CN100566419C - Equipment and method with harmless mode encoding digital image data - Google Patents

Abstract

The method (900) of the signal of a kind of lossless compress and coded representation image information is proposed.Generate lossy compression method data file (922) and residual term compression data file (960).When diminishing the combination of compression data file and residual term compressed file, create consistent with a source data file in fact lossless data file.

Description

Equipment and method with harmless mode encoding digital image data

The application be that July 2, application number in 2002 are 02817150.0 the applying date, denomination of invention divides an application for the application for a patent for invention of " with the equipment and the method for harmless mode encoding digital image data ".

Technical field

The present invention relates to image processing and compression.More particularly, the present invention relates in the frequency field lossless coding to video image and audio-frequency information.

Background technology

In the principle of general Digital Signal Processing, digital picture is handled and is occupied very consequence.The importance of human visual perception has greatly encouraged interest and the progress to this area and digital picture processing science.In the field of transmission and receiving video signals, for example be used for projection film and film, Image Compression has been done multiple improvement.Many video systems present or that propose are used digital coding.These aspects in this field comprise: image encoding, image restoration and characteristics of image are selected.Image encoding is represented to attempt picture is launched on digital communication channel with effective and efficient manner, uses the least possible position to minimize desired bandwidth, meanwhile, distortion is maintained in certain limit.Image restoration represents to make great efforts to recover the work of the true picture of target.Image behind the coding is launched on communication channel can the distortion owing to multiple factor.From target establishment image the time, just may produce distortion sources.Feature selecting is meant some attribute of selecting picture.These attributes may be that the judgement in identification, classification and the wide in range content is desired.

For the digital coding of video, for example in digital camera, carry out, be a field of benefiting from improved Image Compression.Digital image compression generally can be divided into two classes: loseless method and damage method arranged.Lossless image can not lost any information and be resumed.Have damage method can cause some information irretrievably to be lost, this will be according to the quality of compression ratio, compression algorithm and the enforcement of algorithm.Usually, the method for lossy compression method considers to obtain the economic desired compression ratio of digital camera.In order to reach the credit rating of digital camera, this compression method need provide the performance rate of virtually lossless.Like this, still can cause the mathematics of information to be lost although compression is handled, the image fault that is caused by this losing is can not discovered by the audience in general viewing and admiring under the environment.

Existing digital image compression technology is used for other application, i.e. television system by development.These technology have been designed to take into account and have been suitable for required application, but do not satisfy to be used for movie theatre and to show needed quality requirement.

The digital camera compress technique need provide the visual quality of having experienced before the person of seeing a film.Ideally, the visual quality of digital camera will attempt to surmount the film that high-quality distribution is xeroxed.Simultaneously, compress technique should have practical high coding efficiency.As herein defined, code efficiency is meant that the compressed image quality satisfies the needed bit rate of a certain quality standard.Further, this system and coding techniques need built-in enough flexibilities holding different forms, and should be economical, just, and a kind of little size and effectively decoder or coder processes.

Many existing compress techniques can provide significant compression degree, but can cause the decline of video signal quality.Generally speaking, the specification requirement information compressed that is used to send information compressed is sent out with the constant bit rate.

A kind of compress technique of the video signal quality level that significant compression degree can be provided and keep expecting is used the piece and the sub-piece of the self adaptation size of discrete cosine transform (DCT) coefficient data of encoding.Just be called self adaptation block size discrete cosine transform (ABSDCT) method below this technology.This technology is at U.S. Patent number 5,021, open in 891, be entitled as " adaptive block sized images compression method and system " (" AdaptiveBlock SizeImage Compression Method And System "), the assignee who transfers the application is also incorporated herein by reference.The DCT technology is also in U.S. Patent No. 5,107, open in 345, be entitled as " adaptive block sized images compression method and system " (" Adaptive Block Size Image CompressionMethod And System "), the assignee who transfers the application is also incorporated herein by reference.In addition, that uses ABSDCT technology and difference quadtree conversion technology is combined in U.S. Patent No. 5,452, open in 104, be entitled as " adaptive block sized images compression method and system " (" Adaptive Block Size ImageCompression Method And System ") and transfer the application's assignee and incorporated herein by reference equally.Disclosed system uses and is referred to as " in the frame " Methods for Coding in these patents, and the coding of each frame of view data is not considered the content of any other frame in this method.Use the ABSDCT technology, the data rate that reaches can be reduced to about per second 5,000 ten thousand bits from about per second 1,500,000,000 bits, and discernible decline does not appear in picture quality.

The ABSDCT technology can be used for compressing black and white or coloured image, or the signal of presentation video.Input color signal can be the YIQ form, and wherein Y is brightness, or briliancy sampling and I and Q are colourity or color sampling, is 4:4:4 or extended formatting for each.Also can use other known form such as YUV, YC _bC _rOr rgb format.Because eyes are lower for the spatial sensitivity of color, most results of study shows that on vertical and horizontal direction the sub sampling that chrominance component is got 4 times of factors is rational.Thereby a vision signal is represented with 4 luma samples and two chroma samples.

Use ABSDCT, a vision signal generally will be divided into block of pixels to handle.For each piece, brightness and chromatic component are sent to a block size assignment element, perhaps an interleaver.For example, the piece of one 16 * 16 (pixels) is presented to this piece interleaver, and the piece interleaver is arranged in each 16 * 16 or the tissue image sampling is used for data block and the synthetic sub-piece that discrete cosine transform (DCT) is analyzed with generation.The DCT operation is a kind of method that the conversion of signals of time and spatial sampling is become the frequency representation of same signal.By converting frequency representation to, the DCT technique table has revealed and has allowed the very compression of high level, because quantizer can be designed to utilize the frequency distribution characteristic of image.In a preferred embodiment, 16 * 16DCT is applied to ordering for the first time, and 48 * 8DCT are applied to second ordering, and 16 4 * 4DCT are applied to the 3rd ordering and 64 2 * 2DCT are applied to the 4th ordering.

DCT operation has reduced spatial redundancy intrinsic in the video source.After carrying out DCT, most vision signal energy will focus on a few DCT parameter.Additional transformation, the difference quadtree conversion can be used for reducing the redundancy in the DCT parameter.

For 16 * 16 and each height piece, analyze DCT parameter value and DQT value (if using DQT) to determine the quantity of this piece of coding or the desired bit of sub-piece.Then, select to require the bit of minimum number to realize that the piece of encoding or the combination of sub-piece come the presentation video segmentation.For example, 28 * 8 sub-pieces, 64 * 4 sub-pieces and 82 * 2 sub-pieces can be selected for the presentation video segmentation.

The piece of selecting or the combination of sub-piece suitably are aligned in one 16 * 16 then.The DCT/DQT parameter value may pass through frequency weighting, quantification and coding (such as variable length code) to prepare to be used for transmission.Although above-mentioned ABSDCT technology shows quite well, its calculating strength is big.

In addition, be virtually lossless although use ABSDCT, it is wished sometimes accurately to come restore data with the same mode of its coding.For example, be used for controlling with archiving requirement and come packed data in the mode that can accurately return to original territory.

Traditional, the lossless compress system that is used for image comprises fallout predictor, the value of the current pixel that the fallout predictor estimation will be encoded.The residual term pixel is as the variance between the pixel of true and prediction and obtained.The residual term pixel-by-pixel basis is got off to be carried out entropy coding and is preserved or send.Because the association between the pixel has been removed in prediction, the residual term pixel has the dynamic range of reduction, and this dynamic range is that the bilateral index of feature (Laplce) distributes.Therefore should compression.Amount for the compression of residual term depends on prediction and coding method afterwards simultaneously.Most of normally used Forecasting Methodologies are differential pulse coding modulation (DPCM) and its mutation such as adaptive DPCM (ADPCM).

Use is that residual term still has very high energy based on the problem that the forecasting institute of pixel brings.This is owing in forecasting process, in fact only use the very adjacent pixels of smallest number.Therefore, aspect the code efficiency of the prediction scheme of pixel, also has very big space in improvement.

Summary of the invention

Embodiments of the invention have been described and have a kind ofly been carried out digital picture and video data encoding to realize the system of compression in harmless mode.This system is hybrid-type, means this system and has in the mode that diminishes and compress the part of described data and the another part that compresses the remainder data in harmless mode.For diminishing part, this system uses self adaptation block size discrete cosine transform (ABSDCT) algorithm.The described data of this ABSDCT system compresses also provide high visual quality and the high compression ratio.Decompress as source images with from the described ABSDCT system variance of the image that obtains of remainder image is obtained.This remainder uses the Golomb-Rice encryption algorithm to carry out lossless coding.Because based on the adaptive block of vision size with for the quantification of DCT coefficient, residual term has very low energy, and whole preferably lossless compress ratio so just is provided.

This ABSDCT system can realize the high compression ratio qualitatively at movie theatre.Because it is based on piece, it can remove correlation between the pixel better than other any schemes based on pixel.Therefore be used as fallout predictor in the lossless system that it will be described here.Increase a lossless coding system in conjunction with this fallout predictor and just form a mixing lossless compress system.It should be noted that this system can compress the static image and the image of motion.If static image only has the residual term data of this ABSDCT packed data and entropy coding to be used as compression output.For the sequence of motion, can do a judgement to determine to be to use still interframe compression in the frame.For example, if the picture frame of f (t) expression moment t, F (t) and F (t+ Δ t) are illustrated respectively in the DCT of moment t and t+ Δ t picture frame.Notice that Δ t represents two time intervals between the successive frame.

The present invention is embodied in a permission comes restore data in accurately identical with coded data mode the device and method of packed data.Embodiment comprises the system that carries out intraframe coding, interframe encode or both mixing.This system is based on the system of quality, uses the piece and the sub-piece of the self adaptation size of discrete cosine transform coefficient data.A pixel data blocks is imported into an encoder.This encoder comprises a block size assignment (BSA) element, and its block of pixels with input is carried out segmentation to be used for processing.The distribution of block size is based on input block and the variance of the sub-piece further cut apart.Usually, suppose that the mean value of piece and sub-piece falls into different predetermined scopes, then the zone that variance is bigger is divided into littler piece, and the less zone of variance is no longer carried out son and cut apart.Like this, at first the variance thresholding of piece is made amendment from its nominal value according to its mean value, and next the variance yields of piece and this threshold value compare, if variance yields greater than threshold value, then this piece carries out son and cuts apart.

Block size assignment is provided for an inverting element, and it is transformed into frequency domain data with pixel data.This conversion is only carried out on piece of selecting by block size assignment and sub-piece.For the AC element, next transform data is calibrated by quantification and serialization.The quantification of transform data is based on image quality evaluation, for example adjusts scale factor, coefficient counting, code rate distortion, block size assignment density and/or the former scale factor of contrast.Serialization, for example zigzag scanning is based on the longest possible run length of foundation for same value.Next data flow is encoded to prepare to be used for transmission by variable length coder.Coding can be a Huffman coding, or based on the coding of exponential distribution, encodes such as Golomb-Rice.

The hybrid compression system of use such as ABSDCT can be used as the good fallout predictor of pixel or DCT value.Therefore it can obtain than any use based on the higher lossless compress ratio of the prediction of pixel.Diminish part the result of digital camera quality is provided, i.e. the file that compression produces is a virtually lossless.For harmless part, be different from the Huffman coding, the Golomb-Rice coding does not require and generates any priori coding.Therefore, it does not need to preserve a jumbo code book as the Huffman coding.So just can use the resource of chip more efficiently, therefore, can dwindle with hard-wired die size.In addition, Golomb-Rice encoding ratio Huffman coding implements simpler.And because DCT parameter or residual term have the characteristic of the exponential distribution of nature, Golomb-Rice encoding ratio Huffman coding is realized higher code efficiency.Further, because the part that diminishes of compressibility is used the vision important information in the son of piece is cut apart, in the residual term coding, can inherit content model.This point is very important, is not used for the relevant content data that residual term is encoded because so just do not need extra memory register to collect.Because it is also very simple not use any estimation, this system to implement.

A kind of apparatus and method that are used for the signal of lossless compress and coded representation image information are proposed.The signal of presentation video information is compressed to create the image of compressed version.The compressed version of image is quantized, and creates this image that diminishes version.The image of compressed version carries out serialization equally to create the image of serialized a, version that quantize, compression.The image of this version is next decompressed, and the difference between definite source images reconciliation compressed version, and creates the image of a residual term version.Diminish the image of version and the image of remainder version and can export or output in combination with being separated, the image that diminishes version that wherein will decompress combines just identical with source images in fact with the image of residual term version.

The method of the signal of a kind of lossless compress and coded representation image information has been proposed.Generate one and diminish data file and a residual term compression data file.When diminishing data file and remainder compression data file and be combined, create consistent with a source data file in fact lossless data file.

Thereby the aspect of an embodiment provides a kind of equipment and method that lossless compress effectively is provided.

Lossless compress digital picture and audio-frequency information are come with the purpose that is of value to control and file in another aspect of an embodiment.

The another aspect of an embodiment provides a kind of lossless compress system based on interframe.

An embodiment provides a kind of based on the lossless compress system in the frame more on the one hand.

One embodiment of the present of invention have been described and a kind of the data that comprise source images have been carried out apparatus for encoding, this device comprises: the data that are used for the described source images of compression expression are also created the device of the compressed version of described source images thus, and the data that formerly generate by source images being carried out the adaptive block size adjust have been used in wherein said compression; The device that diminishes version that is used to quantize the compressed version of described source images and creates described source images thus; Be used to decompress the compressed version of described source images to create the device of the image through decompressing, the data that formerly generate by source images being carried out the adaptive block size adjust have been used in wherein said decompression; The device that is used for determining the difference between described source images and the described image and creates the residual term data that are associated with described source images thus through decompressing; And the device that diminishes version and described residual term data that is used to export described source images, diminish version and the described residual term data of wherein said source images can be used to create and described source images uniform images in fact.

An alternative embodiment of the invention has been described and has a kind ofly been carried out Methods for Coding to comprising from the data of the multiple source frame of source images, described method comprises: the data of the first source frame of the described multiple source frame of compression expression, and create the compressed version of the described first source frame thus, the data that formerly generate by described multiple source frame being carried out the adaptive block size adjust have been used in wherein said compression; The compressed version that quantizes the described first source frame is also created the version that diminishes of the described first source frame thus; Decompress the compressed version of the described first source frame to create the frame through decompressing, and the data that formerly generate by described multiple source frame being carried out the adaptive block size adjust have been used in wherein said decompression; Determine the difference between the second source frame and the described frame and create the residual term data that are associated with the described first source frame thus through decompressing; And export the described first source frame diminish version and described residual term data, the version that diminishes of the wherein said first source frame can be used to create and the consistent in fact frame of the described first source frame with described residual term data.

Another embodiment of the present invention has described and has a kind ofly carried out apparatus for encoding to comprising from the data of the multiple source frame of source images, described device comprises: be used for the first source frame of the described multiple source frame of compression expression data, and create the device of the compressed version of the described first source frame thus, the data that formerly generate by described multiple source frame being carried out the adaptive block size adjust have been used in wherein said compression; The device that diminishes version that is used to quantize the compressed version of the described first source frame and creates the described first source frame thus; Be used to decompress the compressed version of the described first source frame to create the device of the frame through decompressing, the data that formerly generate by described multiple source frame being carried out the adaptive block size adjust have been used in wherein said decompression; The device that is used for determining the difference between the second source frame and the described frame and creates the residual term data that are associated with the described first source frame thus through decompressing; And the device that diminishes version and described residual term data that is used to export the described first source frame, the version that diminishes of the wherein said first source frame can be used to create and the consistent in fact frame of the described first source frame with described residual term data.

Description of drawings

Feature of the present invention and advantage will be by the detailed description back done below in conjunction with accompanying drawing more obvious, in the accompanying drawings, identical label indicates identical feature all the time, wherein:

The block diagram of the coded portion in image compression of Fig. 1 and the treatment system;

Fig. 2 is the block diagram of the decoded portion in an image compression and the treatment system;

Fig. 3 illustrates the flow chart that relates to based on the treatment step of the block size assignment of variance;

Fig. 4 a illustrates the exponential distribution of Y component run length in the DCT coefficient matrix;

Fig. 4 b illustrates C in the DCT coefficient matrix _bThe exponential distribution of component run length;

Fig. 4 c illustrates C in the DCT coefficient matrix _rThe exponential distribution of component run length;

Fig. 5 a illustrates the exponential distribution of Y component amplitude size in Y component amplitude size or the DCT coefficient matrix;

Fig. 5 b illustrates C _bC in component amplitude size or the DCT coefficient matrix _bThe exponential distribution of component amplitude size;

Fig. 5 c illustrates C _rC in component amplitude size or the DCT coefficient matrix _rThe exponential distribution of component amplitude size;

Fig. 6 illustrates the Golomb-Rice encoding process;

Fig. 7 illustrates the equipment that is used for the Golomb-Rice coding;

Fig. 8 illustrates the processing of encoding D C component value;

Fig. 9 illustrates the equipment that is used for lossless compress; And

Figure 10 illustrates the method for mixing lossless compress.

Embodiment

For the Digital Transmission that realizes digital signal and utilize its advantage, need to use the Signal Compression of some forms usually.When realizing the high compression ratio in a result images, the high-quality of keeping image is important too.And, be used for also expecting its computational efficiency when miniature hardware is realized that this is very important in a lot of the application.

Before in detail explaining one embodiment of the present of invention, need understand that the present invention is not limited to detailed structure that be applied to illustrate in the following description or illustrated in the accompanying drawings and for the arrangement of parts.The present invention can be used for other embodiment and realize in a different manner.Equally, needing to understand wording used herein and term is not to be regarded as restriction for purpose of description.

The image compression that an aspect among embodiment is used is based on discrete cosine transform (DCT) technology, disclosed in for example unsettled U.S. Patent application " the adaptive block size DCT image compression based on variance of contrast sensitivity " (" Contrast Sensitive Variance Based Adaptive Block Size DCTImage Compression "), sequence number is No.09/436,085, be filed on November 8th, 1999, the assignee who transfers the application is also incorporated herein by reference.Use image compression and the decompression systems of DCT in co-pending United States Patent (USP) " based on the image compression of quality " (" Quality BasedImage Compression "), to describe, sequence number is No.09/494,192, be filed on January 28th, 2000, the assignee who transfers the application is also incorporated herein by reference.Usually, an image of handling in digital field is made up of the pixel data of a matrix of the piece that is divided into non-overlapping copies, and size is NxN.On each piece, can carry out the DCT of bidimensional.The DCT of this bidimensional is defined by following relationship:

$X(k,l)=\frac{\mathrm{\α}\left(k\right)\mathrm{\β}\left(k\right)}{\sqrt{N*M}}\underset{m=0}{\overset{N-1}{\mathrm{\Σ}}}\underset{n=0}{\overset{N-1}{\mathrm{\Σ}}}x(m,n)\cos \left[\frac{(2m+1)\mathrm{\πk}}{2N}\right]\cos \left[\frac{(2n+1)\mathrm{\πl}}{2N}\right],0\≤k,l\≤N-1$

Wherein $\mathrm{\α}\left(k\right),\mathrm{\β}\left(k\right)=\left\{\begin{array}{cc}1,\mathrm{if}& k=0\\ \sqrt{2},\mathrm{if}& k\≠0\end{array}\right.,$ And

X (m, n) be in a NxM piece in the position (m, the n) pixel on, and

(k l) is corresponding DCT coefficient to X.

Because pixel value is nonnegative value, DCT divides X (0,0) to be positive always and generally to have most energy.In fact, for typical image, most transformation energy concentrate on component X (0,0) around.The characteristic of this energy compression makes the DCT technology become a kind of attractive compression method.

This Image Compression uses the contrast adaptive coding to realize the reduction of further bit rate.Most of natural images that make discovery from observation are formed by the smooth region that relatively slowly changes with such as numerous change zone of object boundary and high-contrast texture.Contrast adaptive coding scheme utilizes this point to the more position of numerous change region allocation and to the less position of not too busy region allocation.

The contrast adaptive approach uses intraframe coding (spatial manipulation) rather than interframe encode (space-time processing).Interframe encode requires more complicated treatment circuit inherently, also requires a plurality of frame buffer zones.In many application, need to reduce complexity in actual the realization.Intraframe coding also can be used the space-time encoding scheme being lost efficacy and show in the very poor environment.For example, because mechanical shutter makes the time of integration shorter relatively, the film of per second 24 frames just can divide goes into this class.The short time domain aliasing that allows higher degree the time of integration.For motion fast,, thereby destroyed hypothesis for correlation between frame and the frame because it becomes jerking movement.When using the power line frequency of 50Hz and 60Hz simultaneously, intraframe coding also is easy to carry out standardization.TV sends with 50Hz or 60Hz now.Use intraframe coding, can adapt to the operation of 50Hz and 60Hz, perhaps by compromise even can adapt to the film of per second 24 frames to frame rate with respect to spatial resolution as digital scheme.

For image processing, the DCT operation is what to carry out on the pixel data of the piece that is divided into a non-overlapping copies in the matrix.Although the size that it is noted that discussed here is NxN, it all is operable can predicting multiple block size.For example, can use the block size of NxM, wherein N and M are that integer and M can be greater than or less than N.Another important aspect is that this piece can be divided at least one straton piece, for example N/ixN/i, N/ixN/j, N/ixM/j or the like, and wherein i and j are integers.In addition, demonstration block size discussed here is one and has corresponding DCT piece and 16 * 16 block of pixels of sub-piece.It is also envisioned that various other integers for example two even numbers or two odd numbers can use, such as 9 * 9.

Fig. 1 and 2 illustrates an image processing system 100 in conjunction with configurable serial device notion.Image processing system 100 comprises the encoder 104 of the vision signal that compression receives.The signal of compression uses transmission channel or physical medium 108 to send, and is received by decoder 112.Decoder 112 is decoded into image pattern with the coded data that receives, and next image pattern may be demonstrated.

Usually, an image is divided into a plurality of block of pixels to handle.A chrominance signal can use RGB to YC ₁C ₂Transducer 116 is transformed into YC from rgb space ₁C ₂The space, wherein Y is brightness or briliancy component, and C ₁And C ₂Be colourity, or color component.Because eyes are for the low spatial susceptibility of color, many systems come sub sampling C with 4 times of factors in the horizontal and vertical directions ₁And C ₂Component.Yet sub sampling is unwanted.The image of a complete resolution, i.e. the 4:4:4 form known to may be very useful or necessary in some use such as alleged " digital camera ".Two kinds of possible YC ₁C ₂Expression is that YIQ represents to represent with YUV that the both is known in the industry.A kind of variation that also can use YVU to represent is alleged YC _bC _rThis can further be divided into the odd even component.Thereby, use Y-idol, strange, the C of Y-in one embodiment _b-idol, C _b-Qi, C _r-idol, C _r-Qi represents.

In a preferred embodiment, handle odd even Y, C _bAnd C _rIn the component each and do not carry out sub sampling.Like this, the input of each in 6 of one 16 * 16 block of pixels components is provided for encoder 104.In order to illustrate, illustrate the encoder 104 that is used for the Y-even component.Similarly encoder is used to Y-odd component and odd even C _bAnd C _rComponent.Encoder 104 comprises block size assignment device 120, and the branch that it carries out block size is equipped with prepares to be used for video compression.But this block size assignment device 120 is determined decomposition to 16 * 16 based on the Perception Features of the image in this piece.Block size assignment is divided into littler piece with each 16 * 16 son, and such as 8 * 8,4 * 4 and 2 * 2, this is cut apart is with a kind of quaternary tree form according to the activity in 16 * 16.Block size assignment device 120 generates the quaternary tree data, is called the PQR data, and its length can be between 1 to 21 bit.Like this, cut apart if block size assignment is determined 16 * 16 needs, R position PQR data then are set and add 4 additional Q data thereafter, 4 Q data are corresponding to 48 * 8 that are divided into.If block size assignment determines that any 8 * 8 needs carry out son and cut apart, then carry out 8 * 8 that son cuts apart and increase by 4 P data again for each.

With reference now to Fig. 3,, provides the flow chart of the operation of detailed expression block size assignment element 120.The variance of piece is regarded as adjudicating the tolerance when whether son is cut apart a piece.From step 202, read one 16 * 16 block of pixels.In step 204, calculate 16 * 16 variance v16, its variance is calculated with following formula:

$\mathrm{var}=\frac{1}{{\mathrm{<figure-callout\; id="2"\; label="N"\; filenames="C2006101464760026H1.png,C2006101464760027H1.png"\; state="\{\{state\}\}">N</figure-callout>}}^2}\underset{i=0}{\overset{N-1}{\mathrm{\&Sigma;}}}\underset{j=0}{\overset{N-1}{\mathrm{\&Sigma;}}}{x}_{i,j}^2-{\left(\frac{1}{{\mathrm{<figure-callout\; id="2"\; label="N"\; filenames="C2006101464760026H1.png,C2006101464760027H1.png"\; state="\{\{state\}\}">N</figure-callout>}}^2}\underset{i=0}{\overset{N-1}{\mathrm{\&Sigma;}}}\underset{j=0}{\overset{N-1}{\mathrm{\&Sigma;}}}{x}_{i,j}\right)}^2$

N=16 wherein, and x _{I, j}Be to be positioned at i in the NxN piece ^ThRow j ^ThThe pixel of row.In step 206, at first the mean value of if block is then revised variance thresholding T16 so that new variance thresholding T ' 16 to be provided between two predetermined values.Next piece variance and new thresholding T ' 16 compare.

If variance v16 is not more than thresholding T16, then in step 208,16 * 16 initial address is written into temporary storage, and R position PQR data are set to 0 to represent that these 16 * 16 do not have quilt to cut apart.Next this algorithm reads next 16 * 16 block of pixels.If variance v16 is greater than thresholding T16, then in step 210, the PQR data of R position are set to 1 to represent that these 16 * 16 quilts are divided into 48 * 8.

These 48 * 8, i=1: 4, consideration may be carried out further son and cuts apart, shown in step 212.For each 8 * 8, in step 214, calculate variance v8 _iIn step 216, at first the mean value of if block is then revised variance thresholding T8 so that a new thresholding T ' 8 to be provided between two predetermined values, and next piece variance and new thresholding compare.

If variance v8 _iBe not more than thresholding T8, then in step 218,8 * 8 initial address is written into temporary storage, and corresponding Q position, Q _iBe set to 0.Handle next 8 * 8 then.If variance v8 _iGreater than thresholding T8, then in step 220, corresponding Q position, Qi is set to 1 to represent that these 8 * 8 are divided into 44 * 4 with quilt.

These 44 * 4, j _i=1: 4, consideration may be carried out further son and cuts apart, shown in step 222.For each 4 * 4, in step 224, calculate variance v8 _IjIn step 226, at first the mean value of if block is then revised variance thresholding T4 so that a new thresholding T ' 4 to be provided between two predetermined values, and next piece variance and new thresholding compare.

If variance v4 _IjBe not more than thresholding T4, then in step 228, this first address of 4 * 4 is written into, and corresponding P position, P _IjBe set to 0.Handle next 4 * 4 then.If variance v4 _IjGreater than thresholding T4, then in step 230, corresponding P position, P _IjBe set to 1 to represent that these 4 * 4 will be divided into 42 * 2 by further son.In addition, these 42 * 2 are written in the temporary storage.

Thresholding T16, T8 and T4 can be predetermined constants.This is called hard decision.Perhaps, also can use self adaptation or soft-decision.For example, soft-decision changes the thresholding that is used for variance according to the average pixel value of 2N * 2N piece, and wherein N can be 8,4 or 2.Like this, the function of average pixel value can be used as thresholding.

In order to illustrate, consider following example.The variance thresholding of the predetermined Y of being used for component is made as 50,1100 and 880, is respectively applied for 16 * 16,8 * 8 and 4 * 4 piece.In other words, T16=50, T8=1100 and T4=880.The scope of mean value is made as 80 to 100.The variance of supposing to calculate 16 * 16 is 60.Because 60 greater than T16, and mean value 90 is between 80 to 100, and these 16 * 16 quilts are divided into 48 * 8 sub-pieces.Suppose that 8 * 8 the variance that calculates is 1180,935,980 and 1210.Because two 8 * 8 variance has surpassed T8, these two pieces cut apart further quilt to generate 84 * 4 sub-pieces altogether.At last, suppose that 84 * 4 variance is 620,630,670,610,590,525,930 and 690, corresponding average is 90,120,110,115.Because the mean value of first 4 * 4 sub-piece drops within the scope (80,100), its threshold value will be reduced for T ' 4=200, and will be littler than 880.So this 4 * 4 sub-piece is cut apart with the 7th 4 * 4 the same son that will carry out of sub-piece piece.

Notice that similarly processing will be used to distribute the strange and color component C of luminance component Y- _bAnd C _rBlock size.Chrominance component can flatly, vertically carry out selecting, and perhaps both all are.

In addition, be described as a top-down method although should be noted that block size assignment, promptly Zui Da piece (being 16 * 16 in this example) is at first estimated, uses bottom-up method also to be fine.Bottom-up method will at first be estimated minimum piece (being 2 * 2 in this example).

Get back to Fig. 1, the PQR data are provided for DCT element 124 together with selected address.DCT element 124 uses the PQR data to carry out the suitably discrete cosine transform of size on the piece of selecting.Only the piece of Xuan Zeing need carry out the DCT processing.

Image processing system 100 comprises that also DQT element 128 is used for reducing the redundancy of the DC coefficient of DCT.The DC coefficient is positioned at the upper left corner of each DCT piece.Usually, with the AC coefficient ratio, the DC coefficient ratio is bigger.Because making, the variance on the size is difficult to design an effective variable length coder.Therefore, the redundancy that reduces in the DC coefficient is useful.

DQT element 128 carries out the DCT of 2-D, each 2 * 2 on the DC coefficient.Originate in 2 * 2 in 4 * 4, the DCT of a 2-D carries out on 4 DC coefficients.The DCT of this 2-D is called differential quaternary tree conversion, the perhaps DQT of 4 DC coefficients.Next, three adjacent DC coefficients among the DC coefficient of this DQT and 8 * 8 are used to calculate other DQT of next stage.At last, 48 * 8 DC coefficient in 16 * 16 is used to calculate DQT.Like this, in one 16 * 16, has only a real DC and other all are AC coefficients corresponding and DCT and DQT.

Conversion coefficient (DCT and DQT) is provided for quantizer and quantizes.In a preferred embodiment, DCT coefficient frequency of utilization weighting mask (FWMs) and a quantitative calibration factor quantize.FWM is that input DCT coefficient block is at the table with the frequency weight on the one dimension.Frequency weight is used different weights for different DCT coefficients.Weight is designed to the input sample of the frequency content with human vision or optical system sensitivity is strengthened, and weakening does not have the input sample of responsive frequency content to human vision or optical system.Weight also can be based on such as the factor of viewing distance or the like and design.

The selection of weight is based on empirical data.A kind of method of the weight masks that is designed for 8 * 8DCT coefficient is at IS0/IEC JTC1 CD 10918, " digital compression of continuous tone rest image and coding first: require and guilding principle " (" Digital compression and encoding ofcontinuous-tone still images-part 1:Requirements and guidelines "), International Standards Organization, 1994, incorporated herein by reference.Usually, design two FWMs, one is used for luminance component and another is used for chromatic component.Be used for block size and be 2 * 2,4 * 4 FWM table by selecting and obtain to be used for 16 * 16 tables and obtain by interpolation to 8 * 8 tables to being used for 8 * 8 tables.The quality and the bit rate of the coefficient that scale factor control quantizes.

Like this, each DCT coefficient quantizes according to following relationship:

Wherein DCT (i j) is the DCT coefficient of input, fwm (i j) is the frequency weighting mask, and q is a scale factor, and DCTq (i j) is coefficient through quantizing.Notice that according to the symbol of DCT coefficient, first in the bracket rounds up or down.The DCT coefficient also uses suitable weight masks to quantize.Yet, can use a plurality of tables or shielding, and be applied to Y, C _bAnd C _rComponent.

At frame 130, next the AC value is separated and is separately handled from the DC value.For the DC element, be encoded for the DC component value of each section.Each of each section DC component value afterwards is represented as the variance between the DC component of it and its front then, and is encoded, as frame 134.For lossless coding, initial DC component value of each section and difference are used and are encoded in conjunction with the Golomb-Rice shown in Fig. 6,8, as frame 138.Use the difference between the continuous DC component value of Golomb-Rice coding to help making the differential of DC component value to trend towards having bilateral exponential distribution.Next data are used buffer 142 and store temporarily, and transmit or send to decoder 112 by transmission channel 108 then.

Fig. 8 illustrates the processing of encoding D C component value.This processing can be used for equally rest image, video image (such as, but be not limited to moving image or high definition television) and audio frequency.Provide a section in the data, as step 804, a DC component value of this section is retrieved, as step 808.The one DC component value is encoded then, as step 812.Be different from the AC component value, the DC component value does not need to quantize.In one embodiment, use single values regardless of the inefficacy of block size assignment for one 16 * 16.Can predict the piece of any fixed size, such as 8 * 8 or 4 * 4, perhaps any variable-block size by the block size assignment definition can be used.The second, perhaps then, fetch one given section DC component value, as step 816.The 2nd a DC component value and a DC component value compare, its difference, and perhaps residual term is encoded, as step 820.Like this, the 2nd DC component value advances to be represented as the difference between itself and first value.This processing all comes again for the DC component value of each section.Like this, inquire about, step 824 has determined whether to arrive the afterbody (last piece and last DC value) of section.To step 828, fetch the DC value of next section if not then, as step 816 and repeat above-mentioned processing.If then to step 832, fetch next section, as step 804 and till repeating this processings all frame and all sections in all frames in this document and all handling.

Purpose for the lossless coding of DC component value is to generate the remainder value with low variance.Use DCT, DC coefficient component value has been contributed maximum pixel energy.Therefore, by not quantizing the variance that the DC component value reduces residual term.

For the AC component, next data block and frequency weighting mask are calibrated by quantizer 146 or scale factor element.Thereby the quantification of DCT coefficient makes in them most be reduced to 0 reaches compression effects.In a preferred embodiment, 32 scale factors corresponding to mean bit rate are arranged.Be different from other compression methods such as MPEG2 etc., the control of mean bit rate is based on the quality of image of processing rather than the state of target bit rate and buffer memory.

In order further to increase compression, the coefficient of quantification is provided for scanning serializer 150.The coefficient block that 150 scannings of scanning serializer quantize is to produce the serial flow of quantization parameter.Zigzag scanning, column scan or line scanning can be used.Can select the multiple different zigzag scan mode and the mode of non-zigzag scanning.A kind of preferable technology uses 8 * 8 to be used for zigzag scanning.Zigzag scanning for quantization parameter has increased the possibility that runs into the 0 big value distance of swimming.This 0 distance of swimming has the probability that successively decreases inherently, and available Huffman coding carries out efficient coding.

AC coefficient stream serial, that quantize is provided for variable length coder 154.The AC component value can use Huffman coding or Golomb-Rice to encode.For the DC component value, use the Golomb-Rice coding.The run length coding, RLC device has separated the coefficient of 0 coefficient and non-0, and describes in detail in Fig. 6.In one embodiment, use the Golomb-Rice coding.The efficient coding of Golomb-Rice coding energy has the nonnegative integer of exponential distribution.Use the Golomb coding to compress the variable that the coding that shorter length can more preferably be provided is used for exponential distribution.

In the Golomb coded run length, the Golomb sign indicating number undertaken coefficientization by nonnegative integer m.For example, provide a Coefficient m, the Golomb of positive integer n coding adds that by the merchant n/m with monobasic sign indicating number form the remainder of the binary code of modification represents, if remainder smaller or equal to

Then its length is

The position, otherwise length is

The Golomb-Rice coding is a kind of special shape of Golomb coding, and wherein Coefficient m is expressed as m=2 ^kIn this case, merchant n/m obtains by the k position that moves right of the Integer n with binary representation.Like this, the Golomb-Rice sign indicating number is both serial connections.The Golomb-Rice coding can be used for coding and have the positive or negative integer that bilateral how much (index) distributes, and this distribution table is shown

p _α(x)＝cα ^|x| (1)

In (1), α is the characteristic coefficient of the decay of expression probability x, and c is a normalization constants.Because P _α(x) be dull, visible a series of integer value should satisfy

p _α(x _i＝0)≥p _α(x _i＝-1)≥p _α(x _i＝±1)≥p _α(x _i＝-2)≥... (2)

As Fig. 4 a, 4b, 4c and 5a, 5b, shown in the 5c, 0 distance of swimming and amplitude in the DCT coefficient matrix of quantification all have exponential distribution.Distribution shown in these figure is based on the data from true picture.Fig. 4 a illustrates the Y component distribution 400 of 0 run length with respect to correlated frequency.Similarly, Fig. 4 b and 4c illustrate the C of 0 run length with respect to correlated frequency respectively _bAnd C _rComponent distributes 410 and 420.Fig. 5 a illustrates the Y component distribution 500 of amplitude size with respect to correlated frequency.Similarly, Fig. 5 b and 5c illustrate the C of amplitude size with respect to correlated frequency respectively _bAnd C _rComponent distributes 510 and 520.Attention is at Fig. 5 a, the distribution of the curve representation DCT coefficient magnitude of 5b and 5c.The scope of a coefficient value of each size expression.For example, sizes values be 4 have scope for 15 ,-14 ...-8,8..., 14,15} has 16 values.Similarly, sizes values be 10 have scope for 1023 ,-1022 ... ,-512,512 ..., 1022,1023} is 1024 values altogether.From Fig. 4 a, 4b, 4c, 5a, as seen, run length and amplitude size all have exponential distribution among 5b and the 5c.The actual distribution of shown amplitude can be used following equation model:

$p\left(X_{k,l}\right)=\frac{\sqrt{2\mathrm{\&lambda;}}}{2}\exp \{-\sqrt{2\mathrm{\&lambda;}}|X_{k,l}\left|\right\},k,l\&NotEqual;0---\left(3\right)$

In (3), X _{K, l}The expression correspondence is that k and horizontal direction upper frequency are the DCT coefficient of l with the vertical direction upper frequency, and average ${\mathrm{\&mu;}}_x=\frac{1}{\sqrt{2\mathrm{\&lambda;}}},$ Variance ${\mathrm{\&sigma;}}^2=\frac{1}{2\mathrm{\&lambda;}}.$ Thereby, even more ideal with the data processing that described form is used Golomb-Rice to encode to carry out among the DCT.

Although following description is the compression of combining image data, these embodiment can be used for the embodiment of audio compressed data equally.In the compression of view data, image or video data may be, for example, and RGB or YIQ or YUV or have linearity or to the YC of the pixel value components of number encoder _bC _r

Fig. 6 illustrates the processing 600 of coding 0 and non-0 coefficient.When the DCT matrix is scanned, 0 and non-0 coefficient handle and separated with being separated, as step 604.For 0 data, determine the length of 0 distance of swimming, as step 608.Notice that run length is a positive integer.For example, if run length is n, then the Golomb Coefficient m is determined, as step 612.In one embodiment, the Golomb coefficient is confirmed as the function of run length.In another embodiment, Golomb coefficient (m) is determined by following formula (4)

Can be randomly, the length of run length is counted by counter or register with relevant Golomb coefficient, as step 616.In order to encode 0 run length n, the merchant is encoded, as step 620.In one embodiment, this merchant is confirmed as 0 the run length and the function of Golomb coefficient.In another embodiment, merchant (Q) is determined by following formula (5):

In one embodiment, merchant Q is encoded as the monobasic sign indicating number, needs the Q+1 position.Next, remainder is encoded, as step 624.In one embodiment, remainder is encoded as operation function and merchant's function.In another embodiment, remainder (R) uses following formula (6) to determine:

R＝n-2 ^mQ (6)

In one embodiment, remainder R is encoded to the binary coding of m position.Then, determine merchant Q and remainder R, the coding of serial connection Q and R is used for the binary encoding of 0 run length n with expression as step 628.

Non-0 coefficient also uses Golomb-Rice to encode.Because coefficient amplitude can be positive or negative, just need sign bit of use and the absolute value of the amplitude that provides is encoded.The amplitude of non-0 coefficient that provides is x, and its amplitude can be expressed as the function of this absolute value and this symbol.Thereby this amplitude can use following formula (7) to be expressed as y:

Thereby the value of non-0 coefficient also can randomly be counted by counter or register, as step 632.Next determine in step 636 that whether this amplitude is more than or equal to 0.If be encoded as in its value of step 640 and double the value that provides.If not, be encoded as the value of its absolute value two demultiplications 1 in this value of step 644.Can predict other mapping schemes also can use.Its key is not need extra being used to distinguish the sign bit of this value.

With the amplitude coding that equation (7) carries out, its result becomes odd-integral number for positive x value becomes the negative x value of even-integral number.Further, this mapping has kept the probability distribution of the x in (2) formula.The advantage of the coding of (7) formula of use explanation is to avoid using sign bit for representing positive number and negative.After mapping, y encodes with the form same with being used for 0 distance of swimming.All coefficients that processing is proceeded in current block all are scanned.

Although it is important to point out that the function that in the embodiments of the invention value of coefficient and run length is defined as equation (1)-(7), do not need to use definite equation (1)-(7).This is to have utilized the exponential distribution of Golomb-Rice coding and DCT coefficient to accomplish for image and the more efficiently compression of voice data.

Owing to 0 distance of swimming after coding can't distinguish, come first 0 distance of swimming of mark appearance with regard to the special prefix code that needs the use regular length from non-0 amplitude.Usually non-0 amplitude calculates all 0 later on again in having calculated a piece.In some cases, use be called EOB (EOB) sign indicating number may be more effective than using the Golomb-Rice sign indicating number.The EOB sign indicating number also can be the sign indicating number of a special regular length.

According to equation (1) or (3), the amplitude of DCT coefficient matrix or the probability distribution of run length undertaken coefficientizations by α or λ.It has represented can improve for the efficient of the coding that appears at the content under the specific DCT coefficient block.Next can use suitable Golomb-Rice coefficient that interested amount (quantity ofinterest) is encoded.In one embodiment, counter and register are used to each run length and range value to calculate the corresponding number of times that divides other accumulated value and the appearance of this value.For example, the number of elements of accumulating if register is used to preserve accumulated value is respectively R _R1And N _R1, following formula (6) can be used as Rice-Golomb coefficient this run length of encoding:

Similarly handle and can be used for amplitude.

The residual term pixel produces by at first using the decompress data of this compression of ABSDCT decoder, it being deducted from source data again.The dynamic range of residual term is more little, and the ratio of compression is just high more.Because compression is based on piece, residual term is also based on piece.A fact of knowing is that the residual term pixel has bilateral exponential distribution, and the center is generally 0.Because the Golomb-Rice sign indicating number is better for these type of data, the Golomb-Rice encoding process is used to compress the residual term data.Yet,, do not need specific code owing to there is not run length to encode.Further, do not need the EOB coding.Like this, packed data comprises two components.One-component is from the lossy compression method device and another is from the lossless compress device.

When the coding motion sequence, also can utilize temporal correlation.In order to utilize temporal correlation fully, at first estimation is carried out motion compensated prediction to obtain the residual term pixel then owing to the pixel shift that motion causes.Because ABSDCT carries out the adaptive block size coding, block size information replacedly is used as the tolerance of the displacement that is caused by motion.For further simplification, do not use the scene change-detection.As an alternative, at first obtain intra-frame compression data for each frame in the sequence.Next arrive the basis generation present frame DCT of frame and the difference between the previous frame DCT based on frame.This is by U.S. Patent Application Serial Number No.09/877, and 578 further describe, and is filed in June 7 calendar year 2001, and is incorporated herein by reference.Residual term in these DCT territories uses Huffman and Golomb-Rice encoding process to encode.That of the quantity minimum of the bit that final compression output is used corresponding to every frame.

This lossless compression algorithm is a hybrid plan, and it can adapt to utilization or code conversion again by the harmless part of removing wherein.Like this, use the correlation of ABSDCT maximization pixel in space field, can make residual term pixel and those use the pixel of prediction scheme to compare and have lower variance.The part that diminishes of total system allows the user to realize the quality that is used to distribute that needs and data rate and need not handle by means of interframe, thereby has eliminated the complexity that relevant artificial latent shadow with motion disturbed and reduced significantly enforcement.This program that is used for digital cinema applications in distribution seems and is even more important, the quality of grade because the distribution that diminishes part of compression material is had relatively high expectations.

Fig. 9 illustrates and mixes lossless coding equipment 900.Figure 10 illustrates the processing that may operate on this equipment.Source digital information 904 is kept in the memory device, perhaps is launched.Many elements among Fig. 9 are described in detail in Fig. 1 and 2.Frame is sent to compressor reducer 908, and compressor reducer 908 comprises block size assignment element 912, DCT/DQT inverting element 916 and quantizer 920.After data were carried out DCT/DQT, these data were switched to frequency domain.At an output 922, data are quantized by quantizer 920 and are sent to output 924, and output 924 comprises memory and/or switch.All processing described above are on the basis in frame.

The output of quantizer also is sent to decompressor 928.Decompressor 928 carries out the destruction operation of compressor processes, by inverse DCT 932 and IDQT/IDCT 936, together with the knowledge according to the PQR data that defined by BSA.The result of decompressor 940 is provided for subtracter 944 and compares with source data.Subtracter 944 can be multiple element, such as difference engine, calculates the difference of the pixel of residual term pixel after as the unpressed and compressed decompression of each piece.In addition, difference engine can obtain the interframe encode with good conditionsi that residual term in the DCT territory is used for each piece.Comparative result 948 between decompressed data and the source data is exactly a pixel residual term file.That is to say, as a result 948 expressions losing through the data after the compression and decompression.Like this, source data just equal to export 922 with result 948 combination.948 next be serialized 952 and as a result, and be provided as second output 960 by Huffman and/or Golomb Rice coding 956.Huffman and/or Golomb Rice encoder 956 are entropy coders of a kind of form, and it uses Golomb Rice coding that the residual term pixel is encoded.Adjudicate to determine to be to use in the frame or interframe encode based on the least bits that is used for each frame.Use Golomb Rice coding residual term can improve the compression ratio of whole system.

Like this, this can't harm, interframe output is the combination or the mixing of two groups of data, image file that promptly diminish, high-quality (922, or A) and the residual term file (960, or C).

Also can use interframe encode.The output of quantizer is sent to memory 964, and what send together also has knowledge for BSA.Behind the valid data that obtained a frame, subtracter 966 compares frame 964 and the next frame of preserving 968.Its difference generates a DCT residual term 970, and next this residual term carries out serialization and/or Golomb-Rice coding 974, to provide the 3rd output data groups 976 to output 924.Harmless file B of interframe and C like this, have just been compiled.Like this, can select combination (A+C or B+C) based on the consideration of size.Further, for the purpose of editing, may need a pure interframe output.

Get back to Fig. 1, the compressing image signal that encoder 104 generates can use buffer memory 142 to preserve temporarily, uses transmission channel 108 to send to decoder 112 then.Transmission channel 108 can be physical medium, such as magnetic or light storage device, or wired or wireless transmission process or equipment.The PQR data that comprise block size assignment information also are provided for decoder 112 (Fig. 2).Decoder 112 comprises buffer memory 164 and variable-length decoder 168, encoder 168 decoding run lengths and non-0 value.But variable-length decoder 168 with Fig. 6 in the similar opposite mode described operate.

The output of variable-length decoder 168 is provided for de-serializer 172, according to the sweeping scheme array factor that uses.For example, used zigzag scanning, vertical scanning and horizontal sweep if mix, oppositely serialiser 172 will use its employed scan type of knowing to come permutatation coefficient suitably.De-serializer 173 receives the PQR data and helps correctly coefficient is aligned in the synthetic coefficient block.

Synthetic piece is provided for inverse DCT 174, is used to cancel the processing of using quantizer scale factor and frequency weighting mask.

If used the difference quadtree conversion, next coefficient block is provided for IDQT element 186, then is IDCT element 190.Otherwise coefficient block is provided directly to IDCT element 190.IDQT element 186 and IDCT element 190 inverse transformation coefficients are to generate pixel data blocks.Next pixel data must carry out interpolation, transforms to the RGB form, and preservation is used for later demonstration.

Fig. 7 illustrates the equipment 700 that is used for the Golomb-Rice coding.The equipment of Fig. 7 has preferably been implemented the processing described among Fig. 6.Determiner 704 is determined run length (n) and Golomb coefficient (m).Perhaps, counter or register 708 are used to each run length and amplitude sizes values to calculate accumulated value respectively and the corresponding number of times of this value to occur.Encoder 712 coding merchants (Q) are as the function of run length and Golomb coefficient.This encoder 712 function of remainder (R) of also encoding as run length, Golomb coefficient and merchant.In other embodiments, encoder 712 function of non-0 data of also encoding as the symbol of non-0 data value and this non-0 data value.Serial connection device 716 is used to be connected in series Q value and R value.

As an example, various illustrated logic diagram, flow chart and the step that embodiment disclosed herein is correlated with can realize or carry out with hardware or software, with application-specific integrated circuit (ASIC) (ASIC), programmable logic device, discrete gate or transistor logic, discrete hardware components, for example register and FIFO carry out the processor of one group of firmware instructions, any traditional programmable software and processor or their combination.Processor is microprocessor preferably, but also can be any conventional processors, controller, microcontroller or state machine.Software can be kept at RAM memory, flash memory, ROM memory, register, hard disk, moveable magnetic disc, CDROM, DVD-ROM or any other forms of medium of knowing in the industry.

Previously described preferred embodiment provides to those skilled in the art and finishes or use of the present invention.For those skilled in the art, be conspicuous for the multiple change of these embodiment, and defined here General Principle can be applicable to other embodiment and does not need creatively to work.So the present invention should not limit by the embodiment shown in here, and should meet the wideest scope of principle disclosed herein and creative feature.

Other features of the present invention and advantage illustrate in the claims.

Claims (9)

1. one kind is carried out apparatus for encoding to the data that comprise source images, and this device comprises:

The data that are used for the described source images of compression expression are also created the device of the compressed version of described source images thus, and the data that formerly generate by source images being carried out the adaptive block size adjust have been used in wherein said compression;

The device that diminishes version that is used to quantize the compressed version of described source images and creates described source images thus;

Be used to decompress the compressed version of described source images to create the device of the image through decompressing, the data that formerly generate by source images being carried out the adaptive block size adjust have been used in wherein said decompression;

The device that is used for determining the difference between described source images and the described image and creates the residual term data that are associated with described source images thus through decompressing; And

Be used to export the device that diminishes version and described residual term data of described source images, diminish version and the described residual term data of wherein said source images can be used to create and described source images uniform images in fact.

2. device as claimed in claim 1 is characterized in that, the described device that is used to compress be in frame or the basis of interframe on carry out.

3. device as claimed in claim 1 is characterized in that the described device that is used to compress has used the combination of discrete cosine transform and discrete quadtree conversion technology.

4. one kind is carried out Methods for Coding to comprising from the data of the multiple source frame of source images, and described method comprises:

The data of the first source frame of the described multiple source frame of compression expression, and create the compressed version of the described first source frame thus, the data that formerly generate by described multiple source frame being carried out the adaptive block size adjust have been used in wherein said compression;

The compressed version that quantizes the described first source frame is also created the version that diminishes of the described first source frame thus;

Decompress the compressed version of the described first source frame to create the frame through decompressing, and the data that formerly generate by described multiple source frame being carried out the adaptive block size adjust have been used in wherein said decompression;

Determine the difference between the second source frame and the described frame and create the residual term data that are associated with the described first source frame thus through decompressing; And

That exports the described first source frame diminishes version and described residual term data, and the version that diminishes of the wherein said first source frame can be used to create and the consistent in fact frame of the described first source frame with described residual term data.

5. method as claimed in claim 4 is characterized in that, carries out on the described basis that is compressed in interframe.

6. method as claimed in claim 4 is characterized in that the combination of discrete cosine transform and discrete quadtree conversion technology has been used in described compression.

7. one kind is carried out apparatus for encoding to comprising from the data of the multiple source frame of source images, and described device comprises:

Be used for the first source frame of the described multiple source frame of compression expression data, and create the device of the compressed version of the described first source frame thus, the data that formerly generate by described multiple source frame being carried out the adaptive block size adjust have been used in wherein said compression;

The device that diminishes version that is used to quantize the compressed version of the described first source frame and creates the described first source frame thus;

Be used to decompress the compressed version of the described first source frame to create the device of the frame through decompressing, the data that formerly generate by described multiple source frame being carried out the adaptive block size adjust have been used in wherein said decompression;

The device that is used for determining the difference between the second source frame and the described frame and creates the residual term data that are associated with the described first source frame thus through decompressing; And

Be used to export the device that diminishes version and described residual term data of the described first source frame, the version that diminishes of the wherein said first source frame can be used to create and the consistent in fact frame of the described first source frame with described residual term data.

8. device as claimed in claim 7 is characterized in that, the described device that is used to compress carries out on the basis of interframe.

9. device as claimed in claim 7 is characterized in that, the described device that is used to compress uses the combination of discrete cosine transform and discrete quadtree conversion technology.

Images