages and effects are obtained as a result of the features concerned. They are not repeated here.

According to a third aspect of the present invention there is provided apparatus for encoding a sequence of digital images into a bitstream, at least one portion of an image being encoded by motion compensation with respect to a reference image portion, the apparatus comprising: means for obtaining a first set of motion information predictors usable for an image portion to encode; and means for employing the first set of motion information predictors and one or more predetermined motion information predictors to obtain a second set of motion information predictors having a target number of motion information predictors or fewer.

According to a fourth aspect of the present invention there is provided apparatus for decoding a bitstream comprising an encoded sequence of digital images, at least one portion of an image being encoded by motion compensation with respect to a reference image portion, the apparatus comprising: means for obtaining a first set of motion information predictors usable for an image portion to decode; and means for employing the first set of motion information predictors and one or more predetermined motion information predictors to obtain a second set of motion information predictors having a target number of motion information predictors or fewer.

The present invention can also be implemented in software.

According to a fifth aspect of the present invention there is provided a program which, when executed by a computer or processor, causes the computer or processor to carry out the method embodying the aforesaid first or second aspect of the present invention.

Other aspects of the present invention are presented below.

One further aspect of the invention relates to method of encoding a sequence of digital images into a bitstream, at least one portion of an image being encoded by motion compensation with respect to a reference image portion. The method comprises, for at least one image portion to encode, the steps of: -obtaining a target number of motion information predictors to be used for said image portion to encode, and -generating a set of motion information predictors consisting of said target number of motion information predictors, each motion information predictor of the generated set being different from any other motion information predictor of the generated set.

Advantageously, the method of this further aspect of the invention allows the systematic determination of a target number of motion information predictors to be used for encoding motion information, such as a motion vector, associated with an image portion, and the compression is advantageously improved by generating a set of motion information predictors which are all different from one another.

The potential overhead of using a fixed target number of motion information predictors is compensated by the variety of predictors selected which helps improving the compression rate. The target number of different motion information predictors is determined and fixed independently of the actual values of the items of motion information, such as motion vectors, selected as motion information predictors for the current image portion to encode.

An embodiment of the present invention is effective when the number of motion information predictors that is initially generated is a priori unknown, for example as when AMVP is used. For example, if reduction of an initial set is carried out, and the number of initial predictors removed by the reduction process is a priori unknown, an embodiment of the present invention can be used to ensure that the final set of motion information predictors consists of the target number of motion information predictors.

According to an embodiment, the encoding method further comprises the steps of: -selecting a motion information predictor for said image portion to encode from said generated set of motion information predictors, and -encoding an item of information representative of said selected motion information predictor based upon said target number obtained.

Advantageously, a motion information predictor can be selected for a current block to encode and the selected motion vector predictor can be encoded depending on the number of motion information predictors determined. The number of motion information predictors can be systematically retrieved by the decoder, so that the encoded bitstream can be systematically parsed at a decoder even in case of losses.

According to an embodiment, the item of information representative of said selected motion vector predictor is an index of the selected motion vector predictor in the generated set of motion information predictors, and the index is encoded on a number of bits dependent upon said target number obtained.

According to an embodiment, in the step of obtaining a target number of motion information predictors to be used for said image portion to encode, said target number is set equal to a predetermined value for any image portion to encode of the sequence of digital images.

The advantage of this embodiment is that the target number of motion information predictors can be easily obtained, without any supplementary computation or signaling overhead, at both the encoder or the decoder.

According to another embodiment, in the step of obtaining a target number of motion information predictors to be used for said image portion to encode, said target number is determined, for a given image portion to encode, depending upon an encoding information of said given image portion to encode.

Advantageously, such an encoding information can be an encoding parameter, such as for example, if the images are divided into variable size macroblocks for processing, the size of the macroblock to which the image portion to encode belongs. Such an encoding information may also be for example an encoding mode associated with the image portion to encode.

According to yet another aspect, the invention relates to a device for encoding a sequence of digital images into a bitstream, at least one portion of an image being encoded by motion compensation with respect to a reference image portion. The device comprises, for at least one image portion to encode: -means for obtaining a target number of motion information predictors to be used for said image portion to encode, and -means for generating a set of motion information predictors consisting of said target number of motion information predictors, each motion information predictor of the generated set being different from any other motion information predictor of the generated set.

According to yet another aspect, the invention also relates to a computer program product that can be loaded into a programmable apparatus, comprising sequences of instructions for implementing a method for encoding a sequence of digital images as briefly described above, when the program is loaded into and executed by the programmable apparatus. Such a computer program may be transitory or non transitory. In an implementation, the computer program can be stored on a non-transitory computer-readable carrier medium.

The particular characteristics and advantages of the device for encoding a sequence of digital images, of the storage means and of the computer program product being similar to those of the digital video signal encoding method, they are not repeated here.

According to yet another aspect, the invention also relates to a method for decoding a bitstream comprising an encoded sequence of digital images, at least one portion of an image being encoded by motion compensation with respect to a reference image. For at least one said image portion to decode, the method comprises the steps of: -obtaining a target number of motion information predictors to be used for said image portion to decode, and -generating a set of motion information predictors consisting of said target number of motion information predictors, each motion information predictor of the generated set being different from any other motion information predictor of the generated set.

The method for decoding a bitstream has the advantage of allowing determining a target number of motion information predictors and using such a number of different motion information predictors. The target number of motion information predictors can be systematically retrieved, and consequently the bitstream can be parsed systematically, even in case of transmission errors. A further advantage is that in all cases, the parsing of the bitstream is simple, and in particular simpler than with prior art methods which adaptively reduce the number of motion information predictors instead of using a predetermined target number that can be obtained by the decoder.

According to an embodiment, the method further comprises a step of decoding an item of information representative of a selected motion information predictor for said image portion to decode based upon said target number obtained.

Advantageously, in case an encoding dependent on the number of motion information predictors has been applied at the encoder, the item of information representative of the selected motion information predictor for said image portion to decode can be systematically decoded, even in case of transmission errors.

According to yet another aspect, the invention also relates to a device for decoding a bitstream comprising an encoded sequence of digital images, at least one portion of an image being encoded by motion compensation with respect to a reference image portion. The device comprises, for at least one said image portion to decode: -means for obtaining a target number of motion information predictors to be used for said image portion to decode, and -means for generating a set of motion information predictors consisting of said target number of motion information predictors, each motion information predictor of the generated set being different from any other motion information predictor of the generated set.

According to yet another aspect, the invention also relates to an information storage means that can be read by a computer or a microprocessor, this storage means being removable, and storing instructions of a computer program for the implementation of the method for decoding a bitstream as briefly described above.

According to yet another aspect, the invention also relates to a computer program product that can be loaded into a programmable apparatus, comprising sequences of instructions for implementing a method for decoding a bitstream as briefly described above, when the program is loaded into and executed by the programmable apparatus.

Such a computer program may be transitory or non transitory. In an implementation, the computer program can be stored on a non-transitory computer-readable carrier medium.

The particular characteristics and advantages of the device for decoding a bitstream, of the storage means and of the computer program product being similar to those of the decoding method, they are not repeated here.

According to another aspect of the present invention there is provided a method of encoding a sequence of digital images into a bitstream, at least one portion of an image being encoded by motion compensation with respect to a reference image portion, the method comprising generating a set of motion information predictors having controlled diversity and selecting a motion information predictor for said image portion to encode from said generated set of motion information predictors.

Controlled diversity means that the motion information predictors of the set are different from one another but one or more of them are statistically likely to be close to the actual motion information so that a residual (difference between the actual motion information and the predictor concerned) is small and hence efficiently compressible.

The method may comprise: generating first motion information predictors; identifying one or more first motion information predictors as seed predictors; generating one or more second motion information predictors based on the seed predictor(s); and forming said set of motion information predictors from the first and/or second motion information predictors.

In this case, the first motion information predictors can be predictors expected statistically to give good results in terms of compression efficiency. The second motion information predictors, being based on the first predictors, can then be used to explore in an organized or systematic way other predictors in the predictor space neighbouring the first predictors. Such predictors may also be expected to give good results and the more the predictors that are examined the greater the chance of finding a good match to the actual motion information is.

In one embodiment, at least one first motion information predictors is identified as such a seed predictor based on an importance of the first motion information predictor concerned.

The importance may be dependent on a number of times the first motion information predictor concerned appears among the first motion information predictors. The greater the number of times the more important the predictor is considered to be and the more likely it is to be used in the set. As well as looking for identical predictors (duplicates) it can also be effective to look for close matches, too.

Alternatively the importance may be dependent on a measure of how representative the first motion information predictor concerned is of the first motion information predictors as a whole. For example, if the first motion information predictors are averaged, the difference or distance between the average predictor and a given first motion information predictor is a measure of how representative the given predictor is of the first motion information predictors as a whole.

One way of controlhrig the diversity is to generate at east one said second motion information predictor by adding or subtracting an offset from one of said seed predictors. The offset may be fixed. It could also be a pseudo-random value as long as the same seed value is available to the decoder as to the encoder. If the seed predictors are vectors, it is also possible to control the diversity by adding to the seed predictor another vector, e.g. of fixed magnitude and predetermined direction relative to the direction of the seed predictor.

A plurality of said second motion information predictors may be generated based on the same said seed predictor. If the motion information predictors are vectors each having X and Y components, the plurality of second motion information predictors can be obtained by adding and/or subtracting offsets to/from one or both said components of the same said seed predictor. For example, the same offset can be added to and subtracted from the same seed predictor. If the seed predictor is a vector having X and Y components, there are a number of permutations of adding/subtracting offsets to/from one or both of the X and Y components of the same seed predictor. This is an efficient way of generating controlled diversity without a large processing burden.

Another way of controlling the diversity is to generate a plurality of second motion information predictors by forming average of different pairs (or other combinations) of first motion information predictors. For example, if the first motion information predictors are Vi, V2 and V3, three second motion information predictors could be formed from the averages of Vi & V2, V2 & V3 and V3 & Vi. It would also be possible to form different weighted combinations of the same first motion information predictors as different second motion information predictors.

Another way of controlling the diversity is to generate a plurality of second motion information predictors by taking into account the position of the predictor removed from the set of predictors by the reduction process. For example, if a spatial predictor is removed from the set because it is equal to another predictor, the value of the motion vector of the co-located block, the block situated at the same spatial position in the previous frame, can be taken into account. If this motion vector exists and if the value of this motion vector is different from all other predictors in the set, the value of this temporal predictor is added in the predictors set in order to replace the spatial predictor at the same position in the current frame. If the value of this motion vector of the co-located block is the same as the value of any other predictor in the set, the motion vector of the neighboring blocks of the co-located block could be tested. If any motion vector value is different, other alternative way of controlling the diversity could be used.

In a second example, if a temporal predictor is removed from the set because it is equal to another predictor, the value of the motion vector of the co-located block in the current frame can be taken into account. If this vector exists and if the value of this motion vector is different from all other predictors in the set, the value of this spatial predictor is added in the predictors set in order to replace the temporal predictor at the same position in the previous reference frame. If the value of this spatial motion vector is the same as the value of another predictor in the set, the motion vector of the neighboring blocks of this co-located block in the current frame could be tested. If any motion vector value is different, other alternative way of controlling the diversity could be used.

The first motion information predictors may be or include motion information predictors each associated with an image portion having a predetermined spatial and/or temporal relationship with the image portion being encoded. For example the motion information predictors used in AMVP may be first motion information predictors. These are a good source of seed predictors.

In another embodiment, the generation of a set of non-duplicate predictors may consist in pre-determining a fixed set of non-duplicate predictors and replacing eventual duplicate predictor values from the first set of predictors by predictors from this pre-determined fixed set. In order to insure obtaining a final set of non duplicate predictors with a size equal to the target number, the size of the fixed pre-determined set is fixed at least to the target number. The values of these fixed predictors could be determined off line or adaptively for a set of image portions. In that last case the pre-determined fixed set will be the same for all image portions of that set. This set of non duplicate predictors could be systematically added at the end of the first list of predictors.

This pre-determined set of predictors can be generated by adding an offset of a fixed motion vector predictor value as described in the previous paragraph. This fixed value can be the zero motion vector (0,0) or a global motion vector value determined for the current image.

The advantage of this pre-determined non-duplicate predictor is the complexity. The pre-determined set of non-duplicate predictors is computed only once per frame or per sequence or it is fixed. So, no minor additional operations are needed to generate a non-duplicate predictors set according to a target number.

In another embodiment the method comprises: generating first motion information predictors; checking the differences between the generated first motion information predictors; and excluding from said set of motion information predictors one or more first motion information predictors based on the differences.

By taking into account the differences between the first motion information predictors it is possible to control the diversity of the motion information predictors of the set. It is not necessary in this case to identify seed predictors among the first motion information predictors and generate second motion information predictors based on the seed predictors. This can be effective, for example, if a sufficiently high number of first motion information predictors are initially available.

For example, a first motion information predictor having the smallest difference from another first motion information predictor can be removed, as a way of controlling the diversity. The process can be repeated again, as necessary, to successively remove the less diverse predictors.

It is important to note that in this aspect of the invention a number of motion information predictors in said set can be variable.

However, as in the other basic aspect of the present invention the number of motion information predictors in said set can be predetermined at least for a given image portion to encode or even for all image portions (a target number). This makes it possible not only to achieve controlled diversity among the predictors of the set but also to solve the parsing problem noted in the introduction.

This aspect of the invention also provides a corresponding decoding method and corresponding encoding and decoding devices, as well as programs which cause the encoding and decoding.

According to yet another aspect, the invention relates to a method of encoding a sequence of digital images into a bitstream, at least one portion of an image being encoded by motion compensation with respect to a reference image portion. The method comprises, for at least one image portion to encode, the steps of: -obtaining a first set of motion vector predictors to be used for said image portion to encode, and -generating a second set of motion vector predictors from said first set of motion vector predictors, each motion vector predictor of the generated set being different from any other motion vector predictor of the generated second set of motion vector predictors, wherein at least one motion vector predictor of the second set is computed from a selected motion vector predictor of the first set.

Advantageously, the second set of motion vector predictors generated is used for encoding the motion vector associated with the portion of image to encode. The second set of motion vector predictors comprises a variety of different motion vector predictors, which are generated (and possibly selected) so as to enhance the compression efficiency.

According to an embodiment, a motion vector predictor of the first set is selected in a selecting step according to an importance value.

According to an embodiment, the encoding method comprises a step of computing an importance value associated with each motion vector predictor of the first set.

According to an embodiment, a motion vector predictor of the first set is selected in a selecting step according to a distance among the motion vector predictors of the first set.

The various embodiments for selecting a motion vector predictor to generate further additional or virtual motion vector predictors allow applying a controlled diversity selection, which has the advantage of improving the compression efficiency. Indeed, the use of motion vector predictors computed from important motion vector predictors of the initial set allows to more accurately represent the motion vector of the current image portion to encode. Again, it is not essential to have a fixed or target number of predictors in the final set.

According to an embodiment, the motion vector predictors of the first set of motion vector predictors are motion vectors associated with image portions to encode of the image being encoded and/or of a reference image. The first set may be made up of, or include, the predictors used in AMVP.

BRIEF DESCRIPTION OF THE DRAWINGS

Other features and advantages will appear in the following description, which is given solely by way of non-limiting example and made with reference to the accompanying drawings, in which: -Figure 1, already described, illustrates schematically a set of motion vector predictors used in a motion vector prediction scheme; -Figure 2 is a diagram of a processing device adapted to implement an embodiment of the present invention; -Figure 3 is a block diagram of an encoder according to an embodiment of the invention; -Figure 4 illustrates a block diagram of a decoder according to an embodiment of the invention; -Figure 5 details the determination of a set of motion vector predictors according to a first embodiment; -Figure 6 details the determination of a set of motion vector predictors according to a second embodiment; -Figure 7 illustrates schematically a second set of motion vector predictors, -Figure 8 illustrates schematically motion vectors in a coordinates system.

-Figure 9 details the determination of a set of motion vector predictors according to a third embodiment; -Figure 10 details the determination of a set of motion vector predictors according to a fourth embodiment;

DETAILED DESCRIPTION OF THE EMBODIMENTS

Figure 2 illustrates a diagram of a processing device 1000 adapted to implement one embodiment of the present invention. The apparatus 1000 is for example a micro-computer, a workstation or a light portable device.

The apparatus 1000 comprises a communication bus 1113 to which there are preferably connected: -a central processing unit 1111, such as a microprocessor, denoted CPU; -a read only memory 1107 able to contain computer programs for implementing the invention, denoted ROM; -a random access memory 1112, denoted RAM, able to contain the executable code of the method of the invention as well as the registers adapted to record variables and parameters necessary for implementing the method of encoding a sequence of digital images and/or the method of decoding a bitstream; and -a communication interface 1102 connected to a communication network 1103 over which digital data to be processed are transmitted.

Optionally, the apparatus 1000 may also have the following components: -a data storage means 1104 such as a hard disk, able to contain the programs implementing the invention and data used or produced during the implementation of the invention; -a disk drive 1105 for a disk 1106, the disk drive being adapted to read data from the disk 1106 or to write data onto said disk; -a screen 1109 for displaying data and/or serving as a graphical interface with the user, by means of a keyboard 1110 or any other pointing means.

The apparatus 1000 can be connected to various peripherals, such as for example a digital camera 1100 or a microphone 1108, each being connected to an input/output card (not shown) so as to supply multimedia data to the apparatus 1000.

The communication bus affords communication and interoperability between the various elements included in the apparatus 1000 or connected to it. The representation of the bus is not limiting and in particular the central processing unit is able to communicate instructions to any element of the apparatus 1000 directly or by means of another element of the apparatus 1000.

The disk 1106 can be replaced by any information medium such as for example a compact disk (CD-ROM), rewritable or not, a ZIP disk or a memory card and, in general terms, by an information storage means that can be read by a microcomputer or by a microprocessor, integrated or not into the apparatus, possibly removable and adapted to store one or more programs whose execution enables the method of encoding a sequence of digital images and/or the method of decoding a bitstream according to the invention to be implemented.

The executable code may be stored either in read only memory 1107, on the hard disk 1104 or on a removable digital medium such as for example a disk 1106 as described previously. According to a variant, the executable code of the programs can be received by means of the communication network 1103, via the interface 1102, in order to be stored in one of the storage means of the apparatus 1000 before being executed, such as the hard disk 1104.

The central processing unit 1111 is adapted to control and direct the execution of the instructions or portions of software code of the program or programs according to the invention, instructions that are stored in one of the aforementioned storage means. On powering up, the program or programs that are stored in a non-volatile memory, for example on the hard disk 1104 or in the read only memory 1107, are transferred into the random access memory 1112, which then contains the executable code of the program or programs, as well as registers for storing the variables and parameters necessary for implementing the invention.

In this embodiment, the apparatus is a programmable apparatus which uses software to implement the invention. However, alternatively, the present invention may be implemented in hardware (for example, in the form of an Application Specific Integrated Circuit or ASIC).

Figure 3 illustrates a block diagram of an encoder according to an embodiment of the invention. The encoder is represented by connected modules, each module being adapted to implement, for example in the form of programming instructions to be executed by the CPU 1111 of device 1000, a corresponding step of a method implementing an embodiment of the invention.

An original sequence of digital images i0 to i 301 is received as an input by the encoder 30. Each digital image is represented by a set of samples, known as pixels.

A bitstream 310 is output by the encoder 30.

The bitstream 310 comprises a plurality of encoding units or slices, each slice comprising a slice header for encoding values of encoding parameters used to encode the slice and a slice body, comprising encoded video data.

The input digital images are divided into blocks (302), which blocks are image portions and may be of variable sizes (e.g. 4x4, 8x8, 16x16, 32x32). A coding mode is selected for each input block. There are two families of coding modes, spatial prediction coding or Intra coding, and temporal prediction coding or Inter coding. The possible coding modes are tested.

Module 303 implements Intra prediction, in which the given block to encode is predicted by a predictor computed from pixels of the neighbourhood of said block to encode. An indication of the Intra predictor selected and the difference between the given block and its predictor is encoded if the Intra coding is selected.

Temporal prediction is implemented by modules 304 and 305. Firstly a reference image among a set of reference images 316 is selected, and a portion of the reference image, also called reference area, which is the closest area to the given block to encode, is selected by the motion estimation module 304. The difference between the selected reference area and the given block, also called a residual block, is computed by the motion compensation module 305. The selected reference area is indicated by a motion vector.

An information relative to the motion vector and the residual block is encoded if the Inter prediction is selected. To further reduce the bitrate, the motion vector is encoded by difference with respect to a motion vector predictor. A set of motion vector predictors, also called motion information predictors, is obtained from the motion vectors field 318 by a motion vector prediction and coding module 317.

Advantageously, the set of motion vector predictors used to select a best motion vector predictor to encode a current motion vector is generated as explained in more detail hereafter with respect to figures 5 and 6. For a given current block to encode, a predetermined number Nmax of motion vector predictors is set, and consequently the index of the selected motion vector predictor, which is an item of information representative of the selected motion vector predictor, can be encoded using a predetermined number of bits. This predetermined number of bits can be also retrieved by the decoder even in case of losses, therefore it is ensured that the decoder will be able to parse the bitstream even in case of errors or losses. The Nmax motion vector predictors are selected according to various embodiments to be all different from one another so as to enhance the compression efficiency.

The selection of the predetermined number Nmax of motion vector predictors and of the corresponding number of bits to encode the index of the motion vector predictor can be applied either for the entire sequence, or for a group of images of the sequence, or at the block level depending on an encoding parameters such as the block size or the encoding mode. For example, a first predetermined number of motion vector predictors Nmaxi can be used for the blocks encoded using Inter prediction for which a residual block is encoded, and a second predetermined number motion vector predictors Nmax2 can be used for the blocks encoded using the SKIP mode, for which only a motion vector is encoded, but no residual block. The respective numbers of motion vector predictors Nmaxi and Nmax2 can be for example signaled in the bitstream by inserting them in a header, such as the slice header, or in any appropriate metadata field.

The encoder 30 further comprises a module of selection of the coding mode 306, which uses an encoding cost criterion, such as a rate-distortion criterion, to determine which is the best mode among the spatial prediction mode and the temporal prediction mode. A transform 307 is applied to the residual block, the transformed data obtained is then quantized by module 308 and entropy encoded by module 309.

Finally, the encoded residual block of the current block to encode is inserted in the bitstream 310, along with the information relative to the predictor used. For the blocks encoded in SKIP' mode, only a reference to the predictor is encoded in the bitstream, without any residual block.

The encoder 30 further performs the decoding of the encoded image in order to produce a reference image for the motion estimation of the subsequent images. The module 311 performs inverse quantization of the quantized data, followed by an inverse transform 312. The reverse motion prediction module 313 uses the prediction information to determine which predictor to use for a given block and the reverse motion compensation module 314 actually adds the residual obtained by module 312 to the reference area obtained from the set of reference images 316. Optionally, a deblocking filter 315 is applied to remove the blocking effects and enhance the visual quality of the decoded image. The same deblocking filter is applied at the decoder, so that, if there is no transmission loss, the encoder and the decoder apply the same processing.

Figure 4 illustrates a block diagram of a decoder according to an embodiment of the invention. The decoder is represented by connected modules, each module being adapted to implement, for example in the form of programming instructions to be executed by the CPU 1111 of device 1000, a corresponding step of a method implementing an embodiment of the invention.

The decoder 40 receives a bitstream 401 comprising encoding units, each one being composed of a header containing information on encoding parameters and a body containing the encoded video data. As explained with respect to figure 3, the encoded video data is entropy encoded, and the motion vector predictors' indexes are encoded, for a given block, on a predetermined number of bits. The received encoded video data is entropy decoded (402), dequantized (403) and then a reverse transform (404) is applied.

In particular, when the received encoded video data corresponds to a residual block of a current block to decode, the decoder also decodes motion prediction information from the bitstream, so as to find the reference area used by the encoder.

The module 410 applies the motion vector decoding for each current block encoded by motion prediction, comprising determining the number Nmax of motion vector predictors used and retrieving the motion vector predictor index encoded on a number of bits dependent on Nmax.

Similarly to module 317 of figure 3, motion vector decoding module 410 generates a set of Nmax motion vector predictors. The embodiments explained hereafter with respect to figures 5 and 6 apply similarly. If the bitstream is received without losses, the decoder generates exactly the same set of motion vector predictors as the encoder. In case of losses, it may not be possible to generate the set of motion vector predictors and therefore to correctly decode the motion vector associated with the current block. However, the parsing of the bitstream is always possible, even in case of losses, since the number of bits used to encode the index of the motion vectors predictor can be systematically retrieved by the decoder.

Once the index of the motion vector predictor for the current block has been obtained, if no losses have occurred, the actual value of the motion vector associated with the current block can be decoded and used to apply reverse motion compensation (406). The reference area indicated by the decoded motion vector is extracted from a reference image (408) to finally apply the reverse motion compensation 406.

In case an Intra prediction has been applied, an inverse Intra prediction is applied by module 405.

Finally, a decoded block is obtained. A deblocking filter 407 is applied, similarly to the deblocking filter 315 applied at the encoder. A decoded video signal 409 is finally provided by the decoder 40.

Figure 5 details the generation of the set of motion vector predictors or motion vector candidates in a first embodiment of the present invention. All the steps of the algorithm represented in figure 5 can be implemented in software and executed by the central processing unit 1111 of the device 1000.

Figure 5 represents a flowchart applied for a given current block to encode, which has an associated motion vector designating a reference area in a reference image.

An initial set of motion vector predictors Li is firstly obtained in step S500. The set Li is composed of N candidates. In an embodiment, the initial set of motion vector predictors comprises the motion vector candidates selected according to the motion vector prediction scheme AMVP already described with reference to figure 1, for example vectors V0 to V3 of figure 1 and the median vector computed from V1, V2 and V3. Accordingly, N is a maximum of 5.

Preferably, the initial set of motion vector predictors Li contains only motion vectors different from one another. Taking the example of figure 1, only motion vectors V0, V1 and the median vector should be kept in Li, and the number of motion vector predictors is N=3.

In an alternative embodiment, any other scheme for selecting motion vectors already computed and computing other motion vectors from available ones (i.e. average, median etc) to form the initial set of motion vector predictors Li can be applied.

In a yet another alternative embodiment, the initial set of motion vector predictors Li is empty and N=0.

In the following step S502, the target number Nmax of candidate motion vector predictors to use is obtained. Nmax may either be predetermined for the entire sequence of digital images to encode, for example NmaF4, or may be selected according to encoding parameters of the current block to encode or of the encoding unit (for example the slice) to which the current block to encode belongs.

For example, the number Nmax may be dependent on the size of the transform applied to the macroblock to which the current block being processed belongs: e.g. Nmax= p for a transform applied on a block of 21x21.

In an embodiment, the number of bits k to use for encoding the index of the motion vector predictor is directly related to Nmax: kINTsupog2(Nr2x)), where INT(x) is the integer number immediately following the value x. Preferably, Nmax should be a power of 2, Nmax2k, so as to use all the indexes that can be encoded on k bits.

Advantageously, all the bits used for signaling are used so as to be able to designate a variety of motion vector predictors and to improve the compression.

Alternatively, any type of encoding of the indexes representative of the motion vector predictors can be used, after the number of different motion vector predictors Nmax has been determined.

In particular, any type of entropy encoding, such as Huffman encoding or arithmetic encoding can be used. Also, the indexes may be also encoded using a prefix type code, such as a Rice-Golomb or a unary code.

Next, it is tested in step S503 whether the number N of motion vector predictors of set Li is higher than Nmax.

In case of positive answer, test S503 is followed by step S504 of selection of a motion vector predictor candidate from Li, followed by the removal of the selected motion vector predictor candidate from Li in step S506 to form a modified set of motion vector predictors L2.

The selection of step S504 is applied according to a removal criterion, for example a distance criterion. For example, the set Li comprises motion vectors predictors {Vi VN}, wherein each motion vector predictor is a vector represented by its components or coordinates on the X-axis and the Y-axis in a coordinate system, as represented in figure 8.

In an embodiment, step S504 determines the distance between each vector Vk and each other vector V of Li other than Vk itself, by computing the distance as: d(Vk = --where a represents the absolute value of a. Taking the example of figure 8, vector V has the coordinates (3,2), vector V has the coordinates (4,2) and V" has the coordinates (3,3). In this example, d(V,V')=d(V,V")=l, so V and V' are at the same distance from vector V, whereas d(V',V")=2.

Any other type of metric for the calculation of the distances can be alternatively applied.

The minimal distance found d(Vp,Vq) indicates the two closest vectors V, Vq, among the set Li, and therefore one of these two vectors is selected for removal. The selection of one of these two vector can be based on the distance of each one of them to the remaining motion prediction vectors in the modified set Li: the vector between V and Vq which has the smallest distance to another vector of the set Li is selected for removal.

The effect is to ensure that the distance between the remaining vectors in the modified set L2 is maximal, so as to allow the use of motion vector predictors as varied or diverse as possible.

After the removal of the selected vector, the value of N is decreased (S508), and then N is compared to Nmax (S5i0). If the value of N has not reached yet Nmsx (answer no' to test S5i0) steps S504 to S5i0 are repeated. Otherwise, if N has reached Nmax, step S5i0 is followed by step S522 described hereafter.

If the answer to test S503 is no', then test S512 checks whether N is lower than Nmax. In case of negative answer, i.e. if NNmax, test S512 is followed by step S522 described hereafter.

If the answer to test S5i2 is yes', i.e. if N is strictly lower than Nmax, then test S5i2 is followed by step S5i4 of obtaining or generating an additional motion vector predictor candidate. Indeed, starting from the initial set of motion vector predictors Li, it is possible to add other candidates as motion vector predictors, in a predetermined order, to form a modified set of motion vector predictors L2. Taking the example of figure 7, the motion vectors of blocks 710, 720, 730 and 740 can be added as possible motion vector predictors. Further, the 2 predictors among 770, 760, 750 which were not selected in 5500 can be added as possible motion vector predictors.

For each potential motion vector predictor candidate MV, it is checked whether the motion vector predictor MV is different from all the motion vector predictor candidates already stored in the set L2.

If each potential motion vector candidate considered, as for example the motion vectors of blocks 710, 720, 730, 740, 750, 760 and 770 of figure 7, is equal to a motion vector predictor of set L2, new virtual' motion vector predictor candidates are computed in step S514.

Such motion vector predictor candidates are called virtual because they are not motion vectors of other blocks of the current image or of the reference image. The virtual motion vector predictors are computed from existing motion vector predictors, for example by adding offsets. For example, from a motion vector MV of set L2 of coordinates (MVX, MV), it is possible to compute four virtual motion vector predictors by adding/subtracting an offset off to its coordinates: MV'(MV±off, MV±off). Typically, off may be set equal to 1 or 2.

Alternatively, other modifications of the components of the motion vector predictor MV to obtain virtual motion vector predictors can be applied, so as to obtain diverse motion vector predictors starting from the existing motion vector predictors in order to increase the compression efficiency.

For example, the components of the motion vector MV may be modified independently, using respectively two values offx and offy, and either ofix or offy may be set to 0.

In an embodiment, both ofix and offy are proportional to the corresponding component: offx=aMV and offy=bMV with a and b typically smaller than 0.5. If necessary, the modified coordinates MV and MV are rounded to the closest integer value, so as to represent a displacement on the grid of pixels.

In an embodiment, a supplementary motion vector, of predetermined norm, is added to the vector MV, the supplementary vector having the same direction as motion vector MV, as represented in figure 8: supplemental vector 850 is added to vector 820.

In yet another alternative embodiment, a variance of the motion vectors of the set L2 is computed: var = çv1 -MV)2 +(MV -MV)-where MV represents the JtfVeL2 mean value of the MV coordinates of the vectors of L2 and MV represents the mean value of the MV coordinates of the vectors of L2.

Then, the offset off is selected by comparing the calculated value var to a predetermined threshold T. T can be equal to 501L2. If var is lower than T, the value off is small, for example off=1; if var is higher than T, off is set to a larger value, for example ofl=3. Further, in this embodiment also, a differentiated value offx or offy can be computed for each component.

In one embodiment, to generate a virtual motion vector predictor, the virtual motion vector is selected among a pre-determined set of non-redundant predictors. This set can be fixed or changed at sequence level or at image level. It contains Nmax predictor values which are all different. These predictor values can be generated by adding offsets to a fixed value. For example, this fixed value could be the zero motion vector (0,0) or a global motion vector value of the current image.

The advantage of these pre-defined virtual predictors is the low complexity. Indeed, the pre-determined predictors set doesn't need to be computed for each image portion so no operation is needed to generate the predictor compared to adding offsets to the motion vector values of the first set.

One motion vector predictor obtained in step S514 is added to the set of motion vector predictors L2 in step S516, and the number N is increased by 1 (step S518).

Next, it is checked in step S520 is N is equal to Nmax. In case of negative answer, steps S514 to S520 are repeated.

In case of positive answer, the determined target number of motion vector predictor candidates Nmax has been reached, and step S520 is followed, at the encoder, by the step S522 of selection of an optimal motion vector predictor for the current block from set L2. For example, a rate-distortion optimization criterion is applied to select the optimal motion vector predictor MV1 to encode the motion vector of the current block.

At the encoder, the motion residual, i.e. the difference between the motion vector of the current block and the selected motion vector predictor is encoded, as well as an indication of the motion vector predictor selected in step S524. For example, the index i of the selected motion vector predictor MV of L2 is encoded, using klNTsup(log2(Nmsx)) bits.

Alternatively, an entropy encoding of the index i may be applied.

In yet another alternative, the index i can be encoded using a prefix type code, such as the Rice-Golomb code, in which each value i is encoded using i l's followed by a 0'.

The algorithm of figure 5 can also be implemented by a decoder to generate the set of motion vector predictor or motion vector candidates for a given block, without steps S522 and S524.

At the decoder, the index i of the selected motion vector predictor MVi for the given block to decode is obtained from the bitstream, knowing Nmax and therefore the number of bits k on which the index i has been encoded. The steps S500 to S518 are similarly implemented to obtain the set of motion vector predictors L2, so that the index i decoded from the bitstream designates the motion vector predictor actually used by the encoder.

In case of losses during transmission, since the number Nmax can be systematically retrieved by the decoder, the received bitstream can be systematically parsed to extract the index i designating the selected motion vector predictor, even if, depending on the packets lost, the complete set of motion vector predictors L2 may not be obtained at the decoder.

Figure 6 details the generation of the set of motion vector predictors or motion vector candidates in a second embodiment of the present invention. All the steps of the algorithm represented in figure 6 can be implemented in software and executed by the central processing unit 1111 of the device 1000.

Figure 6 represents a flowchart applied for a given current block to encode, which has an associated motion vector designating a reference area in a reference image.

Firstly, similarly to step S502 of figure 5, the target number Nmax of motion vector predictor candidates to use is obtained in step S600.

In an embodiment, Nmax is of the form 2k, so that each index value that can be coded on k bits corresponds to a possible motion vector predictor.

For example, in order to use all motion vector predictors of the AMVP scheme which proposes 5 motion vector predictors, 3 bits are necessary to encode the index of a motion vector predictor. In this case, preferably Nmax=23=8.

An initial set of motion vector predictor candidates Li is obtained in step S602. For example, the initial set of N=5 motion vector predictors of AMVP is selected.

A reduction process is applied on the initial set of motion vector predictors to eliminate duplicates, so as to obtain a reduced set of motion vector predictors containing Ni elements. Preferably, the number of duplicates of each remaining vector after the reduction process is recorded and stored in a memory for a subsequent use in step S6i2 described hereafter.

It is next checked (test S606) whether Ni is higher than or equal to Nmax, the target number of motion vector predictors. It may be pointed out that a positive outcome to this test only occurs if the algorithm starts with a first set of motion vector predictors a greater number of motion vectors than Nmax. In case of positive answer, step S606 is followed by step S630 of selection of the first Nmax motion vector predictor candidates of the set Li to form the set of motion vector predictors L2.

In case of negative answer, i.e. if NI is lower than Nmax, the set of motion vector predictors has to be complemented with additional motion vector predictors.

A second set of motion vector predictor candidates Li' is obtained in step S608.

The second set of motion vector predictors Li' is composed of the remaining motion vector predictors of the first set Li and of additional motion vectors, for example corresponding to the motion vectors of the block 710, 720, 730 and 740 of the reference image as represented on figure 7. Further, the 2 predictors among 770, 760, 750 which were not selected in S600 can be added as possible motion vector predictors. Each motion vector predictor of the set Li' has a corresponding index.

As already mentioned with reference to figure i, a scaling of motion vectors could be necessary to obtain a set of motion vectors with the same scale. However, it may be considered that scaling is a complex process. As a consequence, increasing the number of scaling could increase the complexity of the encoding and decoding processes.

A solution to limit the increase of complexity could be to forbid scaling or limit the number of authorized scalings during the generation of the set of candidates. For instance, in step S608, additional motion vector predictors could be added, if they do not require any scaling or, if a scaling is necessary, an additional motion vector requiring scaling is added only if a budget of authorized scalings is not already used. If this budget is already used, then the additional motion vector predictor is not considered and a new predictor is tested. Another embodiment could be to consider only motion vector predictors with the same POC difference as the current motion vector to predict.

Next, a reduction process is applied to the second set of motion vector predictors in step S61 0 to obtain a reduced second set of motion vector predictors Li" of N2 vectors. The reduction process eliminates the duplicates, so that all motion vector predictors of Li" are different from one another. The number of duplicates of each vector kept ri LI" is recorded and stored in a memory for a subsequent use in step S6i2 described hereafter.

It is then checked in step S628 whether the number of motion vector predictors N2 is higher than or equal to Nmax. In case of positive answer, step S628 is followed by step S630 already described.

In case of negative answer, it is necessary to add more motion vector predictors to the reduced second set of motion vector predictors Ll"to obtain the final set of Nmax motion vector predictors.

Following the test S628, in case of negative answer, an importance value is assigned to each remaining motion vector predictor candidate of the reduced second set of motion vector predictors Li" in step S6i2.

In an alternative embodiment, step S6i2 follows directly test S606, in case of negative answer to test S606.

The importance value is computed in this embodiment as the number of duplicates of a given motion vector predictor, using the number of duplicates of a given motion vector predictor computed and stored during steps S604 and S6i 0. Taking the example of figure 1, two vectors, V0 and V3, are equal, so vector V0 has an importance value equalto2.

In an alternative embodiment, the importance value can be computed as a function of the distance to a representative vector of the set of vectors considered, such as the average value of the vectors of the set or the median of the vectors of the set. Then, the importance may be computed as the inverse of the distance of a given vector of the set Vn to the representative vector: the closer a vector Vn is to the representative vector of the set, the higher the importance of Vn.

Then, the N2 remaining motion vector predictor candidates are ordered in step 8614 according to an order of decreasing importance value. If several motion vector predictors have the same importance value, they can be ordered according to the increasing order of their indexes.

The re-ordered motion vector predictors are re-assigned increasing indexes {V0,Vi VN21}.

In the following step S616, a variable n is initialized to 0 and a variable N is initialized to N2, which is the current number of motion vector predictors in the re-ordered set.

Next, in step 8618 following step 8616, virtual motion vector predictor candidates are added to the re-ordered set. In this embodiment, the virtual motion vector predictors are computed from the remaining motion vector predictors, ordered according to their importance. The motion vector predictor of index n of the re-ordered set, Vn of coordinates (Vn, Vn), is considered. The following list of 8 virtual motion vector predictors, defined by their coordinates, can be computed from Vn, by adding successively i.off and -off to one or the two coordinates of Vn: {(Vn+off, Vn), (Vn-off, Vn), (Vn+off, Vn +off), (Vn-s-off, Vn -off), (Vn-off, Vn i-off), (Vn-off, Vn -off), (Vn, Vn i-off), (Vn, Vn -off) }.

Any alternative computation of the virtual motion vector predictors starting from the motion vector predictor Vn, in particular the alternatives described above with respect to step S51 4 of figure 5, may be used.

This list of virtual motion vector predictors is added to the current set of motion vector predictors.

The duplicates are eliminated in step 8620.

The value N is updated in step 8622 to the remaining number of motion vector predictors after the removal of potential duplicates.

Next, it is checked whether N is higher than or equal to Nmax in step S624. In case of negative answer, step 8624 is followed by step S626 of increasing of the value n by 1, and steps 5618 to S624 are repeated.

In case of positive answer to step S624, sufficient motion vector predictors have been obtained. Step S624 is followed by step 5630 of selection of the first Nmax motion vector candidates to constitute the final set of motion vector predictors L2 of Nmax vectors.

Alternatively, in order to reduce the algorithm complexity, we may limit the number of time we loop in steps S618 to S624 and consequently limit the number of virtual motion vector predictor candidates generated. In that case, if in step 5624 N is above or equal to Nmax or if ii is above or equal to a given maximum number of loops, 5624 is followed by step 5630.

One can note that in this algorithm, the motion vector predictors in the set of N2 motion vector predictors are ordered. If step 5630 follows step S606, then the set follows the order defined by the AMVP algorithm. In an embodiment, AMVF could provide an order depending on statistics of selection of motion vector predictors computed on a set of training sequences.

If step S630 is follows step S628, motion vector predictors selected by AMVP are positioned at the beginning of the set. These first predictors are known to be the most relevant. These predictors are then followed by motion vector predictors extracted from already encoded image portions (and consequently representing the real motion in the video) but which would not have been selected by AMVP.

If step S630 follows step 5624, an order depending of the motion vector importance has been defined in step S614. Additional virtual motion vector predictors are added to the set obtained after steps S6101S6281S612, but the insertion of these virtual motion vector predictors depends on the order obtained during step 5614.

It may be considered as a consequence that step S630 removes the less relevant motion vector predictors.

At the encoder, step S630 is followed by step S632, analogous to step S522 of figure 5, of selection of an optimal motion vector predictor MV1 among the set of motion vector predictors for the current block, according to a predetermined criterion such as a rate-distortion criterion.

Step 8632 is followed by a step 8634 of encoding the motion vector of the current block using the motion vector predictor MVi, similar to step 8524 of figure 5. For example, the index i of the motion vector predictors MVi is encoded using k bits, k being computed from Nmax, k= I NTsup(Iog2(Nmax)).

Alternatively, an entropy encoding of the index i may be applied.

In a further embodiment, one may imagine that the motion vector predictors are stored in memory in a table of size Nmax. created during step S600. Each element of the array contains a motion vector predictor value and a flag indicating whether the value is a duplicate value comparing to a previous element of the table. During step S600, all flags are initialized to "non-duplicate", and all vector predictor values are initialized to the vector (0,0). This array is initialized in step 8602 with the motion vector predictors of the initial set. During steps S606 to S624, each element of this table could be manipulated (associated flag set to "duplicate", insertion of new motion predictor values, etc). The objective is to replace all elements with an associated flag equal to "duplicate" by a new non duplicate element. However, it may occur when the loop S618 to S624 stop in step S624 under the condition "n above or equal to a given maximum number of loops", that the process ends without removing all non duplicates values. In that case, all flags are set to non-duplicate during step S624 (even if some duplicate elements remains) and step 8624 is followed by step 8630.

The algorithm of figure 6 can also be implemented by a decoder to generate the set of motion vector predictors or motion vector candidates for a given block, except steps 8632 and 8634 are omitted on the decoder side.

At the decoder, the index i of the selected motion vector predictor MVi for the given block to decode is obtained from the bitstream, knowing Nmax and therefore the number of bits k on which the index i has been encoded. The steps S600 to 8630 are similarly implemented to obtain the set of motion vector predictors L2, so that the index i decoded from the bitstream designates the motion vector predictor actually used by the encoder.

The embodiments described above are based on block partitions of input images, but more generally, any type of image portions to encode or decode can be considered, in particular rectangular portions or more generally geometrical portions.

Figure 9 details a third embodiment for the generation of the set of motion vector predictors or motion vector candidates according to the invention. This third embodiment is a fast motion vector predictors set generation.

Firstly, similarly to step 8502 of figure 5 or to step 8600 of figure 6, the target number Nmax of motion vector predictor candidates to use is obtained in step 8900.

An initial set of motion vector predictor candidates Li is obtained in step S902 similarly to step 8602 of figure 6.

Then in step 8904, a pre-determined set of non-redundant predictors is added to the initial set of predictors. The pre-determined set of non-redundant predictors can be fixed or changed at sequence level or at image level. It contains Nmax predictor values which are different from one another. These predictor candidate values can be generated by adding offsets to a fixed value. For example, this fixed value could be the zero motion vector (0,0) or a global motion vector value of the current image.

A reduction process is applied in step S906 on the set resulting from the combination of the initial set with the pre-determined set of motion vector predictors to eliminate duplicate motion vectors so as to obtain a reduced set U of motion vector predictors containing Ni elements. In this process, the reduced set of predictor candidates contains necessarily at least Nmax candidates. Note that in a faster implementation, the elimination of duplicates candidates is not performed. In That case, Li is just the combination of the initial set with the pre-determined set of motion vector predictors.

The first Nmax motion vector predictor candidates of the set Li are selected to form the set of motion vector predictors L2 in step 908.

At encoder side, the optimal motion vector is selected in step 5910 and its index i is encoded using k bits, k being computed from Nmax, k=lNTsup(log2(Nmax)) in step 5912. Steps 5910 and S9i2 are similar to respectively steps S632 and S634 of figure 6.

Figure 10 details a fourth embodiment for the generation of the set of motion vector predictors or motion vector candidates according to the invention. This fourth embodiment also involves fast motion vector predictors set generation.

Firstly, similarly to step S502 of figure 5 or to step 5600 of figure 6, the target number Nmax of motion vector predictor candidates to use is obtained in step Si 000.

An initial set of motion vector predictor candidates Li is obtained in step Si 002 similarly to step S602 of figure 6. This set may correspond to the set obtained with the AMVP algorithm. In the set obtained with the AMVP algorithm the motion vector predictors are ordered in accordance with AMVP and the present embodiment advantageously makes use of this ordering to establish a priority of the motion vector predictors, as will be described later.

In this embodiment, it is not necessary to apply a reduction process on the initial set of motion vector predictors to eliminate duplicates. However, optionally, such reduction may be applied in S1004, so as to obtain a reduced set Li of motion vector predictors containing Ni elements (Si004) similarly to step S604 in figure 6.

During step S1005, the motion vector predictors candidates contained in the set Li are ordered in an order reflecting the importance (i.e. priority) of the motion vector predictors. The order could be the order assigned to the motion vector predictors in accordance with the AMVF algorithm, as noted above. Alternatively, the order could be given by a statistical analysis of the most selected motion vector predictors on a set of training sequences. This analysis could allow ordering the motion vector predictors according to their position. For instance a co-located motion vector predictor could be statistically more important than the above motion vector predictor, and the above motion vector predictor could be statistically more important than the predictor on the left.

Step Si005 is followed by step Si006 checking if Ni is above or equal to Nmax. If yes, this step is followed by step Si030 which selects the first Nmax candidates in the ordered set Li. This step S1030 is described in more detail below.

If Ni is below Nmax, step Si006 is followed by step Si008.

During step Si008 a so-called fixed set of motion vector predictors is obtained. This fixed set is made up of predetermined non-redundant predictors. This pre-determined set is similar to the one used in step S904 of figure 9.The fixed set of non-redundant predictors can be fixed at sequence level, at image level or at the image portion level. It contains at least Nmax predictor values which are different from one another. As already described these predictor candidate values can be generated by adding offsets to a fixed motion vector predictor such as the motion vector (0,0).

This step is followed by step SiOi2 where a subset of motion vector predictors of the pre-determined set are added to the ordered set Li. In this embodiment each subset contains more than one motion vector predictor, for example 2 or 3 motion vector predictors. However, it is also possible to add just one motion vector predictor in each iteration of S1012.

The result of Si 012 is a combined set having the first set of motion vector predictors obtained in steps 81002 to 81005 end one or more motion vector predictors from the fixed set, as added, iteratively or once and for all, in step 51012.

Step 81014, like S1004, is optional in the present embodiment. However, if carried out, during step 81014, a reduction process eliminating duplicates candidates is applied on the combined set.

The size N2 of the combined set, after the reduction process if applied, is then compared with Nmax (step 1020). When N2 is greater than or equal to Nm2x processing moves to step 81030 or, optionally, to step 51016.

Note also that instead of applying step S1012 (and, if used, 81014) iteratively, the entire fixed set could be added in one go, in which case 81012 is executed once only and the check in 51020 can be dispensed with.

The motion vectors from the fixed set, added in 81012, may have priorities. For example, they may be generated in order of decreasing priority so that the motion vector predictors of the first subset have higher priorities than the motion vector predictors of the second subset, and so on for each subsequent subset. In other words the order of generation accords with the priority of the motion vector predictors of the fixed set. In that case, it is not necessary to assign any further priority to the motion vector predictors of the fixed set. The motion vector predictors already have priorities as determined in step 81005, for example the priorities in AMVP, and the fixed-set motion vector predictors added in step 81012 will have lower priorities, for example according to the order of generation thereof.

However, if desired, in step Si 016 an importance can be assigned to each candidate of the combined set of N2 candidates. For example, a step similar to step S612 can be carried out, where the priority (importance) is based on the number of duplicates associated to each candidate.

This step is followed by step S1018 where motion vector predictor candidates are ordered in function of their associated importance. Step S1030 is then carried out.

In step S1030, the first Nmax motion vector predictor candidates in the order thereof within the combined set are selected to obtain a second set of motion vector predictors having Nmax motion vector predictors.

If S1030 is reached from S1005, there will only be first-set motion vector predictors to select from. If the number of them is equal to Nmax then no selection is required at all. If the number of them is greater than Nmax then the Nmax higher-priority motion vector predictors are selected and the lower-priority motion vector predictors are discarded.

If, on the other hand, S1030 is reached via S1012, there will be a mixture of motion vector predictors from the first set and motion vector predictors from the fixed set to select from If the number N2 of motion vector predictors in the combined set is equal to Nmax then no selection is required at all. If N2 is greater than Nmax then the Nmax higher-priority motion vector predictors are selected and the lower-priority motion vector predictors are discarded. Unless S1016 and S1018 are applied, the first-set motion vector predictors will have higher priority than the fixed-set motion vector predictors, so all the first-set motion vector predictors will be selected in S1030, the remaining motion vector predictors being taken from the fixed set in the order of generation (priority) thereof.

It should be noted that in this embodiment, the second set can contain duplicates.

This step is followed by step 51032 where an optimal motion vector predictor is selected similarly to step 5632 and by step 51034 where the index of the optimal motion vector predictor candidate is encoded similarly to step S634.

It should also be noted that in step Si 030 fewer than Nmax predictors could be selected, if desired.Other alternative embodiments may be envisaged, such as for example starting with a large set of motion vector predictor candidates, comprising a number of vectors N larger than the target number of motion vector predictors determined Nmax, and applying a clustering type algorithm to reduce the set of vectors. For example, a VoronoI partition can be applied to reduce the set to the Nmax most representative vectors of the set according to a predefined distance between the vectors of the set.

More generally, any modification or improvement of the above-described embodiments, that a person skilled in the art may easily conceive should be considered as falling within the scope of the invention.

Claims

<claim-text>CLAIMS1. A method of encoding a sequence of digital images into a bitstream, at least one portion of an image being encoded by motion compensation with respect to a reference image portion, wherein, for at least one image portion to encode, the method comprises: obtaining a first set of motion information predictors usable for said image portion to encode; and employing the first set of motion information predictors and one or more predetermined motion information predictors to obtain a second set of motion information predictors having a target number of motion information predictors or fewer.</claim-text> <claim-text>2. A method as claimed in claim 1, wherein the first set of motion information predictors are or include motion information predictors each associated with an image portion having a predetermined spatial and/or temporal relationship with the image portion being encoded.</claim-text> <claim-text>3. A method as claimed in claim 1 or 2, wherein the first set of motion information predictors are the motion information predictors used in Advanced Motion Vector Prediction.</claim-text> <claim-text>4. A method as claimed in any preceding claim, comprising obtaining said one or more predetermined motion information predictors from a fixed set of predetermined motion information predictors.</claim-text> <claim-text>5. A method as claimed in claim 4, wherein the number of predetermined motion information predictors in said fixed set is greater than or equal to said target number.</claim-text> <claim-text>6. A method as claimed in claim 4 or 5, wherein the predetermined motion information predictors in said fixed set are all different.</claim-text> <claim-text>7. A method as claimed in claim 4, 5 or 6, wherein the fixed set of motion information predictors is the same for at least two different image portions to be encoded.</claim-text> <claim-text>8. A method as claimed in claim 7, wherein said first set of motion information predictors differs from one said image portion to the next.</claim-text> <claim-text>9. A method as claimed in any one of claims 4 to 8, comprising adding the fixed set of motion information predictors to said first set to form a combined set of motion information predictors.</claim-text> <claim-text>10. A method as claimed in claim 9, further comprising, if the number of motion information predictors in said combined set is greater than target number, removing one or more motion information predictors of said combined set to obtain said second set having said target number of motion information predictors.</claim-text> <claim-text>11. A method as claimed in claim 10, wherein the removing is carried out in dependence upon priorities of the motion information predictors.</claim-text> <claim-text>12. A method as claimed in claim 11, wherein the second set is obtained by selecting, from said combined set, Nrnax motion information predictors having higher priority and discarding one or more remaining motion information predictors having lower priority.</claim-text> <claim-text>13. A method as claimed in claim 10, wherein the motion information predictors in said combined set are ordered, and the second set is obtained by selecting from said combined set the first Nmax motion information predictors in the order thereof, where Nrnax is said target number.</claim-text> <claim-text>14. A method as claimed in claim 13, wherein the first set of motion information predictors are the motion information predictors used in Advanced Motion Vector Prediction (AMVP) and are ordered in conformity with AMVP.</claim-text> <claim-text>15. A method according to claim to any one of claims 4 to 14, wherein the fixed set is the same for any sequence of digital images.</claim-text> <claim-text>16. A method according to claim 15, wherein the fixed set is the same for a set of consecutive image portions of digital images of said sequence of digital images.</claim-text> <claim-text>17. A method of decoding a bitstream comprising an encoded sequence of digital images, at least one portion of an image being encoded by motion compensation with respect to a reference image portion, wherein, for at least one image portion to decode, the method comprises: obtaining a first set of motion information predictors usable for said image portion to decode; and employing the first set of motion information predictors and one or more predetermined motion information predictors to obtain a second set of motion information predictors having a target number of motion information predictors or fewer.</claim-text> <claim-text>18. A method as claimed in claim 17, wherein the first set of motion information predictors are or include motion information predictors each associated with an image portion having a predetermined spatial and/or temporal relationship with the image portion being encoded.</claim-text> <claim-text>19. A method as claimed in claim 17 or 18, wherein the first set of motion information predictors are the motion information predictors used in Advanced Motion Vector Prediction.</claim-text> <claim-text>20. A method as claimed in any one of claims 17 to 19, comprising obtaining said one or more predetermined motion information predictors from a fixed set of predetermined motion information predictors.</claim-text> <claim-text>21. A method as claimed in claim 20, wherein the number of predetermined motion information predictors in said fixed set is greater than or equal to said target number.</claim-text> <claim-text>22. A method as claimed in claim 20 or 21, wherein the predetermined motion information predictors in said fixed set are all different.</claim-text> <claim-text>23. A method as claimed in claim 20, 21 or 22, wherein the fixed set of motion information predictors is the same for at least two different image portions to be encoded.</claim-text> <claim-text>24. A method as claimed in claim 23, wherein said first set of motion information predictors differs from one said image portion to the next.</claim-text> <claim-text>25. A method as claimed in any one of claims 20 to 24, comprising adding the fixed set of motion information predictors to said first set to form a combined set of motion information predictors.</claim-text> <claim-text>26. A method as claimed in claim 25, further comprising, if the number of motion information predictors in said combined set is greater than target number, removing one or more motion information predictors of said combined set to obtain said second set having said target number of motion information predictors.</claim-text> <claim-text>27. A method as claimed in claim 26, wherein the removing is carried out in dependence upon priorities of the motion information predictors.</claim-text> <claim-text>28. A method as claimed in claim 27, wherein the second set is obtained by selecting, from said combined set, Nrnax motion information predictors having higher priority and discarding one or more remaining motion information predictors having lower priority.</claim-text> <claim-text>29. A method as claimed in claim 26, wherein the motion information predictors in said combined set are ordered, and the second set is obtained by selecting from said combined set the first Nmax motion information predictors in the order thereof, where Nmax is said target number.</claim-text> <claim-text>30. A method as claimed in claim 29, wherein the first set of motion information predictors are the motion information predictors used in Advanced Motion Vector Prediction (AMVP) and are ordered in conformity with AMVP.</claim-text> <claim-text>31. A method according to claim to any one of claims 20 to 30, wherein the fixed set is the same for any sequence of digital images.</claim-text> <claim-text>32. A method according to claim 31, wherein the fixed set is the same for a set of consecutive image portions of digital images of said sequence of digital images.</claim-text> <claim-text>33. Apparatus for encoding a sequence of digital images into a bitstream, at least one portion of an image being encoded by motion compensation with respect to a reference image portion, the apparatus comprising: means for obtaining a first set of motion information predictors usable for an image portion to encode; and means for employing the first set of motion information predictors and one or more predetermined motion information predictors to obtain a second set of motion information predictors having a target number of motion information predictors or fewer.</claim-text> <claim-text>34. Apparatus for decoding a bitstream comprising an encoded sequence of digital images, at least one portion of an image being encoded by motion compensation with respect to a reference image portion, the apparatus comprising: means for obtaining a first set of motion information predictors usable for an image portion to decode; and means for employing the first set of motion information predictors and one or more predetermined motion information predictors to obtain a second set of motion information predictors having a target number of motion information predictors or fewer.</claim-text> <claim-text>35. A program which, when executed by a computer or processor, causes the computer or processor to carry out the method of any one of claims ito 32.</claim-text>