CN102082950A - Methods, devices and systems for compressing and decompressing images - Google Patents

Abstract

The invention discloses a method for compressing images, comprising the steps of: carrying out support vector regression treatment on an original image, and obtaining a support vector by taking the coordinate values of a pixel of the original image as an input sample and the gray value of the pixel as a monitoring signal; calculating a regression image according to the support vector; calculating a residual error image according to the original image and the regression image; and encoding the support vector and the residual error image. The invention also discloses a method for decompressing the images, corresponding to the method for compressing the images, and devices for compressing and decompressing the images. In the invention, a support vector technology is adopted in image compressing for predicating an image, so that parameters are obviously decreased and the compression rate is very high. Meanwhile, a vector supporting X-Y coordinates is used for expressing a medical image pixel value, which is more stable for compression treatment of medical images in comparison with a compression method carrying out predication based on image pixel gradient, therefore, the compression method provided by the invention is more stable.

Description

The compression of image, decompression method, Apparatus and system

Technical field

The present invention relates to image compression and decompression technique, particularly relate to a kind of compression, the method for decompression, Apparatus and system of the medical image based on machine learning.

Background technology

Nowadays, most of medical images obtain and store with digital form.The data volume of these images is very big, especially in radiology is used.Though the ability of the calculating of computer, storage and transmission constantly increases,, still press for the method for the high and stable lossless compress of compression ratio, especially aspect teleradiology and file for distributed image management.

Up to the present, there is several different methods to be considered with the problem that solves the medical image lossless compress, such as: based on contextual adaptability lossless image coding (CALIC, Context-based, Adaptive, Lossless Image Codec), and the JPEG-LS method.These two kinds of methods have description in many documents.

These two kinds of methods are compressed based on the gradient prediction of neighbor.For example, CALIC uses and is called as gradient adjustment prediction (GAP, Gredient-Adjusted Prediction) fallout predictor is done simple a, self adaptation, nonlinear prediction, and this method is used the prediction window of a size as 3*3, comprises 7 pixels on three image scanning lines in the window; JPEG-LS then utilizes the window of same size, comprises 4 pixels of preceding two image scanning lines in the window.These two kinds popular Lossless Image Compression methods, CALIC and JPEG-LS use very small amount of pixel in prediction window.

There is certain limitation in said method.Obviously, the pixel in the prediction window is less.If the pixel in the window increases, predicting the outcome can be better.Moreover the gradient predicted value of these two kinds of methods can not cover the grey scale pixel value of all medical images, such as the maximum prefetch measured value number of JPEG-LS is 364, and this just can't cover 12 bit depth (2 ¹²=4096, the gray value of possible pixel) medical image.In addition, if medical image comprises the complicated pixel (as chest x-ray sheet and many lung marking ray images) that can't predict with the gradient forecast method, generally can not obtain good result based on the compression of gradient prediction.

The practice that machine learning techniques is applied to compressed image is also arranged now.The some of them method is used for lossy compression method, such as disclosed method based on the image compression of learning in U.S. Pat 20090067491, because this method can't recover original image, is not therefore adopted by Medical Image Compression usually.Other methods are used for lossless compress, these methods are with machine learning model original image to be compressed, but, because they adopt the very complicated machine learning model that has comprised quantity of parameters, and the shared space of these parameters almost is similar to the size of original image, cause compression ratio very low, so the compression result of this method also is difficult to be accepted by Medical Image Compression.

Summary of the invention

The invention provides a kind of compression method of image, in Medical Image Compression, adopt the support vector regression technology, obtain preferable compression result, have higher compression ratio simultaneously.

The present invention also provides a kind of decompression method of image, and the medical image that adopts the support vector regression technology to compress is decompressed.

The present invention also provides a kind of compression set of image, adopts the support vector regression technology in Medical Image Compression, obtains preferable compression effectiveness, has higher compression ratio simultaneously.

The present invention also provides a kind of decompressing device of image, and the medical image that adopts the support vector regression technology to compress is decompressed.

Simultaneously, the present invention also provides a kind of compression and decompression system of image, comprises the compression set and the corresponding decompressing device that adopt the support vector regression technology.

A kind of compression method of image comprises: original image is carried out support vector regression handle, the coordinate figure of the pixel of described original image is the input sample, and the gray value of pixel is a supervisory signals, supported vector; Calculate the regression figure picture according to described support vector; According to described original image and recurrence image calculation residual image; To described support vector and described residual image coding.

Wherein, described input sample comprises the coordinate figure of all pixels of described original image.

Wherein, describedly calculate regression figure according to support vector and look like to comprise: as input, through the processing of SVMs, the output pixel gray value is to form the regression figure picture with each coordinate figure of regression figure picture.

Wherein, described according to original image with return the image calculation residual image and comprise: as, to generate initial residual image with described original image and described recurrence image subtraction; Described initial residual image is carried out add operation, to form non-negative residual image.

Wherein, described algorithm to support vector and residual image coding is: Huffmann coding, arithmetic coding, CALIC, JPEG-LS or JPEG2000.

A kind of decompression method of image is used for comprising decompressing through above-mentioned method processed images: to decoding supported vector sum residual image through the support vector and the residual image of coding; Calculate the regression figure picture according to described support vector; Described regression figure picture and residual image are merged, generate original image.

A kind of compression method of image comprises: original image is divided into a plurality of original image blocks; For all images block, carry out following steps respectively: image block is carried out support vector regression handle, the coordinate figure of the pixel of described image block is the input sample, and the gray value of pixel is a supervisory signals, obtains the block support vector; According to described block support vector calculation block regression figure picture; According to described original image block and block regression figure as the calculation block residual image; To described block support vector and block residual image coding.

Wherein, for each image block, described input sample comprises the coordinate figure of all pixels in this image block.

Wherein, describedly look like to comprise according to block support vector calculation block regression figure: as input, through the processing of block SVMs, the output pixel gray value is to form block regression figure picture with each coordinate figure of block regression figure picture.

Wherein, describedly comprise as the calculation block residual image: described original image block and described block are returned image subtraction, generate initial block residual image according to original image block and block regression figure; Described initial block residual image is carried out add operation, to form non-negative block residual image.

Wherein, described algorithm to block support vector and block residual image coding is: Huffmann coding, arithmetic coding, CALIC, JPEG-LS or JPEG2000.

A kind of decompression method of image, be used for decompressing through above-mentioned method processed images, comprise: for all images block, carry out following steps respectively:, obtain the support vector and the block residual image of image block to decoding through the support vector and the block residual image of image encoded block; Support vector calculation block regression figure picture according to described image block; Described block is returned trend and the merging of block residual image, generate the original image block; All original image blocks are merged, generate original image.

A kind of compression set of image comprises: support vector regression module 6011, and be used for that original image is carried out support vector regression and handle, the coordinate figure of the pixel of described original image is the input sample, the gray value of pixel is a supervisory signals, supported vector; Residual image generation module 6012 is used for calculating the regression figure picture according to described support vector, then according to described original image and recurrence image calculation residual image; Coding module 6013 is used for described support vector and described residual image coding.

Wherein, also comprise: memory module 701 is used to store support vector and residual image behind the coding of described coding module 6013 outputs.

A kind of decompressing device of image is used for above-mentioned compression set processed images is decompressed, and comprising: decoder module 6021 is used for decoding supported vector sum residual image through the support vector and the residual image of coding; Support vector modular converter 6022 is used for calculating the regression figure picture according to described support vector; Image merges module 6023, is used for the residual image of decoder module 6021 outputs and the regression figure of support vector modular converter 6022 outputs are looked like to merge, to generate original image.

A kind of compression and decompression system of image comprises: the compression set 601 of described image; Decompressing device 602 with described image.

A kind of compression set of image comprises: block is divided module 8014, is used for original image is divided into a plurality of original image blocks; Support vector regression module 8011 is used for that all images block is carried out support vector regression respectively and handles, and the coordinate figure of the pixel of described image block is the input sample, and the gray value of pixel is a supervisory signals, obtains the block support vector; Residual image generation module 8012 is used for according to described block support vector calculation block regression figure picture, then according to described original image block and block regression figure as the calculation block residual image; Coding module 8013 is used for described block support vector and block residual image coding.

Wherein, the compression and decompression system of described image also comprises: memory module 901 is used to store support vector and residual image behind the coding of described coding module 8013 outputs.

A kind of decompressing device of image, be used for above-mentioned compression set processed images is decompressed, comprise: decoder module 8021 is used for obtaining block support vector and block residual image to decoding through the block support vector and the block residual image of coding; Support vector modular converter 8022, the block support vector that is used for that decoding is obtained is converted to block regression figure picture; Image merges module 8023, and the block regression figure picture that is used for being converted to combines with the block residual image, obtains the original image block; Piece merges module 8024, is used for after receiving all images block of original image all images block being merged, to generate original image.

A kind of compression and decompression system of image comprises: the compression set 801 of described image; Decompressing device 802 with described image.

As can be seen from the above technical solutions, the present invention is in the compression process of medical image, adopt the support vector technology that image is predicted, the sample of input is not limited to tradition based on the pixel in the window in the compression method of image pixel gradient, even can adopt whole image pixels, it is more accurate to predict the outcome, but parameter significantly reduces, and compression ratio is very high.Simultaneously, compress technique of the present invention is used and is supported the vector of X and Y coordinates to represent the medical image pixel value, and X and Y coordinates are more stable than the image pixel gradient, therefore compress technique of the present invention makes compression method of the present invention also more stable for more more stable than traditional compression method based on the prediction of image pixel gradient in the processing of medical image.

Description of drawings

To make clearer above-mentioned and other feature and advantage of the present invention of those of ordinary skill in the art by describing the preferred embodiments of the present invention in detail with reference to accompanying drawing below, identical label is represented identical parts, in the accompanying drawing:

Fig. 1 is the method flow diagram that image is carried out compression and decompression according to the embodiment of the invention one;

Fig. 2 has represented to set up the schematic diagram of supporting vector machine model;

Fig. 3 shows original image (a), regression figure picture (b) and residual image (c);

Fig. 4 is the method flow diagram that image is carried out compression and decompression according to the embodiment of the invention two;

Fig. 5 is the schematic diagram of in the embodiment of the invention two image division being handled;

Fig. 6 is the structural representation according to the compression and decompression system of the embodiment of the invention three;

Fig. 7 is the structural representation according to the compression and decompression system of the embodiment of the invention four;

Fig. 8 is the structural representation according to the compression and decompression system of the embodiment of the invention five;

Fig. 9 is the structural representation according to the compression and decompression system of the embodiment of the invention six.

Embodiment

In order to make technical scheme of the present invention and advantage clearer,, the present invention is further elaborated below in conjunction with drawings and Examples.Should be appreciated that specific embodiment described herein only in order to explanation the present invention, and be not used in qualification the present invention.

In the face of prior art problems the time, the inventor is not confined to existing method, merely increases more pixel to obtain more excellent compression effectiveness in prediction window, perhaps the prior art of machine learning is improved, such as reducing learning parameter or the like, to improve compression efficiency.The inventor has proposed the method and system of the lossless compress of novelty after having studied at the possible solution based on the harmless Medical Image Compression of machine learning.In this new method and system, machine learning and pattern classification technology are combined, be used for " prediction " (study) of image, when prediction, adopt more pixel (even entire image), utilize the method for support vector regression (SVR) to learn and predicted picture, then residual image and support vector are encoded.

The gray scale medical image can be expressed as { x _n, y _n, i _n, n=1 ... N, wherein x and y represent X and Y coordinates respectively, i is corresponding pixel intensity value, i.e. gray scale, N is the pixel count of image.The function representation that coordinate is relevant with pixel grey scale is:

f(x，y)→i. (1)

From equation (1) as can be seen, there is correlation between coordinate and the gray scale.If with equation (1) study is a simple high-accuracy formula, we can only preserve coordinate, are used to predict gray scale.Like this, can greatly reduce image, because coordinate is regularly arranged with what fix very much.Yet the curved surface of most medical images is very complicated and irregular.Therefore, in most of the cases, the correlation of equation (1) is non-linear, and can not be expressed as one or several conspicuous formula.The present invention utilizes the technology-support vector regression of improvement to solve this problem.Recurrence only is used for " prediction ", and is not to be used for accurate Calculation.

SVMs (SVM) is a kind of machine learning method based on Statistical Learning Theory.SVMs is used for the input sample is classified, for nonlinear situation, by nonlinear transformation the input space is transformed to a high-dimensional feature space with the inner product function definition, classify at high-dimensional feature space, search out the optimal classification face, the optimal classification face is the face of the class interval maximum that makes two class samples.Support vector is the closer vector of sample middle distance optimal classification face, generally is the vector that is in the edge of class interval.Utilize support vector to carry out support vector regression, image is carried out " prediction ".Usually the support vector in the high-dimensional feature space is corresponding to the universal characteristics of image.SVMs output is the linear combination of intermediate node, the inner product of a corresponding input sample of each intermediate node and a support vector.

Below, in conjunction with the accompanying drawings the specific embodiment of the present invention is elaborated.

Fig. 1 is the method flow diagram that image is carried out compression and decompression according to the embodiment of the invention one.As can be seen from Figure 1, the method for embodiment one mainly comprises the steps:

Step 101 is carried out support vector regression to original image and is handled, and obtains support vector.Be specially, X, the Y coordinate set up with image are the supporting vector machine model of input, use the supporting vector machine model of being set up to come the presentation video interior pixels regularity of distribution, promptly go to explain relation between coordinate figure and the pixel distribution with this model.The input sample is with (in addition, the pairing grey scale pixel value of input sample for supervisory signals (also claiming teacher signal) input, is used for exporting the desired value that need approach in learning process for x, y) expression.

Sample as study can be a partial pixel in the image herein, also the sample of whole pixels as study can be decided on required concrete condition.

Fig. 2 has represented to set up the schematic diagram of supporting vector machine model.As can be seen from Figure 2, and generator G generation input sample (x, y).Training aids S will import sample (x, y) corresponding with output valve (signal of promptly supervising and guiding) i.Learning machine LM to from input sample-gray value of generator G and training aids S to learning, select optimum function to reflect input (coordinate figure) and export relation between (gray value).After study, LM is supporting vector machine model.

Process through support vector regression is handled has produced support vector.

Step 102 is calculated residual image according to support vector.

Support vector is used for " prediction " to image, can obtain the image of " prediction ", i.e. regression figure picture according to support vector.Then, calculate residual image according to regression figure picture and original image.Residual image has reflected poor between original image and the predicted picture.

The method of calculating the regression figure picture according to support vector be coordinate figure with the regression figure picture as input, through the processing of SVMs, the output pixel gray value forms the regression figure picture.The coordinate figure of described regression figure picture promptly is the coordinate figure of the respective pixel that the process support vector regression is handled in the original image.

We can utilize regression figure picture and original image to calculate initial residual image:

$I_{\mathrm{residual}}^0=I-I_p---\left(2\right)$

Wherein I is an original image, shown in Fig. 3 (a).Ip represents the regression figure picture, shown in Fig. 3 (b).

Be initial residual image, shown in Fig. 3 (c).Clearly, in the initial residual image in the equation (2) element have on the occasion of and negative value, should represent by the numerical value of INT type that has the negative information value or SHORT type.Can adjust this initial residual plot with the ADD operation is a non-negative residual image I _ResidualProcess is as follows:

$I_{\mathrm{residual}}=I_{\mathrm{residual}}^0+\left|i_{\min }\right|---\left(3\right)$

Wherein

$i_{\min }=\underset{n=l,...,N}{\min }{I}_{\mathrm{residual}}^0---\left(4\right)$

Usually, the value that comprises in the residual image is smaller.That is to say span _rValue smaller:

${\mathrm{span}}_r=\underset{n=1,...,N}{\max }{I}_{\mathrm{residual}}---\left(5\right)$

For general medical image, span _rValue normally less than 32.

Step 103 is encoded to support vector and residual image.

Can adopt conventional Image Compression that residual image is encoded, for example Huffmann coding, arithmetic coding, CALIC, JPEG-LS or JPEG2000.

Support vector and residual image are carried out after the encoding process, and image compression process is promptly finished.

Step 104, through the coding support vector and residual image can temporarily be stored in the memory, or via Network Transmission to decoding end.

Step 105, to decoding through the support vector and the residual image of coding, the result of decoding processing is supported vector sum residual image.

Step 106, the support vector that decoding is obtained is converted to the regression figure picture.

The method of calculating the regression figure picture according to support vector be coordinate figure with the regression figure picture as input, through the processing of SVMs, the output pixel gray value forms the regression figure picture.

Step 107 combines the regression figure picture that is converted to residual image, obtain original image.

So far, complete image compression and decompression process are promptly finished.

Because the pixel value of medical image is many usually, the speed of study can be slow when therefore support vector regression was handled, in order to address this problem, the inventor optimizes aforesaid embodiment, be about to the original image piecemeal, block carries out the compression and decompression processing one by one, then all is merged through the image block that decompress, thereby obtains original image.Describe below in conjunction with accompanying drawing.

Fig. 4 is the method flow diagram that image is carried out compression and decompression of the embodiment of the invention two.As can be seen from Figure 4, the method for embodiment two mainly comprises the steps:

Step 401 is carried out block to original image and is divided.

Fig. 5 is the schematic diagram that image division is handled.As can be seen from Figure 5, a view picture original image is divided into the L*M piece, and wherein L is maximum line number, and M is maximum columns.The size of image block is not made as fixed value, adjusts the block size because in general need according to the intensity and the content of original image.

Then,, carry out the process of embodiment one described compression and decompression respectively for each block, promptly carry out with step 101 to 107 essentially identical steps 402 to 408, obtain each original image block.

Step 402 is carried out support vector regression to the original image block and is handled, and has produced the block support vector.

Step 403 is according to support vector calculation block residual image.

Step 404 is encoded to block support vector and block residual image.Block support vector and block residual image are carried out after the encoding process, and the image block compression process is promptly finished.

Step 405, through the coding block support vector and block residual image can temporarily be stored in the memory, or via Network Transmission to decoding end.

Step 406, to decoding through the block support vector and the block residual image of coding, the result of decoding processing obtains block support vector and block residual image.

Step 407, the block support vector that decoding is obtained is converted to block regression figure picture.

Step 408 combines the block regression figure picture that is converted to the block residual image, obtain the original image block.So far, image block compression and decompression process is promptly finished.

After finishing for the compression and decompression process of all original image blocks, execution in step 409 merges all images block, obtains original image.

Embodiment two is better than embodiment one part and mainly is the image division block, makes that the learning time in the support vector regression processing procedure reduces, thereby has further improved the efficient of compression and decompression of the present invention.

The present invention is not limited to the method for compression and decompression, also provides and the corresponding system of compression and decompression method, that is to say, carries out the system of compression and decompression method of the present invention.Be elaborated below in conjunction with accompanying drawing.

Fig. 6 is the structural representation of the compression and decompression system of the embodiment of the invention three.As can be seen from Figure 6, on the whole, the compression and decompression system of embodiment three comprises compression set 601 and decompressing device 602.

Compression set 601 comprises support vector regression module 6011, residual image generation module 6012 and coding module 6013.

6011 pairs of original images of support vector regression module carry out support vector regression to be handled, and has produced support vector, and concrete processing sees step 101 for details.

Residual image generation module 6012 generates residual image according to support vector, and concrete processing sees step 102 for details.

6013 pairs of support vectors of coding module and residual image are encoded, and generate support vector and residual image through coding, and concrete processing sees step 103 for details.

Decompressing device 602 comprises that decoder module 6021, support vector modular converter 6022 and image merge module 6023.

6021 pairs of support vector and residual images through coding of decoder module are decoded.

Support vector modular converter 6022 is converted to the regression figure picture with support vector.

Image merges module 6023 residual image of decoder module 6021 outputs and the regression figure of support vector modular converter 6022 outputs is looked like to merge, and obtains original image.

Fig. 7 is the structural representation of the compression and decompression system of the embodiment of the invention four.Embodiment four only is that with the difference of embodiment three embodiment four has increased a memory module 701 on the basis of embodiment three.The support vector and the residual image through coding of 601 outputs of memory module 701 store compressed devices, and after obtaining transfer instruction, will be sent to decompressing device 602 through the support vector and the residual image of coding.Compression set 601 is identical with embodiment three with each functions of modules of decompressing device 602 inside, does not repeat them here.

Fig. 8 is the structural representation of the compression and decompression system of the embodiment of the invention five.As can be seen from Figure 8, the compression set 801 of embodiment five has also increased a block and has divided module 8014 except comprising support vector regression module 8011, residual image generation module 8012 and coding module 8013.Block is divided module 8014 original image is divided into into some original image blocks.Each original image block all passes through the processing of support vector regression module 8011, residual image generation module 8012 and coding module 8013.

8011 pairs of original image blocks of support vector regression module carry out support vector regression to be handled, and has produced the block support vector.

Residual image generation module 8012 is according to support vector calculation block residual image.

8013 pairs of block support vectors of coding module and block residual image are encoded.

As can be seen from Figure 8, the decompressing device 802 of embodiment five comprises that also piece merges module 8024 except comprising decoder module 8021, support vector modular converter 8022 and image merging module 8023.

8021 pairs of block support vector and block residual images through coding of decoder module are decoded, and obtain block support vector and block residual image.

The block support vector that support vector modular converter 8022 obtains decoding is converted to block regression figure picture.

Image merges module 8023 the block regression figure picture that is converted to is combined with the block residual image, obtains the original image block.

Piece merges module 8024 after receiving all images block of original image, and all images block is merged, and obtains original image.

Fig. 9 is the structural representation of the compression and decompression system of the embodiment of the invention six.Embodiment six only is that with the difference of embodiment five embodiment six has increased a memory module 901 on the basis of embodiment five.The block support vector and the block residual image through coding of 801 outputs of memory module 901 store compressed devices, and after obtaining transfer instruction, will be sent to decompressing device 802 through the block support vector and the block residual image of coding.Compression set 801 is identical with embodiment five with each functions of modules of decompressing device 802 inside, does not repeat them here.

New method and system of the present invention, the machine learning techniques of application of advanced-support vector regression technology in Medical Image Compression.Utilize more pixel (or even pixel of entire image) information to come " prediction ", it is more accurate to predict the outcome, and parameter significantly reduces, and compression ratio is very high.The compression ratio of this system approaches CALC and JPEG-LS usually, compression ratio even higher when handling complicated medical image or medical image block.

Another advantage of the present invention is this compression method and system for more more stable than traditional compression method based on the prediction of image pixel gradient in the processing of medical image, and this is to support the vector of X and Y coordinates to represent the medical image pixel value because use.X and Y coordinates are more stable than the image pixel gradient.

The above only is preferred embodiment of the present invention, not in order to restriction the present invention, all any modifications of being done within the spirit and principles in the present invention, is equal to and replaces and improvement etc., all should be included within protection scope of the present invention.

Claims (20)

1. the compression method of an image is characterized in that, comprising:

Original image is carried out support vector regression handle, the coordinate figure of the pixel of described original image is the input sample, and the gray value of pixel is a supervisory signals, supported vector;

Calculate the regression figure picture according to described support vector;

According to described original image and recurrence image calculation residual image;

To described support vector and described residual image coding.

2. method according to claim 1 is characterized in that described input sample comprises the coordinate figure of all pixels of described original image.

3. method according to claim 1 is characterized in that, describedly calculates regression figure according to support vector and looks like to comprise:

As input, through the processing of SVMs, the output pixel gray value is to form the regression figure picture with each coordinate figure of regression figure picture.

4. method according to claim 1 is characterized in that, and is described according to original image with return the image calculation residual image and comprise:

With described original image and described recurrence image subtraction, generate initial residual image;

Described initial residual image is carried out add operation, to form non-negative residual image.

5. method according to claim 1 is characterized in that, described algorithm to support vector and residual image coding is:

Huffmann coding, arithmetic coding, CALIC, JPEG-LS or JPEG2000.

6. the decompression method of an image is used for it is characterized in that decompressing through the described method processed images of claim 1, comprising:

To decoding supported vector sum residual image through the support vector and the residual image of coding;

Calculate the regression figure picture according to described support vector;

Described regression figure picture and residual image are merged, generate original image.

7. the compression method of an image is characterized in that, comprising:

Original image is divided into a plurality of original image blocks;

For all images block, carry out following steps respectively:

Image block is carried out support vector regression handle, the coordinate figure of the pixel of described image block is the input sample, and the gray value of pixel is a supervisory signals, obtains the block support vector;

According to described block support vector calculation block regression figure picture;

According to described original image block and block regression figure as the calculation block residual image;

To described block support vector and block residual image coding.

8. method according to claim 7 is characterized in that, for each image block, described input sample comprises the coordinate figure of all pixels in this image block.

9. method according to claim 7 is characterized in that, describedly looks like to comprise according to block support vector calculation block regression figure:

As input, through the processing of block SVMs, the output pixel gray value is to form block regression figure picture with each coordinate figure of block regression figure picture.

10. method according to claim 7 is characterized in that, describedly comprises as the calculation block residual image according to original image block and block regression figure:

Described original image block and described block are returned image subtraction, generate initial block residual image;

Described initial block residual image is carried out add operation, to form non-negative block residual image.

11. method according to claim 7 is characterized in that, described algorithm to block support vector and block residual image coding is:

Huffmann coding, arithmetic coding, CALIC, JPEG-LS or JPEG2000.

12. the decompression method of an image is used for it is characterized in that decompressing through the described method processed images of claim 6, comprising:

For all images block, carry out following steps respectively:

To decoding, obtain the support vector and the block residual image of image block through the support vector and the block residual image of image encoded block;

Support vector calculation block regression figure picture according to described image block;

Described block is returned trend and the merging of block residual image, generate the original image block;

All original image blocks are merged, generate original image.

13. the compression set of an image is characterized in that, comprising:

Support vector regression module (6011) is used for that original image is carried out support vector regression and handles, and the coordinate figure of the pixel of described original image is the input sample, and the gray value of pixel is a supervisory signals, supported vector;

Residual image generation module (6012) is used for calculating the regression figure picture according to described support vector, then according to described original image and recurrence image calculation residual image;

Coding module (6013) is used for described support vector and described residual image coding.

14. device according to claim 13 is characterized in that, also comprises:

Memory module (701) is used to store support vector and residual image behind the coding of described coding module (6013) output.

15. the decompressing device of an image is used for the described compression set processed images of claim 13 is decompressed, and it is characterized in that, comprising:

Decoder module (6021) is used for decoding supported vector sum residual image through the support vector and the residual image of coding;

Support vector modular converter (6022) is used for calculating the regression figure picture according to described support vector;

Image merges module (6023), is used for the residual image of decoder module (6021) output and the regression figure of support vector modular converter (6022) output are looked like to merge, to generate original image.

16. the compression and decompression system of an image is characterized in that, comprising:

The compression set of image as claimed in claim 11 (601); With

The decompressing device of image as claimed in claim 13 (602).

17. the compression set of an image is characterized in that, comprising:

Block is divided module (8014), is used for original image is divided into a plurality of original image blocks;

Support vector regression module (8011) is used for that all images block is carried out support vector regression respectively and handles, and the coordinate figure of the pixel of described image block is the input sample, and the gray value of pixel is a supervisory signals, obtains the block support vector;

Residual image generation module (8012) is used for according to described block support vector calculation block regression figure picture, then according to described original image block and block regression figure as the calculation block residual image;

Coding module (8013) is used for described block support vector and block residual image coding.

18. device according to claim 17 is characterized in that, also comprises:

Memory module (901) is used to store support vector and residual image behind the coding of described coding module (8013) output.

19. the decompressing device of an image is used for the described compression set processed images of claim 17 is decompressed, and it is characterized in that, comprising:

Decoder module (8021) is used for obtaining block support vector and block residual image to decoding through the block support vector and the block residual image of coding;

Support vector modular converter (8022), the block support vector that is used for that decoding is obtained is converted to block regression figure picture;

Image merges module (8023), and the block regression figure picture that is used for being converted to combines with the block residual image, obtains the original image block;

Piece merges module (8024), is used for after receiving all images block of original image all images block being merged, to generate original image.

20. the compression and decompression system of an image is characterized in that, comprising:

The compression set of image as claimed in claim 17 (801); With

The decompressing device of image as claimed in claim 19 (802).

Images