CN103761753A - Decompression method based on texture image similarity - Google Patents

Abstract

The invention discloses a decompression method based on texture image similarities. The decompression method based on the texture image similarities comprises the steps that a texture image to be compressed is transformed into a plurality of tiles in a YCrCb color space, a code book and a code table of the texture image to be compressed are determined according to the similarities among the different tiles, codewords in the code book are coded by adopting coding similar to Huffman coding, the code table is transformed into an index file according to coding results, then the compression of the texture image is completed, the compressing process is further reversed, and the decompression is carried out on the compressed texture image according to the code book and the index file. According to the decompression method based on the texture image similarities, interlacing sampling is carried out on chrominance information, the effect of the reconstruction image is not influenced, and the compression ratio of the texture image is improved to a certain extent. Furthermore, when the compression code book of the texture image is established automatically, both the integral similarities of the texture tiles and the local similarities of texture are taken into consideration and correspond to a thread of a GPU processor, parallel decompression of data is achieved, the computing capability of the GPU is fully used, and real-time decompression is achieved.

Description

Decompression method based on texture image similarity

Technical field

The present invention relates to computer picture field, relate in particular to a kind of decompression method based on texture image similarity.

Background technology

Thereby compression of images object is the redundant information reducing in view data stores and data transmission with realizing efficient data.At present, compression of images is mainly divided into lossy compression method and Lossless Compression, and popular coding standard is just like JPEG, MPEG etc.

In large-scale scene rendering, need texture image to call in video memory, this process can not disposablely complete rapidly, so we need to study compression and the decompression algorithm of texture image, to reduce the storage space of texture image, reduce transmission delay, reduce the loss of data texturing simultaneously.The compression that realizes texture image be take traditional images compression method as basis, so both have certain difference because texture image itself presents very high similarity.

Texture compression has had a lot of solutions at present, and Main Means is the code book that produces image, then according to code book, realizes the reconstruct of image.Recent years, effect is reasonable increment type code book generating algorithm, and this algorithm can dynamically increase code book, and it is simple to generate step, and compression speed is also very fast.But generate code book by random mode in this algorithm, the code word that probably causes some to have more representative is left out, and can cause like this increase of number of codewords, reduce the ratio of compression of texture image, the visual effect of the image that reconstruct obtains also can reduce.In order to improve the quality of ratio of compression and the reconstructed image of image, another kind of popular algorithm is the Self-organizing Maps algorithm based on neural network, this algorithm is to carry out unsupervised study by double-layer structure network, final effect is relatively good, but this algorithm need to be through iteration many times, calculated amount is very large, and compression process is very slow.

Very early, the decompression algorithm of texture image is only used CPU to carry out decompress(ion), delivers to main memory in external memory after compressing again, and the effect that this decompression algorithm is drawn real-time decompress(ion) is poor especially.At present, along with the maturation of the graphic hardwares such as GPU, a lot of algorithms, all in conjunction with CPU and GPU, promote the effect of real-time decompress(ion) by scheduling between the two.Wherein, CPU and GPU data communication meeting between the two reduces the speed of algorithm, and this mode also exists certain drawback.

Summary of the invention

The present invention is directed to the deficiencies in the prior art, provide a kind of based on texture image similar solution compression method, this decompression method can access good texture image ratio of compression, calculated amount is little, the picture quality that reconstruct simultaneously obtains is better, this algorithm takes full advantage of the computation capability of GPU in addition, greatly improves the speed of understanding embossing reason.

A decompression method based on texture image similarity, comprises compression process and decompression procedure,

Described compression process comprises:

(1) texture image to be compressed is transformed into YC from rgb color space _rc _bcolor space obtains YC _rc _bimage, and by YC _rc _bimage is cut into several tiles; (2) utilize tile similarity, generate code book and the code table of image to be compressed, described code book comprises several code words, and described code table is for recording the numbering of the code word that each tile is corresponding;

(3) according to code book and code table, all code words in code book are carried out to class Huffman encoding and obtain codeword coding, described codeword coding comprises identification burst and code field, and generating gauge outfit information, described gauge outfit information comprises the size of texture image to be compressed and the length of identification burst;

(4) by the numbering of corresponding codewords in codeword coding replacement code table, obtain code table coding, gauge outfit information is added to the head of code table coding, form the index file of texture image to be compressed;

Described decompression procedure comprises:

(S1) scanning index file, according to the length of the size of the texture image recording in gauge outfit information and identification burst, obtain the size of texture image, and the codeword coding corresponding with each position in this texture image, according to the codeword coding inquiry code book obtaining, obtain the tile of each position, and by GPU, all tiles that obtain are reconstructed, obtain YC _rc _bimage;

(S2) YC to reconstruct _rc _bthe colourity of image is carried out the operation of interlacing bilinear interpolation, and is transformed into rgb color space and obtains the texture image after decompress(ion).

Decompression method of the present invention is applicable to the decompression of texture image.YC in the present invention _rc _bit is mutually not overlapped between each tile when image is cut into several tiles.Code word described in the present invention is actually tile images.

Code word is interpreted as the tile that can represent this texture image in texture image, and a texture image is generally to there being a plurality of code words, and the set of all code words is code book.

The present invention is first transformed into YCrCb color space texture image to be compressed from rgb color space, and chrominance information is carried out to partiting row sampling, and but image is cut into formed objects not overlapped texture tile.This decompression method is based on AOP algorithm (Automatic Organizing Process, from rebuilding) automatically build the compressed codebook of texture image, according to the similarity between different tiles, determine code book and the code table of texture image to be compressed, then adopt class Huffman encoding to encode and obtain the codeword coding of all code words the code word in this code book, and code table is converted into index file, and then completing texture image compression, compression result is code book and index file.And one step compression process is converse, the texture image according to code book and index file after to compression decompresses.

The present invention has designed the AOP method of processing structure based on generating, this algorithm calculated amount is little, simple to operate easy to implement, in the visual effect that guarantees reconstructed image, improve ratio of compression simultaneously, and adopt class Huffman encoding to obtain each code word to encode, the code word size obtaining is different, in code table, occur that code word degree corresponding to maximum codeword number is the shortest, increase successively, so just greatly reduce the storage space of index file, improve the ratio of compression of texture image, be specially adapted to the data dispatch of the scheduling of distributed network and internal memory.Meanwhile, decompress(ion) is directly carried out in this invention on GPU, makes full use of the computation capability of GPU, thereby improve decompress(ion) speed, realizes real-time decompress(ion).

Described tile size is 2 * 2～6 * 6 pixels.Tile is generally square, if texture image to be compressed can not be divided into while setting big or small tile just, finally at the edge of texture image, may exist pixel to be less than the tile of setting value.If it is too large that tile is set, the code book finally obtaining can be larger, and the ratio of compression of image is lower.If but set too little, the computational data amount of algorithm is large than very, the picture quality that while decompress(ion) obtains is also poor.

Described step (1) is by YC _rc _bimage is cut into before several tiles also to described YC _rc _bthe chrominance information of image is carried out partiting row sampling.

YC _rc _by passage, C that image comprises _rpassage and C _bpassage, Y passage represents brightness value, C _rpassage and C _bpassage is chromatic value.Affect a lot of because have of picture quality, as the contrast of image, monochrome information, color information etc., but for human eye, than more sensitive be brightness because human eye is more responsive to monochrome information, not too responsive to chrominance information comparatively speaking.Therefore chrominance information Cr and two passages of Cb are carried out to partiting row sampling, preserve the chrominance information of half, on the basis that does not affect visual effect, reduced like this size of image, further improve the ratio of compression of texture image.

In described step (2), the generation code book of image to be compressed and the detailed process of code table are as follows:

(2-1) null set of initialization is as code book, and array of initialization is as code table, and the size of array is identical with the sum of tile;

(2-2) choose arbitrarily a tile as initial tile, using initial tile as code word, and this code word is numbered, then the numbering of this code word is write to position corresponding with this tile in code table;

(2-3) for other any one tiles, the phase knowledge and magnanimity that calculate respectively each code word in current tile and code table and each adjacent tile are poor, the tile of the poor minimum of phase knowledge and magnanimity (is comprised to code word, code word is in fact also a tile) as the most similar tile of current tile, and make the following judgment:

If the minimum poor threshold value that is less than setting of phase knowledge and magnanimity, and this most similar tile is adjacent tile,, using this most similar tile as code word, this code word be numbered and this code word is write to code book, then the numbering of this code word being write to position corresponding with current tile in code table;

If the minimum poor threshold value that is less than setting of phase knowledge and magnanimity, and this most similar tile is the code word in code book, and numbering corresponding to this code word write to position corresponding with current tile in code table;

Otherwise, using current tile as code word, this code word is numbered and this code word is write to code book, then the numbering of this code word is write to position corresponding with current tile in code table.

If while there is the tile of the poor minimum of a plurality of phase knowledge and magnanimity, consider that these minimum phase knowledge and magnanimity are poor whether to there being code word, if there is code word, make this code word for the most similar tile, if there is no (being that the poor minimum of phase knowledge and magnanimity is all corresponding to adjacent tile), from tile corresponding to these minimum phase knowledge and magnanimity random one get as the most similar tile.

Initialization code book for null set, travel through each tile and obtain gradually new code word, then write code book, thereby travel through all code words that all tiles obtain this texture image, obtained the code book completing.The code table of initialization code table is an array, the size of array is identical with the sum of tile, in its array, the position of each element and the position of tile are to corresponding, when similarity based on tile obtains code book also to again to all elements assignment in this array, in the present invention, by the coding of code word corresponding to each tile again assignment, be corresponding element, obtain for representing the code table of the corresponding relation of tile and code word.

Described adjacent tile is YC _rc _bin image centered by current tile, the tile in the region of some 1～3 tiles of extension.

The number of the code word comprising in code book determines by the similarity between all tiles, and the element number that code table comprises is consistent with the number of tile, and each element represents respectively the numbering (being codeword number) of the code word that corresponding tile is corresponding.Successively all tiles are traveled through to code book and the code table that can obtain this texture image, determine that the code word in code book is general by two-layer judgement, ground floor is used the code word of increment type to generate the similarity of the code word in i.e. more current tile and code book, the similarity of the more current tile of the second layer and adjacent tile, the local similarity of the texture of also considering by the two-layer global similarity of more not only having considered texture.

It is poor that described step (2-3) is calculated phase knowledge and magnanimity according to phase knowledge and magnanimity function, comprises the following steps:

(a) according to formula:

$M_{\mathrm{ij}}^k=a\×{(Y_i^k-Y_j^k)}^2+(1-a){(C_{\mathrm{ri}}^k-C_{\mathrm{rj}}^k)}^2+(1-a)\×{(C_{\mathrm{bi}}^k-C_{\mathrm{bj}}^k)}^2$

Calculate, wherein M _ijrepresent that k the pixel phase knowledge and magnanimity of k pixel and j tile in i tile are poor; A represents the coefficient of brightness adjusting information, (Y _i, C _ri, C _bi) and (Y _j, C _rj, C _bj) represent respectively the tile information of i and j tile, Y _i,

represent that respectively i tile is at YC _rc _bthe Y passage of color space, C _rpassage and C _bthe value of passage, Y _j,

represent that respectively j tile is at YC _rc _bthe Y passage of color space, C _rpassage and C _bthe value of passage, k=0,1,2 ..., N, N is the pixel sum in tile;

(b) it is poor that the value of traversal k obtains in i tile in each pixel and j tile the phase knowledge and magnanimity of respective pixel point, and averaged, obtains i tile poor with the phase knowledge and magnanimity of j tile.

The threshold value of setting in described step (2-3) is 200-400.If Threshold is too little, finally the code book obtaining after compression is very large, and ratio of compression is lower.If Threshold is too large, although can improve the ratio of compression of image, the mass ratio of the image that after last decompress(ion), reconstruct obtains is poor.

In described step (3), class Huffman encoding comprises the following steps:

(3-1) occurrence number of each codeword number in the code word number of statistics in code book and code table, and according to the occurrence number of the numbering of each code word from how to sort to few;

(3-2) ask for and satisfy condition:

2 ¹+2 ²+…+2 ⁿ≥M，

The value of n minimum positive integer, and using this value as the length of identification burst, wherein, M is the code word number in code book;

(3-3) according to the sequence of the numbering occurrence number of code word, set the length of code field, occurrence number is more, and the length of code field is larger;

(3-4) according to the length of identification burst and the length of each code field set, each code word is carried out to the codeword coding that binary coding obtains each code word.

In the present invention, codeword coding is divided into identification burst and two parts of code field, identification burst is front, length for the code field that represents, in code book, the number of code word is more, the length of identification burst is longer, code field corresponding to occurrence number is more in seasonal code table the codeword number of encoding is shorter, and the overall length of codeword coding is shorter, can obviously improve like this ratio of compression of texture image.

Described decompression process also comprises carries out the nl-means noise reduction process based on GPU to described texture image.Adopt AOP algorithm to carry out texture image compression, the texture image that the reconstruct that decompresses obtains tends to occur mosaic effect.By nl-means noise reduction process algorithm, and in conjunction with GPU computing power at a high speed, can substantially obtain in real time the reasonable reconstructed image of visual effect.The nl-means noise reduction process algorithm document that sees reference wherein: Roimela K, Aarnio T, j.High dynamic range texture compression, ACM Transactions on Graphics (TOG) .ACM, 2006,25 (3): 707-712.

Compared with prior art, beneficial effect of the present invention is:

(1) texture image to be compressed is transformed into YC from rgb color space _rc _bcolor space obtains YC _rc _bimage, carries out partiting row sampling to Cr and two passages of Cb, afterwards at YC after partiting row sampling _rc _bimage graph looks like to carry out texture image compression, because human eye is insensitive to colourity, Cr and two passages of Cb is carried out to partiting row sampling, neither affects the effect of reconstructed image, has improved to a certain extent the ratio of compression of texture image;

(2) adopt class Huffman encoding to carry out binary coding to code word, occurrence number is more, and corresponding codeword coding is shorter, and therefore, for whole texture image to be compressed, the index file finally obtaining is less, has greatly improved ratio of compression;

(3) the AOP algorithm of the present invention's design can build the compressed codebook of texture image automatically, by bilayer, process structure, by once traveling through and just can obtain code book, computing cost is also smaller, and double-decker has not only been considered the local similarity of the texture that the global similarity of texture tile is also considered simultaneously;

(4) decompression method of the present invention is each data element is corresponded on the thread of GPU processor, simultaneously by dispatch deal, has realized the parallel decompress(ion) of data.This decompression algorithm makes full use of GPU computation capability, can realize real-time decompress(ion).

Embodiment

Below in conjunction with embodiment, the present invention is described in detail.

A decompression method based on texture image similarity, comprises compression process and decompression procedure, and compression process comprises:

(1) texture image to be compressed is transformed into YC from rgb color space _rc _bcolor space obtains YC _rc _bimage, by YC _rc _bimage is cut into before several tiles also to described YC _rc _bthe chrominance information of image is carried out partiting row sampling, then and by YC _rc _bimage is cut into several tiles (in the present embodiment, tile size is 4 * 4 pixels, cuts apart and obtains m tile),

Wherein according to formula:

$\left[\begin{array}{c}Y\\ C_r\\ C_b\end{array}\right]=\left[\begin{array}{ccc}0.299& 0.578& 0.114\\ 0.500& -0.4187& -0.0813\\ -0.1687& -0.3313& 0.500\end{array}\right]\×\left[\begin{array}{c}R\\ G\\ B\end{array}\right]+\left[\begin{array}{c}0\\ 128\\ 128\end{array}\right]$

Texture image to be compressed is transformed into YCrCb color space from rgb color space, and wherein R, G and B are respectively a pixel in texture image to be compressed at the value of R passage, G passage and B passage, Y, C _rand C _bthis pixel is at YC respectively _rc _bthe Y passage of color space, C _rpassage and C _bthe value of passage.All pixels in texture image to be compressed are transformed into the conversion that YCrCb color space can complete whole texture maps image to be compressed from rgb color space.

(2) utilize tile similarity, generate code book and the code table of image to be compressed, described code book comprises several code words, and described code table is for recording the numbering of the code word that each tile is corresponding, and detailed process is as follows:

(2-1) null set of initialization is as code book, and array of initialization is as code table, and the size of array is identical with the sum of tile;

(2-2) choose arbitrarily a tile as initial tile, using initial tile as code word, and this code word is numbered, then the numbering of this code word is write to position corresponding with this tile in code table;

(2-3), for other any one tiles, the phase knowledge and magnanimity that calculate respectively each code word in current tile and code table and each adjacent tile are poor, the most similar tile using the tile of the poor minimum of phase knowledge and magnanimity as current tile, and make the following judgment:

If the minimum poor threshold value that is less than setting of phase knowledge and magnanimity, and this most similar tile is that (in the present embodiment, adjacent tile is YC to adjacent tile _rc _bin image centered by current tile, tile in the region of some 2 tiles of extension),, using this most similar tile as code word, this code word be numbered and this code word is write to code book, then the numbering of this code word being write to position corresponding with current tile in code table;

If the minimum poor threshold value that is less than setting of phase knowledge and magnanimity, and this most similar tile is code book code word, and numbering corresponding to this code word write to position corresponding with current tile in code table;

Otherwise using current tile as code word, this code word be numbered and this code word write to code book, then the numbering of this code word being write to position corresponding with current tile in code table,

It is poor that described step (2-3) is calculated phase knowledge and magnanimity according to phase knowledge and magnanimity function, comprises the following steps:

(a) according to formula:

$M_{\mathrm{ij}}^k=a\×{(Y_i^k-Y_j^k)}^2+(1-a){(C_{\mathrm{ri}}^k-C_{\mathrm{rj}}^k)}^2+(1-a)\×{(C_{\mathrm{bi}}^k-C_{\mathrm{bj}}^k)}^2$

Calculate, wherein,

M _ijrepresent that k the pixel phase knowledge and magnanimity of k pixel and j tile in i tile are poor; A represents the coefficient (a=0.9 in the present embodiment) of brightness adjusting information,

with

the Pixel Information that represents respectively i k pixel and k pixel of j tile,

with

represent respectively in i tile that k pixel is at YC _rc _bthe Y passage of color space, C _rpassage and C _bthe value of passage,

with represent that respectively j tile is at YC _rc _bthe Y passage of color space, C _rpassage and C _bthe value of passage, k=0,1,2 ..., N, N is the pixel sum (N=16 in the present embodiment) in tile;

(b) it is poor that the value of traversal k obtains in i tile in each pixel and j tile the phase knowledge and magnanimity of respective pixel point, and averaged, obtains i tile poor with the phase knowledge and magnanimity of j tile;

(3) according to code book and code table, all code words in code book are carried out to class Huffman encoding and obtain codeword coding, described codeword coding comprises identification burst and code field, and generating gauge outfit information, described gauge outfit information comprises the size of texture image to be compressed and the length of identification burst;

Class Huffman encoding is specific as follows:

(3-1) occurrence number of each codeword number in the code word number of statistics in code book and code table, and according to the occurrence number of the numbering of each code word from how to sort to few;

(3-2) ask for and satisfy condition:

2 ¹+2 ²+…+2 ⁿ≥M，

The value of n minimum positive integer, and using this value as the length of identification burst, wherein, M is the code word number in code book;

(3-3) according to the sequence of the numbering occurrence number of code word, set the length of code field, occurrence number is more, and the length of code field is larger;

(3-4) according to the length of identification burst and the length of each code field set, each code word is carried out to the codeword coding that binary coding obtains each code word.

If have 5 code words in code book, the length of setting identification burst is 2, with code word occurrence number order from high to low in code table, each code word be encoded to 010,011,1000,1001,1010, finally obtain the codeword coding of all code words.

(4) by the numbering of corresponding codewords in codeword coding replacement code table, obtain code table coding, gauge outfit information is added to the head of code table coding, form the index file of texture image to be compressed.

Decompression procedure comprises:

(S1) scanning index file, according to the length of the size of the texture image recording in gauge outfit information and identification burst, obtain the size of texture image, and the codeword coding corresponding with each position in this texture image, according to the codeword coding inquiry code book obtaining, obtain the tile of each position, and by GPU, all tiles that obtain are reconstructed, obtain YC _rc _bimage;

(S2) YC to reconstruct _rc _bthe colourity of image is carried out the operation of interlacing bilinear interpolation, and is transformed into rgb color space and obtains the texture image after decompress(ion), is expressed as follows:

$\left[\begin{array}{c}R\\ G\\ B\end{array}\right]=\left[\begin{array}{ccc}1& 1.4020& 0\\ 1& -0.7141& 0.3441\\ 1& 0& 1.7720\end{array}\right]\×\left[\begin{array}{c}Y\\ C_r-128\\ C_b-128\end{array}\right]$

(S3) texture image step (S2) being obtained carries out the nl-means noise reduction process based on GPU.

Claims (9)

1. the decompression method based on texture image similarity, comprises compression process and decompression procedure, it is characterized in that,

Described compression process comprises:

(1) texture image to be compressed is transformed into YC from rgb color space _rc _bcolor space obtains YC _rc _bimage, and by YC _rc _bimage is cut into several tiles;

(2) utilize tile similarity, generate code book and the code table of image to be compressed, described code book comprises several code words, and described code table is for recording the numbering of the code word that each tile is corresponding;

(3) according to code book and code table, all code words in code book are carried out to class Huffman encoding and obtain codeword coding, described codeword coding comprises identification burst and code field, and generating gauge outfit information, described gauge outfit information comprises the size of texture image to be compressed and the length of identification burst;

(4) by the numbering of corresponding codewords in codeword coding replacement code table, obtain code table coding, gauge outfit information is added to the head of code table coding, form the index file of texture image to be compressed;

Described decompression procedure comprises:

(S1) scanning index file, according to the length of the size of the texture image recording in gauge outfit information and identification burst, obtain the size of texture image, and the codeword coding corresponding with each position in this texture image, according to the codeword coding inquiry code book obtaining, obtain the tile of each position, and by GPU, all tiles that obtain are reconstructed, obtain YC _rc _bimage;

(S2) YC to reconstruct _rc _bthe colourity of image is carried out the operation of interlacing bilinear interpolation, and is transformed into rgb color space and obtains the texture image after decompress(ion).

2. the decompression method based on texture image similarity as claimed in claim 1, is characterized in that, described tile size is 2 * 2～6 * 6 pixels.

3. the decompression method based on texture image similarity as claimed in claim 2, is characterized in that, described step (1) is by YC _rc _bimage is cut into before several tiles also to described YC _rc _bthe chrominance information of image is carried out partiting row sampling.

4. the decompression method based on texture image similarity as claimed in claim 3, is characterized in that, in described step (2), the generation code book of image to be compressed and the detailed process of code table are as follows:

(2-1) null set of initialization is as code book, and array of initialization is as code table, and the size of array is identical with the sum of tile;

(2-2) choose arbitrarily a tile as initial tile, using initial tile as code word, and this code word is numbered, then the numbering of this code word is write to position corresponding with this tile in code table;

(2-3), for other any one tiles, the phase knowledge and magnanimity that calculate respectively each code word in current tile and code table and each adjacent tile are poor, the most similar tile using the tile of the poor minimum of phase knowledge and magnanimity as current tile, and make the following judgment:

If the minimum poor threshold value that is less than setting of phase knowledge and magnanimity, and this most similar tile is adjacent tile,, using this most similar tile as code word, this code word be numbered and this code word is write to code book, then the numbering of this code word being write to position corresponding with current tile in code table;

If the minimum poor threshold value that is less than setting of phase knowledge and magnanimity, and this most similar tile is the code word in code book, and numbering corresponding to this code word write to position corresponding with current tile in code table;

Otherwise using current tile as code word, this code word be numbered and this code word is write to code book, then the numbering of this code word being write to position corresponding with current tile in code table.

5. the decompression method based on texture image similarity as claimed in claim 4, is characterized in that, described adjacent tile is YC _rc _bin image centered by current tile, the tile in the region of some 1～3 tiles of extension.

6. the decompression method based on texture image similarity as claimed in claim 5, is characterized in that, it is poor that described step (2-3) is calculated phase knowledge and magnanimity according to phase knowledge and magnanimity function, comprises the following steps:

(a) according to formula:

$M_{\mathrm{ij}}^k=a\×{(Y_i^k-Y_j^k)}^2+(1-a){(C_{\mathrm{ri}}^k-C_{\mathrm{rj}}^k)}^2+(1-a)\×{(C_{\mathrm{bi}}^k-C_{\mathrm{bj}}^k)}^2$

Calculate, wherein M _ijrepresent that k the pixel phase knowledge and magnanimity of k pixel and j tile in i tile are poor; A represents the coefficient of brightness adjusting information, (Y _i, C _ri, C _bi) and (Y _j, C _rj, C _bj) represent respectively the tile information of i and j tile, Y _i,

represent that respectively i tile is at YC _rc _bthe Y passage of color space, C _rpassage and C _bthe value of passage, Y _j,

represent that respectively j tile is at YC _rc _bthe Y passage of color space, C _rpassage and C _bthe value of passage, k=0,1,2 ..., N, N is the pixel sum in tile;

(b) it is poor that the value of traversal k obtains in i tile in each pixel and j tile the phase knowledge and magnanimity of respective pixel point, and averaged, obtains i tile poor with the phase knowledge and magnanimity of j tile.

7. the decompression method based on texture image similarity as claimed in claim 6, is characterized in that, the threshold value of setting in described step (2-3) is 200～400.

8. the decompression method based on texture image similarity as described in claim 1～7, is characterized in that, the class Huffman encoding in described step (3) comprises the following steps:

(3-1) occurrence number of each codeword number in the code word number of statistics in code book and code table, and according to the occurrence number of the numbering of each code word from how to sort to few;

(3-2) ask for and satisfy condition:

2 ¹+2 ²+…+2 ⁿ≥M，

The value of n minimum positive integer, and using this value as the length of identification burst, wherein, M is the code word number in code book;

(3-3) according to the sequence of the numbering occurrence number of code word, set the length of code field, occurrence number is more, and the length of code field is larger;

(3-4) according to the length of identification burst and the length of each code field set, each code word is carried out to the codeword coding that binary coding obtains each code word.

9. the decompression method based on texture image similarity as claimed in claim 8, is characterized in that, described decompression process also comprises carries out the nl-means noise reduction process based on GPU to described texture image.