CN103501437B - A kind of based on fractal and H.264 method for compressing high spectrum image - Google Patents

Abstract

The present invention proposes a kind of based on fractal and H.264 method for compressing high spectrum image。First high spectrum image cubic data body is converted to the video of yuv format, sends into encoder。The I frame of EO-1 hyperion video, uses the H.264 fast intra-mode prediction mode discrimination method based on macro block flatness to carry out infra-frame prediction, to remove the spatial coherence of high spectrum image；The P frame of EO-1 hyperion video carries out block motion estimation/compensation fractal image, to remove the Spectral correlation of high spectrum image。First the various tree piecemeals of P frame coded macroblocks are found best matching blocks by MSE criterion in reference frame, determine the iterated function system coefficient of each piece, then the rate distortion costs of all pieces of dividing mode is compared, using block dividing mode minimum for rate distortion costs as final interframe encoding mode, record final fractal parameter。Finally, the residual error data of I frame and P frame is encoded write code stream by entropy code CABAC after dct transform, quantization, and the fractal parameter of P frame is also carried out CABAC entropy code。

Description

A kind of based on fractal and H.264 method for compressing high spectrum image

Technical field:

The invention belongs to Compression of hyperspectral images field, for the spatial coherence existed in high spectrum image and Spectral correlation, propose a kind of based on fractal and H.264 method for compressing high spectrum image, under the premise ensureing picture quality, it is greatly accelerated the compression speed of high spectrum image, improves compression ratio。

Background technology:

One of focus that hyperspectral technique is the frontier development of 21 century remote sensing technology and current remote sensing circle is paid close attention to, it organically combines traditional two-dimensional imaging remote sensing technology and spectral technique, while obtain the spatial information of measured object by imaging system, become the spectrum of different wave length to radiate the radiolysis of measured object by spectrometer system, each pixel tens even hundreds of continuous print carrier wave segment information can be obtained in a spectrum range。High spectrum resolution remote sensing technique is own through being successfully applied in geology, ecology, atmospheric research, a lot of field of soil investigation, it is shown that very big potentiality and vast potential for future development。

High spectrum image, on the basis of ground image two-dimensional signal, adds third dimension spectral information。High-spectral data is the cube of a spectrum picture, and its spatial image dimension describes earth's surface two-dimensional space feature, and its spectrum dimension discloses the curve of spectrum feature of each pixel of image。Spatial image in hyperspectral image data ties up Existential Space redundancy, and spectrum dimension exists redundancy between spectrum。Therefore Compression of hyperspectral images is should consider to remove spatial coherence with the difference of Ordinary image compression, considers again to remove Spectral correlation。At present, conventional method for compressing high spectrum image is divided into three major types: based on the method for prediction, the method based on conversion and the method based on vector quantization。Such as propose the compression scheme based on linear model optimum prediction with Zhang Ye etc. during Chen Yu, utilize the thought that recurrence is bi-directional predicted, by setting up the linear model between bands of a spectrum, show that the optimum prediction under signal to noise ratio is (referring to Chen Yushi, Zhang Ye, Zhang Jun is put down. based on the Compression of hyperspectral images [J] of linear model optimum prediction. and Nanjing Aero-Space University's journal, 2007,39 (3): 368-372.)。PennaB extracts the amount of calculation of strategy reduction KLT by adopting to count, and the KLT of this improvement is used for decorrelation between high spectrum image spectrum, and reconstructed image quality is not significantly affected (referring to PennaB, TilloT, MagliE, etal.Transformcodingtechniquesforlossyhyperspectraldatac ompression [J] .IEEETransactionsonGeoscienceandRemoteSensing, 2007,45 (5): 1408-1421.)。Shen-EnQian proposes a kind of mensuration algorithm of quick vector to improve the efficiency of the generation of code book, at this algorithm without carrying out full search, greatly reduce the complexity (referring to Shen-EnQian.Hyperspectraldatacompressionusingafastvector quantizationalgorithm [J] .IEEETransactionsonGeoscienceandRemoteSensing.2004,42 (8): 1791-1798.) of computing。Except above-mentioned traditional coded method, fractal image is also applied to the compression of high spectrum image in recent years, as Xia Lili devises a three-dimensional fractal Coding Compression Algorithm, the dependency between band image is eliminated while removing image space dependency, obtain good compression effectiveness (referring to Xia Lili. based on the research [J] of the Hyperspectral image compression algorithm of fractal theory. computer and digital engineering, 2011,39 (9): 132-135.)。

Present invention incorporates predictive coding, fractal image, dct transform coding, CABAC entropy code, the H.264 fast intra-mode prediction coded method based on macro block flatness is utilized to reduce the spatial coherence of high spectrum image, fractal image is utilized to reduce its Spectral correlation, and coded residual data are carried out dct transform, quantization, code stream is write with CABAC entropy code, fractal parameter is also carried out CABAC entropy code, it is achieved that being effectively compressed of high spectrum image。

Summary of the invention:

The present invention proposes a kind of based on fractal and H.264 method for compressing high spectrum image。First high spectrum image cubic data body is converted to the video of yuv format, sends into encoder。The I frame of EO-1 hyperion video, uses the H.264 fast intra-mode prediction mode discrimination method based on macro block flatness to carry out infra-frame prediction, to remove the spatial coherence of high spectrum image；The P frame of EO-1 hyperion video carries out block motion estimation/compensation fractal image, to remove the Spectral correlation of high spectrum image。First the various tree piecemeals of P frame coded macroblocks are found best matching blocks by MSE criterion in reference frame, determine the iterated function system coefficient of each piece, then the rate distortion costs of all pieces of dividing mode is compared, using block dividing mode minimum for rate distortion costs as final interframe encoding mode, record final fractal parameter。Finally, the residual error data of I frame and P frame is encoded write code stream by entropy code CABAC after dct transform, quantization, and the fractal parameter of P frame is also carried out CABAC entropy code。

A kind of based on fractal and H.264 method for compressing high spectrum image, it is characterised in that to realize step as follows:

Step one: high spectrum image cube metadata is converted to the video of yuv format；

Step 2: if (the first frame is necessary for I frame to the I frame of EO-1 hyperion video, whether other frame can be arranged is I frame), the H.264 fast intra-mode prediction mode discrimination method based on macro block flatness is used to carry out Intra prediction mode selection, predicted macroblock, can obtain the prediction frame of I frame after completing the prediction of all pieces。Obtain coding side residual frame by the difference of primitive frame and prediction frame, proceed to step 4 coded residual；If the P frame of EO-1 hyperion video, forward step 3 to；

Step 3: if the P frame of EO-1 hyperion video, successively all macro blocks of present frame are carried out fractal image。In search window in reference frame, current macro being carried out Block-matching, the position of sub-block is as the initiating searches point of father's block, and the size of father's block is identical with the size of sub-block。Each macro block is carried out tree piecemeal, and namely piecemeal can be divided into 16 × 16,16 × 8,8 × 16,8 × 8,8 × 8 down (sub-macroblock partition) can be divided into 8 × 4,4 × 8,4 × 4 from big to small。First the various tree piecemeals of coded macroblocks are found best matching blocks by MSE criterion, it is determined that the iterated function system coefficient of each piece and IFS coefficient；Then the rate distortion costs of all pieces of dividing mode is compared；Finally using block dividing mode minimum for rate distortion costs as final interframe encoding mode。Record final IFS coefficient, proceed to step 5 and obtain the reconstructed block of this block。If all of macro block of present frame is all encoded complete, all of reconstructed block composition rebuilds image (i.e. the reference frame of next frame), obtains coding side residual image by original image and the difference rebuilding image, forwards step 4 coded residual to。Described search window is the rectangular search region in reference frame；Described IFS coefficient includes the position of father's block and sub-block and offsets, namely motion vector (x, y) and scale factor s, displacement factor o；

Step 4: the data of residual image coefficient after DCT, quantization carries out Zig-Zag scanning on the one hand, is then encoded write code stream with entropy code CABAC；Decoding end residual frame is obtained on the other hand after inverse quantization, inverse DCT conversion。Reconstruction frames (i.e. the reference frame of next frame) is obtained by prediction frame and decoding end residual frame sum。If all IFS coefficients then also to be carried out CABAC entropy code by P frame。Judge whether present frame is last frame, if last frame terminates coding；Otherwise, return step 2 and continue with next frame image；

Step 5: substitute into decoding equation by the IFS coefficient preserved

r_i=s·d_i+o(1)

Calculating obtains predictive value, the difference of original block and prediction block obtain coding side residual block, and coding side residual block obtains decoding end residual block through the conversion of dct transform, quantization, inverse quantization and inverse DCT, then is obtained reconstructed block by prediction block and decoding end residual block sum。Proceed to step 3 coding depth graphic sequence next macro block of P frame。R in formula_iFor predicting the pixel value of block, d_iPixel value for the corresponding father's block of reference frame。

The video that high spectrum image cube metadata converts in described step one yuv format includes following two step:

1) hyperspectral image data is normalized to the integer in [0,255] interval。Hyperspectral image data represents the spectral reflectivity of atural object, and its bit depth is generally 16 bits/pixel, it is necessary to convert 8 bits/pixel to。Conversion method is that first the value of spectral reflectivity is normalized to [0,255] interval, is then converted into the integer without symbol 8 bit, and final numerical value is the integer that [0,255] is interval；

2) using each wave band of hyperspectral image data as a frame, the reflection coefficient after the normalization of each wave band is as the luminance elements (Y-component) of this frame, and chromatic component Cb and Cr is all set to intermediate value 128。Circulation performs step 2) until all wave bands of high spectrum image are disposed。Namely all frames form the video file of a yuv format。

Described step 2 includes following five steps based on the H.264 fast intra-mode prediction mode discrimination method of macro block flatness:

1) calculate current coding macro block two-dimensional gray histogram, and record the maximum z of z-axis in two-dimensional gray histogram_max。

The two-dimensional gray histogram of image is based on the three-dimensional description figure that the Joint Distribution of image pixel gray level and neighborhood of pixels gray average is constituted。If image f (x, y) be sized to M × N, gray level is L。By f, (x, y) adopting k × k dot matrix to smooth the neighborhood averaging gray level image obtained is that (x y), is designated as g

$g(x,y)=\frac{1}{k^2}\underset{m=-\frac{k}{2}}{\overset{\frac{k}{2}}{\mathrm{\Σ}}}\underset{n=-\frac{k}{2}}{\overset{\frac{k}{2}}{\mathrm{\Σ}}}f(x+m,y+n)---\left(2\right)$

In formula, 1≤x+m≤M, 1≤y+n≤N, k takes odd number。

Then g (x, image size y) and gray level and f (and x, y) identical。By f, (x, y) (x, (i, j), each two tuples belong to a point on two dimensional surface y) to may be constructed two tuples with g。Definition two-dimensional histogram be N (i, j), represent image f (x, y) grey scale pixel value is i, and when the gray value of neighborhood of pixels average gray image is j, two-dimensional points (i, number of times j) occurred (i, j=0,1 ..., L-1)。Z-axis in two-dimensional gray histogram is N, and (i j), represents two-dimensional points (i, number of times j) occurred；

2) capping threshold value Th_highWith lower threshold Th_low, Th_highAnd Th_lowIt is the integer between [1,256]；

3) if z_max≥Th_high, it is believed that macro block is smooth, removes Intra4 × 4 patterns, selects Intra16 × 16 pattern, and using pattern minimum for rate distortion expense as optimal frames inner estimation mode；Update higher limit simultaneouslyOtherwise proceed to step 4)；

4) if z_max≤Th_low, it is believed that macro block details is enriched, and removes Intra16 × 16 patterns, selects Intra4 × 4 pattern, and using pattern minimum for rate distortion expense as optimal frames inner estimation mode；Update lower limit simultaneouslyOtherwise proceed to step 5)；

5) if Th_low<z_max<Th_high, it is believed that macro block flatness feature is not notable, adopts H.264 standard intraframe prediction algorithm。

In described step 3, P frame macro block fractal image includes following ten steps:

1) first in unit search window in reference frame, current macro is carried out Block-matching with 16 × 16 macro blocks。The position of sub-block is as the initiating searches point of father's block, and the size of father's block is identical with the size of sub-block。Travel through whole search window according to full search mode, find the minimum father's block position of matching error MSE。Record this position iterated function system coefficient and IFS coefficient (include motion vector (x, y), scale factor s, displacement factor o), and calculate the rate distortion expense cost1 of this pattern；

2) this 16 × 16 macro block is divided into the fritter of 2 16 × 8, each fritter is all traveled through corresponding whole search window according to full search mode, find the minimum corresponding father's block position of matching error MSE。The IFS coefficient of record the two position, and calculate the rate distortion expense sum cost2 of two blocks；

3) this 16 × 16 macro block is divided into the fritter of 28 × 16, each fritter is all traveled through corresponding whole search window according to full search mode, find the minimum corresponding father's block position of matching error MSE。The IFS coefficient of record the two position, and calculate the rate distortion expense sum cost3 of two blocks；

4) this 16 × 16 macro block is divided into the fritter of 48 × 8, each fritter is all traveled through corresponding whole search window according to full search mode, find the minimum corresponding father's block position of matching error MSE。Record the IFS coefficient of these 4 positions, and calculate the rate distortion expense cost4_1 of these 4 blocks, cost4_2, cost4_3, cost4_4 and four rate distortion expense sum cost4 respectively；

5) comparison step 1) to step 4) in rate distortion expense cost1, cost2, cost3, cost4, if minima is cost4, then needs to continue to divide by each 8 × 8 pieces further, forward step 6 to)；The dividing mode that otherwise rate distortion expense is minimum is the P frame encoding mode that this macro block is final, retains corresponding IFS coefficient, terminates；

6) it is first divided into 28 × 4 fritters by the 1st 8 × 8 pieces, each fritter is all traveled through corresponding whole search window according to full search mode, find the minimum corresponding father's block position of matching error MSE。The IFS coefficient of record the two position, and calculate the rate distortion expense sum cost5 of the two block；

7) it is divided into 24 × 8 fritters by the 1st 8 × 8 pieces, each fritter is all traveled through corresponding whole search window according to full search mode, find the minimum corresponding father's block position of matching error MSE。The IFS coefficient of record the two position, and calculate the rate distortion expense sum cost6 of the two block；

8) it is divided into 44 × 4 fritters by the 1st 8 × 8 pieces, each fritter is all traveled through corresponding whole search window according to full search mode, find the minimum corresponding father's block position of matching error MSE。Record the IFS coefficient of these 4 positions, and calculate the rate distortion expense sum cost7 of these 4 blocks；

9) compare cost4_1, cost5, cost6, cost7, using block dividing mode corresponding for minima therein as this coding mode 8 × 8 pieces final, retain corresponding IFS coefficient；

10) successively to 16 × 16 macro blocks each 8 × 8 pieces according to step 6) to step 9) and the final coding mode of method choice, retain corresponding IFS coefficient。

Wherein being calculated as follows of matching error MSE:

$\mathrm{MSE}(R,D)=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{[r_i-(s\&CenterDot;d_i+o)]}^2=\frac{1}{N}[\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{r}_i^2+s(s\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{d}_i^2-2\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{r}_i{d}_i+2o\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{d}_i^2)+o(N\&CenterDot;o-2\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{r}_i^2)]---\left(3\right)$

In formula, the calculating formula of s, o is respectively as follows:

$s=\frac{[N\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{r}_i{d}_i-\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{r}_i\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{d}_i]}{[N\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{d}_i^2-{\left(\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{d}_i\right)}^2]}---\left(4\right)$

$o=\frac{1}{N}[\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{r}_i-s\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{d}_i]---\left(5\right)$

Wherein, r_iIt is each point pixel value of current block, d_iBeing each point pixel value of match block, N is the number of pixels of current block, and s is scale factor, and o is displacement factor。

Being calculated as follows of rate distortion expense:

cost=MSE(R,D)+λ_MODE·Rate(x,y,s,o,MODE|QP)(6)

Wherein, R represents current block, and D represents match block, and MODE represents macro block dividing mode, and QP represents quantization parameter, λ_MODEIt is the LaGrange parameter relevant with QP。

Proposed by the invention it is in that based on fractal and method for compressing high spectrum image H.264 advantage:

(1) first high spectrum image cube metadata has been carried out form conversion by this method, converts the video of yuv format to, thus can encode high spectrum image as coding ordinary video；

(2) this method adopts the H.264 fast intra-mode prediction based on macro block flatness to encode the spatial coherence existed in reduction high spectrum image。Adopt two-dimensional gray histogram to characterize macro block flatness feature, and be provided with dual threshold, smooth macro block is removed Intra4 × 4 pattern；The macro block that details is abundant removes Intra16 × 16 pattern；The unconspicuous macro block of flatness feature is adopted H.264 canonical algorithm。Threshold value Dynamic iterations updates, and makes judgment threshold can adjust in real time according to the fluctuation rebuilding macro block flatness。Both ensure that the quality that high spectrum image decompresses, accelerate again compression speed；

(3) this method adopts fractal image to reduce the Spectral correlation existed in high spectrum image。Block-matching support 16 × 16,16 × 8,8 × 16,8 × 8,8 × 4,4 × 8,4 × 4 totally 7 kinds of block dividing mode, every kind of block dividing mode first looks for the minimum match block of matching error MSE, then the rate distortion costs of all patterns is compared, finally using pattern minimum for rate distortion costs as optimum fractal image pattern。Improve the precision of fractal image。

Accompanying drawing illustrates:

Fig. 1 is that the present invention is a kind of based on fractal and H.264 method for compressing high spectrum image flow chart；

Fig. 2 is the 17th wave band grayscale sub-image of high spectrum image " DCMall "；

Fig. 3 is that the present invention is a kind of based on H.264 fast intra-mode prediction mode discrimination method flow chart based on macro block flatness in fractal and method for compressing high spectrum image step 2 H.264

Fig. 4 is that the present invention is a kind of based on P frame macro block fractal image model selection flow chart in fractal and method for compressing high spectrum image step 3 H.264；

Fig. 5 be through the present invention a kind of based on fractal and H.264 method for compressing high spectrum image compression again rebuild after " DCMall " high spectrum image the 17th wave band gray-scale map；

Fig. 6 (a) is the comparison diagram of a kind of Y-PSNR that " DCMall " high spectrum image is compressed coding based on fractal and method for compressing high spectrum image H.264 and Lucana et al. method of the present invention；

Fig. 6 (b) is a kind of based on fractal and H.264 method for compressing high spectrum image and Lucana et al. the method comparison diagram to the bit number that " DCMall " high spectrum image is compressed for the present invention；

Fig. 6 (c) is a kind of based on fractal and H.264 method for compressing high spectrum image and Lucana et al. the method comparison diagram to the time that " DCMall " high spectrum image is compressed for the present invention。

Detailed description of the invention:

Below in conjunction with accompanying drawing, the inventive method is described in further detail。

The present invention proposes a kind of based on fractal and H.264 method for compressing high spectrum image。First high spectrum image cubic data body is converted to the video of yuv format, sends into encoder。The I frame of EO-1 hyperion video, uses the H.264 fast intra-mode prediction mode discrimination method based on macro block flatness to carry out infra-frame prediction, to remove the spatial coherence of high spectrum image；The P frame of EO-1 hyperion video carries out block motion estimation/compensation fractal image, to remove the Spectral correlation of high spectrum image。First the various tree piecemeals of P frame coded macroblocks are found best matching blocks by MSE criterion in reference frame, determine the iterated function system coefficient of each piece, then the rate distortion costs of all pieces of dividing mode is compared, using block dividing mode minimum for rate distortion costs as final interframe encoding mode, record final fractal parameter。Finally, the residual error data of I frame and P frame is encoded write code stream by entropy code CABAC after dct transform, quantization, and the fractal parameter of P frame is also carried out CABAC entropy code。

As shown in Figure 1, a kind of based on fractal and H.264 method for compressing high spectrum image flow chart。Shopping center, Washington DC (DCMall) high spectrum image provided for spectral information technical applications center, Virginia, intercepts 256 × 1024 pixel subimages。Removing the wave band of 0.9 to 1.4 bigger μ m of its medium cloud coverage rate, the final wave band number retained is 191。Accompanying drawing 2 is the 17th wave band grayscale sub-image of high spectrum image " DCMall "。

Step one: high spectrum image cube metadata is converted to the video of yuv format。Specifically comprise the following steps that

1st step, hyperspectral image data are normalized to the integer in [0,255] interval。Hyperspectral image data represents the spectral reflectivity of atural object, and its bit depth is generally 16 bits/pixel, it is necessary to convert 8 bits/pixel to。Conversion method is that first the value of spectral reflectivity is normalized to [0,255] interval, is then converted into the integer without symbol 8 bit, and final numerical value is the integer that [0,255] is interval；

2nd step, using each wave band of hyperspectral image data as a frame, the reflection coefficient after the normalization of each wave band is as the luminance elements (Y-component) of this frame, and chromatic component Cb and Cr is all set to intermediate value 128。Circulation performs the 2nd step until all wave bands of high spectrum image are disposed。Namely all frames form the video file of a yuv format。

Step 2: judge whether present encoding EO-1 hyperion frame of video is that (the first frame is necessary for I frame to I frame, whether other frame can be arranged is I frame), if I frame, the H.264 fast intra-mode prediction mode discrimination method based on macro block flatness is used to carry out Intra prediction mode selection, it was predicted that macro block。Accompanying drawing 3 is the H.264 fast intra-mode prediction mode discrimination method flow chart based on macro block flatness。Specifically include following 5 steps:

1st step, calculating current coding macro block two-dimensional gray histogram, and record the maximum z of z-axis in two-dimensional gray histogram_max。

The two-dimensional gray histogram of image is based on the three-dimensional description figure that the Joint Distribution of image pixel gray level and neighborhood of pixels gray average is constituted。If image f (x, y) be sized to M × N, gray level is L。By f, (x, y) adopting k × k dot matrix to smooth the neighborhood averaging gray level image obtained is that (x y), is designated as g

$g(x,y)=\frac{1}{k^2}\underset{m=-\frac{k}{2}}{\overset{\frac{k}{2}}{\mathrm{\&Sigma;}}}\underset{n=-\frac{k}{2}}{\overset{\frac{k}{2}}{\mathrm{\&Sigma;}}}f(x+m,y+n)---\left(7\right)$

In formula, 1≤x+m≤M, 1≤_y+ n≤N, k takes odd number。

Then g (x, image size y) and gray level and f (and x, y) identical。By f, (x, y) (x, (i, j), each two tuples belong to a point on two dimensional surface y) to may be constructed two tuples with g。Definition two-dimensional histogram be N (i, j), represent image f (x, y) grey scale pixel value is i, and when the gray value of neighborhood of pixels average gray image is j, two-dimensional points (i, number of times j) occurred (i, j=0,1 ..., L-1)。Z-axis in two-dimensional gray histogram is N, and (i j), represents two-dimensional points (i, number of times j) occurred；

2nd step, capping threshold value Th_high(empirical value is set to 175) and lower threshold Th_low(empirical value is set to 100), Th_highAnd Th_lowIt is the integer between [1,256]；

If the 3rd step z_max≥Th_high, it is believed that macro block is smooth, removes Intra4 × 4 patterns, selects Intra16 × 16 pattern, and using pattern minimum for rate distortion expense as optimal frames inner estimation mode；Update higher limit simultaneouslyOtherwise proceed to the 4th step；

If the 4th step z_max≤Th_low, it is believed that macro block details is enriched, and removes Intra16 × 16 patterns, selects Intra4 × 4 pattern, and using pattern minimum for rate distortion expense as optimal frames inner estimation mode；Update lower limit simultaneouslyOtherwise proceed to the 5th step；

If the 5th step Th_low<z_max<Th_high, it is believed that macro block flatness feature is not notable, adopts H.264 standard intraframe prediction algorithm。

The prediction frame of I frame can be obtained after completing the prediction of all pieces。Obtain coding side residual frame by the difference of primitive frame and prediction frame, proceed to step 4 coded residual；If being the P frame of EO-1 hyperion video, forward step 3 to。

Step 3: if being the P frame of EO-1 hyperion video, successively all macro blocks of present frame are carried out fractal image。In search window in reference frame, current macro being carried out Block-matching, the position of sub-block is as the initiating searches point of father's block, and the size of father's block is identical with the size of sub-block。Each macro block is carried out tree piecemeal, and namely piecemeal can be divided into 16 × 16,16 × 8,8 × 16,8 × 8,8 × 8 down (sub-macroblock partition) can be divided into 8 × 4,4 × 8,4 × 4 from big to small。First the various tree piecemeals of coded macroblocks are found best matching blocks by MSE criterion, it is determined that the iterated function system coefficient of each piece and IFS coefficient；Then the rate distortion costs of all pieces of dividing mode is compared；Finally using block dividing mode minimum for rate distortion costs as final interframe encoding mode。Accompanying drawing 4 is P frame macro block fractal image model selection flow chart, specifically includes following 10 steps:

The first step, first in unit search window in reference frame, current macro is carried out Block-matching with 16 × 16 macro blocks。The position of sub-block is as the initiating searches point of father's block, and the size of father's block is identical with the size of sub-block。Travel through whole search window according to full search mode, find the minimum father's block position of matching error MSE。Record this position iterated function system coefficient and IFS coefficient (include motion vector (x, y), scale factor s, displacement factor o), and calculate the rate distortion expense cost1 of this pattern；

Second step, this 16 × 16 macro block is divided into the fritter of 2 16 × 8, each fritter is all traveled through corresponding whole search window according to full search mode, finds the minimum corresponding father's block position of matching error MSE。The IFS coefficient of record the two position, and calculate the rate distortion expense sum cost2 of two blocks；

3rd step, this 16 × 16 macro block is divided into the fritter of 28 × 16, each fritter is all traveled through corresponding whole search window according to full search mode, finds the minimum corresponding father's block position of matching error MSE。The IFS coefficient of record the two position, and calculate the rate distortion expense sum cost3 of two blocks；

4th step, this 16 × 16 macro block is divided into the fritter of 48 × 8, each fritter is all traveled through corresponding whole search window according to full search mode, finds the minimum corresponding father's block position of matching error MSE。Record the IFS coefficient of these 4 positions, and calculate the rate distortion expense cost4_1 of these 4 blocks, cost4_2, cost4_3, cost4_4 and four rate distortion expense sum cost4 respectively；

5th step, compare the first step to the rate distortion expense cost1 in the 4th step, cost2, cost3, cost4, if minima is cost4, then needs to continue to divide by each 8 × 8 pieces further, forward the 6th step to；The dividing mode that otherwise rate distortion expense is minimum is the P frame encoding mode that this macro block is final, retains corresponding IFS coefficient, terminates；

6th step, it is first divided into 28 × 4 fritters by the 1st 8 × 8 pieces, each fritter is all traveled through corresponding whole search window according to full search mode, find the minimum corresponding father's block position of matching error MSE。The IFS coefficient of record the two position, and calculate the rate distortion expense sum cost5 of the two block；

7th step, it is divided into 24 × 8 fritters by the 1st 8 × 8 pieces, each fritter is all traveled through corresponding whole search window according to full search mode, find the minimum corresponding father's block position of matching error MSE。The IFS coefficient of record the two position, and calculate the rate distortion expense sum cost6 of the two block；

8th step, it is divided into 44 × 4 fritters by the 1st 8 × 8 pieces, each fritter is all traveled through corresponding whole search window according to full search mode, find the minimum corresponding father's block position of matching error MSE。Record the IFS coefficient of these 4 positions, and calculate the rate distortion expense sum cost7 of these 4 blocks；

9th step, compare cost4_1, cost5, cost6, cost7, using block dividing mode corresponding for minima therein as this coding mode 8 × 8 pieces final, retain corresponding IFS coefficient；

Tenth step, successively to 16 × 16 macro blocks each 8 × 8 pieces according to the final coding mode of the 6th step to the method choice of the 9th step, retain corresponding IFS coefficient, terminate。

Wherein being calculated as follows of matching error MSE:

$\mathrm{MSE}(R,D)=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{[r_i-(s\&CenterDot;d_i+o)]}^2=\frac{1}{N}[\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{r}_i^2+s(s\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{d}_i^2-2\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{r}_i{d}_i+2o\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{d}_i^2)+o(N\&CenterDot;o-2\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{r}_i^2)]---\left(8\right)$

In formula, the calculating formula of s, o is respectively as follows:

$s=\frac{[N\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{r}_i{d}_i-\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{r}_i\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{d}_i]}{[N\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{d}_i^2-{\left(\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{d}_i\right)}^2]}---\left(9\right)$

$o=\frac{1}{N}[\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{r}_i-s\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{d}_i]---\left(10\right)$

Wherein, r_iIt is each point pixel value of current block, d_iBeing each point pixel value of match block, N is the number of pixels of current block, and s is scale factor, and o is displacement factor。

Being calculated as follows of rate distortion expense:

cost=MSE(R,D)+λ_MODE·Rate(x,y,s,o,MODE|QP)(11)

Wherein, R represents current block, and D represents match block, and MODE represents macro block dividing mode, and QP represents quantization parameter, λ_MODEIt is the LaGrange parameter relevant with QP。

Obtained, by above-mentioned steps, the coding mode that P frame macro block is final, record final IFS coefficient, proceed to step 5 and obtain the reconstructed block of this block。If all of macro block of present frame is all encoded complete, all of reconstructed block composition rebuilds image (i.e. the reference frame of next frame), obtains coding side residual image by original image and the difference rebuilding image, forwards step 4 coded residual to。Described search window is the rectangular search region in reference frame；Described IFS coefficient includes the position of father's block and sub-block and offsets, namely motion vector (x, y) and scale factor s, displacement factor o。

Step 4: the data of residual image coefficient after DCT, quantization carries out Zig-Zag scanning on the one hand, is then encoded write code stream with entropy code CABAC；Decoding end residual frame is obtained on the other hand after inverse quantization, inverse DCT conversion。Reconstruction frames (i.e. the reference frame of next frame) is obtained by prediction frame and decoding end residual frame sum。If all IFS coefficients then also to be carried out CABAC entropy code by P frame。Judge whether present frame is last frame, if last frame terminates coding；Otherwise, return step 2 and continue with next frame image。

Step 5: substitute into decoding equation by the IFS coefficient preserved

r_i=s·d_i+o(12)

Calculating obtains predictive value, the difference of original block and prediction block obtain coding side residual block, and coding side residual block obtains decoding end residual block through the conversion of dct transform, quantization, inverse quantization and inverse DCT, then is obtained reconstructed block by prediction block and decoding end residual block sum。Proceed to step 3 coding EO-1 hyperion video next macro block of P frame。R in formula_iFor predicting the pixel value of block, d_iPixel value for the corresponding father's block of reference frame。

This method selects visual c++ 6.0 as the platform that realizes of described method, and CPU is IntelCore^TM2DuoT8300,2.4GHz dominant frequency, memory size is 2G, 191 wave band cube metadatas of high spectrum image " DCMall " are carried out based on fractal and H.264 Compression of hyperspectral images experiment, after converting the video file of yuv format to, totalframes is 191 frames equal to total wave band number, every two field picture is sized to 256 × 1024 pixels, coded image group structure is IPPP ..., I frame code period is 50, and namely the first frame is encoded to I frame, encodes 1 I frame every 49 P frames afterwards, hunting zone is ± 7, and quantization parameter QP is 20。Accompanying drawing 5 be through the present invention a kind of based on fractal and H.264 method for compressing high spectrum image compression again rebuild after " DCMall " high spectrum image the 17th wave band gray-scale map。

It is respectively adopted Lucana et al. method (referring to LucanaSantos, Sebasti á nL ó pez, GustavoM.Callic ó, etal.PerformanceevaluationoftheH.264/AVCvideocodingstand ardforlossyhyperspectralimagecompression [J] .IEEEJournalofSelectedTopicsinAppliedEarthObservationsan dRemoteSensing, 2012, 5 (2): 451-461.) and the inventive method to the comparison diagram of the Y-PSNR that " DCMall " high spectrum image is compressed such as shown in accompanying drawing 6 (a)；It is respectively adopted Lucana et al. method and the inventive method comparison diagram to the bit number that " DCMall " high spectrum image is compressed such as shown in accompanying drawing 6 (b)；It is respectively adopted Lucana et al. method and the inventive method comparison diagram to the time that " DCMall " high spectrum image is compressed such as shown in accompanying drawing 6 (c)。

It is respectively adopted Lucana et al. method and the inventive method 191 wave band compression performance average comparing results of " DCMall " high spectrum image are as shown in table 1。Wherein △ PSNR, △ bit rate, △ compression time definition as follows:

△PSNR=PSNR_OURS-PSNR_Lucana(13)

△ bit rate=(bit rate_OURS-bit rate_Lucana)/bit rate_Lucana(14)

△ compression time=(compression time_OURS-compression time_Lucana)/compression time_Lucana(15)

Depth map sequence	△PSNR/dB	△ bit rate	△ compression time
				DC Mall	0.67	-90.50%	-90.00%

Table 1Lucana et al. method and the inventive method are to 191 the wave band compression performance average contrasts of " DCMall " high spectrum image

Can be seen that from accompanying drawing 6 and table 1, the inventive method is compared with traditional International video coding standard H.264 standard test models JM18.1 method, Y-PSNR PSNR on average improves 0.67dB, encoding code stream bit rate on average reduces by 17.73%, compression time on average reduces 90.00%, shows superior compression performance。This is because the inventive method is on depth map sequence I frame encodes, have employed the H.264 fast intra-mode prediction coded method based on macro block flatness of improvement, remove the spatial coherence of high spectrum image；On depth map sequence P frame encodes, have employed block motion estimation/compensation fractal coding algorithm, remove the Spectral correlation of high light image。

Claims (4)

1. one kind based on fractal and H.264 method for compressing high spectrum image, it is characterised in that following steps:

Step one: high spectrum image cube metadata is converted to the video of yuv format；

Step 2: if the I frame of EO-1 hyperion video, use the H.264 fast intra-mode prediction mode discrimination method based on macro block flatness to carry out Intra prediction mode selection, it was predicted that macro block, the prediction frame of I frame can be obtained after completing the prediction of all pieces；Obtain coding side residual frame by the difference of primitive frame and prediction frame, proceed to step 4 coded residual；If the P frame of EO-1 hyperion video, forward step 3 to；

Step 3: if the P frame of EO-1 hyperion video, successively all macro blocks of present frame are carried out fractal image；In search window in reference frame, current macro being carried out Block-matching, the position of sub-block is as the initiating searches point of father's block, and the size of father's block is identical with the size of sub-block；Each macro block is carried out tree piecemeal, and namely piecemeal can be divided into 16 × 16,16 × 8,8 × 16,8 × 8,8 × 8 down to carry out sub-macroblock partition from big to small and can be divided into 8 × 4,4 × 8,4 × 4；First the various tree piecemeals of coded macroblocks are found best matching blocks by MSE criterion, it is determined that the iterated function system coefficient of each piece and IFS coefficient；Then the rate distortion costs of all pieces of dividing mode is compared；Finally using block dividing mode minimum for rate distortion costs as final interframe encoding mode；Record final IFS coefficient, proceed to step 5 and obtain the reconstructed block of this block；If all of macro block of present frame is all encoded complete, all of reconstructed block composition rebuilds image, the i.e. reference frame of next frame, obtains coding side residual image by original image and the difference rebuilding image, forwards step 4 coded residual to；Described search window is the rectangular search region in reference frame；Described IFS coefficient includes the position of father's block and sub-block and offsets, namely motion vector (x, y) and scale factor s, displacement factor o；

Step 4: the data of residual image coefficient after DCT, quantization carries out Zig-Zag scanning on the one hand, is then encoded write code stream with entropy code CABAC；Decoding end residual frame is obtained on the other hand after inverse quantization, inverse DCT conversion；Reconstruction frames is obtained, i.e. the reference frame of next frame by prediction frame and decoding end residual frame sum；If all IFS coefficients then also to be carried out CABAC entropy code by P frame；Judge whether present frame is last frame, if last frame terminates coding；Otherwise, return step 2 and continue with next frame image；

Step 5: substitute into decoding equation by the IFS coefficient preserved

r_i=s d_i+o(1)

Calculating obtains predictive value, the difference of original block and prediction block obtain coding side residual block, and coding side residual block obtains decoding end residual block through the conversion of dct transform, quantization, inverse quantization and inverse DCT, then is obtained reconstructed block by prediction block and decoding end residual block sum；Proceed to step 3 coding depth graphic sequence next macro block of P frame；R in formula_iFor predicting the pixel value of block, d_iPixel value for the corresponding father's block of reference frame。

2. a kind of based on fractal and H.264 method for compressing high spectrum image according to claim 1, it is characterised in that: the video that high spectrum image cube metadata converts in described step one yuv format includes following two step:

1) hyperspectral image data is normalized to the integer in [0,255] interval；Hyperspectral image data represents the spectral reflectivity of atural object, and its bit depth is generally 16 bits/pixel, it is necessary to convert 8 bits/pixel to；Conversion method is that first the value of spectral reflectivity is normalized to [0,255] interval, is then converted into the integer without symbol 8 bit, and final numerical value is the integer that [0,255] is interval；

2) using each wave band of hyperspectral image data as a frame, the reflection coefficient after the normalization of each wave band is as the luminance elements Y-component of this frame, and chromatic component Cb and Cr is all set to intermediate value 128；Circulation performs step 2) until all wave bands of high spectrum image are disposed；Namely all frames form the video file of a yuv format。

3. a kind of based on fractal and H.264 method for compressing high spectrum image according to claim 1, it is characterised in that: described step 2 includes following five steps based on the H.264 fast intra-mode prediction mode discrimination method of macro block flatness:

1) calculate current coding macro block two-dimensional gray histogram, and record the maximum z of z-axis in two-dimensional gray histogram_max；

The two-dimensional gray histogram of image is based on the three-dimensional description figure that the Joint Distribution of image pixel gray level and neighborhood of pixels gray average is constituted；If image f (x, y) be sized to M × N, gray level is L；By f, (x, y) adopting k × k dot matrix to smooth the neighborhood averaging gray level image obtained is that (x y), is designated as g

In formula, 1≤x+m≤M, 1≤y+n≤N, k takes odd number；

Then g (x, image size y) and gray level and f (and x, y) identical；By f, (x, y) (x, (i, j), each two tuples belong to a point on two dimensional surface y) to may be constructed two tuples with g；Definition two-dimensional histogram be N (i, j), represent image f (x, y) grey scale pixel value is i, and when the gray value of neighborhood of pixels average gray image is j, two-dimensional points (i, number of times j) occurred, i, j=0,1 ..., L-1；Z-axis in two-dimensional gray histogram is N, and (i j), represents two-dimensional points (i, number of times j) occurred；

2) capping threshold value Th_highWith lower threshold Th_low, Th_highAnd Th_lowIt is the integer between [1,256]；

3) if z_max≥Th_high, it is believed that macro block is smooth, removes Intra4 × 4 patterns, selects Intra16 × 16 pattern, and using pattern minimum for rate distortion expense as optimal frames inner estimation mode；Update higher limit simultaneouslyOtherwise proceed to step 4)；

4) if z_max≤Th_low, it is believed that macro block details is enriched, and removes Intra16 × 16 patterns, selects Intra4 × 4 pattern, and using pattern minimum for rate distortion expense as optimal frames inner estimation mode；Update lower limit simultaneouslyOtherwise proceed to step 5)；

5) if Th_low<z_max<Th_high, it is believed that macro block flatness feature is not notable, adopts H.264 standard intraframe prediction algorithm。

4. a kind of based on fractal and H.264 method for compressing high spectrum image according to claim 1, it is characterised in that: in described step 3, P frame macro block fractal image includes following ten steps:

1) first in unit search window in reference frame, current macro is carried out Block-matching with 16 × 16 macro blocks；The position of sub-block is as the initiating searches point of father's block, and the size of father's block is identical with the size of sub-block；Travel through whole search window according to full search mode, find the minimum father's block position of matching error MSE；Record this position iterated function system coefficient and IFS coefficient, including motion vector (x, y), scale factor s, displacement factor o, and calculate the rate distortion expense cost1 of this pattern；

2) this 16 × 16 macro block is divided into the fritter of 2 16 × 8, each fritter is all traveled through corresponding whole search window according to full search mode, find the minimum corresponding father's block position of matching error MSE；The IFS coefficient of record the two position, and calculate the rate distortion expense sum cost2 of two blocks；

3) this 16 × 16 macro block is divided into the fritter of 28 × 16, each fritter is all traveled through corresponding whole search window according to full search mode, find the minimum corresponding father's block position of matching error MSE；The IFS coefficient of record the two position, and calculate the rate distortion expense sum cost3 of two blocks；

4) this 16 × 16 macro block is divided into the fritter of 48 × 8, each fritter is all traveled through corresponding whole search window according to full search mode, find the minimum corresponding father's block position of matching error MSE；Record the IFS coefficient of these 4 positions, and calculate the rate distortion expense cost4_1 of these 4 blocks, cost4_2, cost4_3, cost4_4 and four rate distortion expense sum cost4 respectively；

5) comparison step 1) to step 4) in rate distortion expense cost1, cost2, cost3, cost4, if minima is cost4, then needs to continue to divide by each 8 × 8 pieces further, forward step 6 to)；The dividing mode that otherwise rate distortion expense is minimum is the P frame encoding mode that this macro block is final, retains corresponding IFS coefficient, terminates；

6) it is first divided into 28 × 4 fritters by the 1st 8 × 8 pieces, each fritter is all traveled through corresponding whole search window according to full search mode, find the minimum corresponding father's block position of matching error MSE；The IFS coefficient of record the two position, and calculate the rate distortion expense sum cost5 of the two block；

7) it is divided into 24 × 8 fritters by the 1st 8 × 8 pieces, each fritter is all traveled through corresponding whole search window according to full search mode, find the minimum corresponding father's block position of matching error MSE；The IFS coefficient of record the two position, and calculate the rate distortion expense sum cost6 of the two block；

8) it is divided into 44 × 4 fritters by the 1st 8 × 8 pieces, each fritter is all traveled through corresponding whole search window according to full search mode, find the minimum corresponding father's block position of matching error MSE；Record the IFS coefficient of these 4 positions, and calculate the rate distortion expense sum cost7 of these 4 blocks；

9) compare cost4_1, cost5, cost6, cost7, using block dividing mode corresponding for minima therein as this coding mode 8 × 8 pieces final, retain corresponding IFS coefficient；

10) successively to 16 × 16 macro blocks each 8 × 8 pieces according to step 6) to step 9) and the final coding mode of method choice, retain corresponding IFS coefficient；

Wherein being calculated as follows of matching error MSE:

In formula, the calculating formula of s, o is respectively as follows:

Wherein, r_iIt is each point pixel value of current block, d_iBeing each point pixel value of match block, N is the number of pixels of current block, and s is scale factor, and o is displacement factor；

Being calculated as follows of rate distortion expense:

Cost=MSE (R, D)+λ_MODE·Rate(x,y,s,o,MODE|QP)(6)

Wherein, R represents current block, and D represents match block, and MODE represents macro block dividing mode, and QP represents quantization parameter, λ_MODEIt is the LaGrange parameter relevant with QP。