CN101710985A - Image brightness compensation method for image coding - Google Patents

Abstract

The invention discloses an image brightness compensation method for image coding, belongs to the field of image processing, and solves the problems of complexity, low speed and low compression efficiency of the conventional algorithm for performing brightness compensation for multi-viewpoint video coded images. The image brightness compensation method comprises the following steps: determining brightness difference between a current block P0 and a reference block PS by comparing the brightness difference between a current block template X0 and a reference block template XS, thereby realizing the brightness compensation for the current block; and utilizing a series of solved candidate weight matrixes B alpha, solving the optimal Bj and assigning the optimal Bj to weight matrixes B of the reference block PS, thereby solving the brightness difference between the current block template X0 and the reference block template XS according to B(X0T-XST), which serves as the brightness difference between the current block P0 and the reference block PS for brightness compensation.

Description

A kind of image brightness compensation method that is used for image encoding

Technical field

The present invention relates to a kind of image brightness compensation method that is used for image encoding, belong to image processing field.

Background technology

Because the progress that video obtains technology and video display technology, 3 D video is accepted by the consumer gradually with its distinctive advantage.By 3 D video technology people can be mutual the selection viewpoint, thereby the scene that can remove to observe real world with the angle that people like.Usually, 3 D video represents by many viewpoints, and each viewpoint is one and traditional passes through the two dimensional video sequence that video camera obtains.Multiple view video coding technology (MVC) is the key technology that adopts in the 3 D video transmission system that many viewpoints represent, can effectively reduce the computation complexity of transmission bandwidth and decoding end.

In the multiple view video coding system, because the influence of camera parameter and the variation of camera position and angle, the brightness value of same target object in different points of view can change.When carrying out predicting between the looking of multiple view video coding, these variations can cause mate in the difference estimation process inaccurate, cause the decline of code efficiency.People adopt the luminance compensation technology to solve difference estimation on the encoding and decoding aspect to mate inaccurate problem, promptly remove the luminance difference between different points of view in the difference estimation process.This luminance difference is called as local brightness variation.In decoding end, according to the difference predicted value, residual values and brightness changing value sum are reconstructed the piece that has carried out luminance compensation:

I(i，j)＝R(i，j)+r(i+x，j+y)+C，

Wherein ((i j) is residual signals to R to I, and (i+x j+y) is reference block to r, and C is the brightness changing value of this piece for i, j) expression reconstructed blocks.

Luminance difference between the different points of view that exists in the multiple view video coding has caused influence to coding, and at present, the algorithm that the image of multiple view video coding is carried out luminance compensation is too complicated, and speed is slow, and compression efficiency is low.

Summary of the invention

The present invention seeks to that to carry out the algorithm of luminance compensation too complicated in order to solve at present image to multiple view video coding, speed is slow, and the problem that compression efficiency is low provides a kind of image brightness compensation method that is used for image encoding.

Method of the present invention may further comprise the steps:

The a certain macro block of definition current frame image is current block P ₀, in present frame with described current block P ₀Adjacent L shaped neighborhood is current block template X ₀, current block template brightness value matrix is X ₀, current block template brightness value matrix is X ₀Transposed matrix be X ₀ ^T,

Definition former frame image and current block P ₀The macro block that is in same position is reference block P _S, in the former frame image with described reference block P _SAdjacent L shaped neighborhood is reference block template X _S, reference block P _SThe brightness value matrix be P _S, reference block template brightness value matrix is X _S, reference block template brightness value matrix X _STransposed matrix be X _S ^T,

Step 1, with reference block template X _SIn all pixels and reference block P _SCorrelation normalization form a series of candidate's weight matrix B _α:

$B_{\mathrm{\α}}=(\frac{l_{\mathrm{\α},0}}{\underset{i=0}{\overset{m-1}{\mathrm{\Σ}}}{l}_{\mathrm{\α},i}},\frac{l_{\mathrm{\α},1}}{\underset{i=0}{\overset{m-1}{\mathrm{\Σ}}}{l}_{\mathrm{\α},i}},...,\frac{l_{\mathrm{\α},m-1}}{\underset{i=0}{\overset{m-1}{\mathrm{\Σ}}}{l}_{\mathrm{\α},i}})$

Reference block template X _SIn m pixel altogether, each pixel on the angle [alpha] direction to reference block P _SCorrelation be expressed as l respectively _{α, 0}, l _{α, 1}..., l _{α, m-1},

Wherein, α is reference block template X _SIn pixel and reference block P _SThe angle that intersects is the value of Discrete Change, and the Discrete Change amount of α is 5 °～20 °, 0 °≤α≤180 °,

Step 2, obtain reference block P _SWeight matrix B=B _j, wherein, parameter j is determined by following formula:

$j=\arg \underset{\mathrm{\α}}{\min }|P_S-B_{\mathrm{\α}}{X_S}^{T}E|,$

Step 3, according to B (X ₀ ^T-X _S ^T) E acquisition current block P ₀With reference block P _SBetween luminance difference, to current block P ₀Carry out luminance compensation, wherein E is 1 complete 1 matrix for value, and its exponent number and reference block P _SThe brightness value matrix be P _SExponent number identical.

Advantage of the present invention: according to the luminance correlation of adjacent area in the video image, utilize the brightness of adjacent area to change the brightness variation of deriving between current block and the reference block, thereby realize luminance compensation to current block, algorithm of the present invention is simple, the speed of service is fast, has improved the multiple view video coding compression efficiency.

Description of drawings

Fig. 1 is the inventive method flow chart, Fig. 2 is current frame image distribution structure figure, Fig. 3 is former frame image distribution structure figure, and Fig. 4 is the correlation schematic diagram of pixel and reference block in the reference block template in the former frame image, and Fig. 5 utilizes the Race1 sequence to carry out the experimental result picture of emulation.

Embodiment

Embodiment one: below in conjunction with Fig. 1 to Fig. 5 present embodiment is described, present embodiment may further comprise the steps:

The a certain macro block of definition current frame image is current block P ₀, in present frame with described current block P ₀Adjacent L shaped neighborhood is current block template X ₀, current block template brightness value matrix is X ₀, current block template brightness value matrix is X ₀Transposed matrix be X ₀ ^T,

Definition former frame image and current block P ₀The macro block that is in same position is reference block P _S, in the former frame image with described reference block P _SAdjacent L shaped neighborhood is reference block template X _S, reference block P _SThe brightness value matrix be P _S, reference block template brightness value matrix is X _S, reference block template brightness value matrix X _STransposed matrix be X _S ^T,

Step 1, with reference block template X _SIn all pixels and reference block P _SCorrelation normalization form a series of candidate's weight matrix B _α:

$B_{\mathrm{\α}}=(\frac{l_{\mathrm{\α},0}}{\underset{i=0}{\overset{m-1}{\mathrm{\Σ}}}{l}_{\mathrm{\α},i}},\frac{l_{\mathrm{\α},1}}{\underset{i=0}{\overset{m-1}{\mathrm{\Σ}}}{l}_{\mathrm{\α},i}},...,\frac{l_{\mathrm{\α},m-1}}{\underset{i=0}{\overset{m-1}{\mathrm{\Σ}}}{l}_{\mathrm{\α},i}})---\left(1\right)$

Reference block template X _SIn m pixel altogether, each pixel on the angle [alpha] direction to reference block P _SCorrelation be expressed as l respectively _{α, 0}, l _{α, 1}..., l _{α, m-1},

Wherein, α is reference block template X _SIn pixel and reference block P _SThe angle that intersects is the value of Discrete Change, and the Discrete Change amount of α is 5 °～20 °, 0 °≤α≤180 °,

Step 2, obtain reference block P _SWeight matrix B=B _j, wherein, parameter j is determined by following formula:

$j=\arg \underset{\mathrm{\α}}{\min }|P_S-B_{\mathrm{\α}}{X_S}^{T}E|---\left(2\right)$

Step 3, according to B (X ₀ ^T-X _S ^T) E acquisition current block P ₀With reference block P _SBetween luminance difference, to current block P ₀Carry out luminance compensation, wherein E is 1 complete 1 matrix for value, and its exponent number and reference block P _SThe brightness value matrix be P _SExponent number identical.

According to the continuity of adjacent area pixel brightness value in the video image, as shown in Figure 2, we define current block P ₀L shaped neighborhood X ₀Be the current block template, and current block template X ₀Brightness value and current block P ₀Brightness value between have stronger correlation.

Same, reference block P _SWith reference block template X _SBetween also have stronger correlation.

Therefore, method of the present invention is exactly by comparing current block template X ₀With reference block template X _SBetween luminance difference determine current block P ₀With reference block P _SBetween luminance difference.The establishment of formula (3) is promptly arranged.

Diff(P ₀，P _S)≈Diff(X ₀，X _S) (3)

P wherein ₀, P _S, X ₀And X _SBe the brightness value matrix.When calculating luminance difference, we think the luminance difference value basically identical in the piece, and the brightness value in the template is got weighted sum.That is:

P ₀≈P _s+(AX ₀ ^T-BX _S ^T)E (4)

AX wherein ₀ ^T-BX _S ^TBe the luminance difference between two pieces estimating to obtain according to template, E is 1 complete 1 matrix for value, current block P ₀Weight matrix A and reference block P _SWeight matrix B obtain according to the direction character of each pixel.Like this, in difference estimation and difference compensation process, can make up one and compensate the reference block that brightness changes, reference block P _SThe matrix that forms after the compensate for brightness is:

P _S′＝P _S+(AX ₀ ^T-BX _S ^T)E (5)

Improve the code efficiency of prediction between looking, improved the multiple view video coding compression efficiency.

Further, current block P ₀With reference block P _SThe direction character unanimity, A in the formula (5) and the value of B equate A=B in theory substantially, therefore, we only need obtain reference block P _SWeight matrix B get final product, and formula (5) is reduced to:

P _S′＝P _S+B(X ₀ ^T-X _S ^T)E (6)

P _S' back by way of compensation current block P ₀Brightness value matrix P ₀, that is:

P ₀＝P _S+B(X ₀ ^T-X _S ^T)E (7)

Introduce its computational process below in conjunction with Fig. 4:

The key of luminance compensation is whether luminance difference calculating is correct, and current block template X ₀With reference block template X _SWhen determining, luminance difference is determined by the luminance compensation weights.The present invention proposes and a kind ofly determine candidate's weights of template according to direction, and according to reference block P _SIn direction character derive current block P ₀Direction character.

For reference block template X _SCertain interior pixel p defines its coordinate for (x y), as shown in Figure 4, puts ray and reference block P that p sends _SWhat may intersect has a lot of bars, the direction character α that corresponding ray is different, its length of shining upon in piece is l, then defining point p on direction α to reference block P _SCorrelation be l _{α, p}, reference block template X _SIn m pixel (from 0 to m) arranged, they and reference block P _SCorrelation be respectively l _{α, 0}, l _{α, 1}..., l _{α, m-1}, on the α direction, with all reference block template X _SThe correlation normalization of middle m pixel obtains each candidate's weight matrix corresponding on the α direction:

$B_{\mathrm{\α}}=(\frac{l_{\mathrm{\α},0}}{\underset{i=0}{\overset{m-1}{\mathrm{\Σ}}}{l}_{\mathrm{\α},i}},\frac{l_{\mathrm{\α},1}}{\underset{i=0}{\overset{m-1}{\mathrm{\Σ}}}{l}_{\mathrm{\α},i}},...,\frac{l_{\mathrm{\α},m-1}}{\underset{i=0}{\overset{m-1}{\mathrm{\Σ}}}{l}_{\mathrm{\α},i}}),$

α is reference block template X _SIn pixel can with reference block P _SThe angle that intersects is the value of Discrete Change, and the Discrete Change amount of α is 5 °～20 °, 0 °≤α≤180 °,

Like this, just try to achieve a series of candidate's weight matrix B _α, we select optimum weight matrix according to formula (2), can make | P _S-B _αX _S ^TThe α assignment of E| minimum is given parameter j, again with B _jAssignment is given reference block P _SWeight matrix B, like this, just can be according to B (X ₀ ^T-X _S ^T) E obtains current block template X ₀With reference block template X _SBetween luminance difference, and as current block P ₀With reference block P _SBetween luminance difference, be used for luminance compensation.

Current block template brightness value matrix X ₀, reference block template brightness value matrix is X _SWith reference block P _SWeight matrix B all be the matrix of 1 row m row, current block template brightness value transpose of a matrix matrix is X ₀ ^TBe the matrix of capable 1 row of m, then B (X ₀ ^T-X _S ^T) be a number, E multiplies each other with matrix, obtains exponent number and reference block P _SIdentical matrix is so that by formula (7) are to reference block P _SLuminance compensation is carried out in addition.

The present invention is applicable to video compression system, includes but not limited to H.264/AVC, AVS, VC-1 etc.

Embodiment two: the difference of present embodiment and execution mode one is, is divided into 10 °～18 ° mutually between adjacent two discrete points of angle [alpha], and other is identical with execution mode one.

Fig. 5 has provided the rate distortion comparative result on the Race1 sequence, and wherein transverse axis is represented the bit number of average every frame, and the longitudinal axis is represented the Y-PSNR (PSNR) of average every frame.MVC is not for adding the result under the brightness compensation, and MVC+ICT (Illumination Compensation Template) is for having added the result based on the luminance compensation of template.As can be seen from the figure, the present invention has a significant effect to the compression efficiency that improves multiple view video coding.

Embodiment three: the difference of present embodiment and execution mode one is, is divided into 15 ° mutually between adjacent two discrete points of angle [alpha], and other is identical with execution mode one.

Present embodiment provides a specific embodiment, be divided into 15 ° mutually between adjacent two discrete points of angle [alpha], then the angle [alpha] value is 0,15,30,45,60,75,90,105,120,135,150,165, according to these discrete numerical value, tries to achieve 12 candidate's weight matrix B _αValue is for use.

Embodiment four: the difference of present embodiment and execution mode one is, is divided into 18 ° mutually between adjacent two discrete points of angle [alpha], and other is identical with execution mode one.

Present embodiment provides a specific embodiment, be divided into 18 ° mutually between adjacent two discrete points of angle [alpha], then the angle [alpha] value is 0,18,36,54,72,90,108,126,144,162, according to these discrete numerical value, tries to achieve 10 candidate's weight matrix B _αValue is for use.

Claims (5)

1. an image brightness compensation method that is used for image encoding is characterized in that, this method may further comprise the steps:

The a certain macro block of definition current frame image is current block P ₀, in present frame with described current block P ₀Adjacent L shaped neighborhood is current block template X ₀, current block template brightness value matrix is X ₀, current block template brightness value matrix is X ₀Transposed matrix be X ₀ ^T,

Definition former frame image and current block P ₀The macro block that is in same position is reference block P _S, in the former frame image with described reference block P _SAdjacent L shaped neighborhood is reference block template X _S, reference block P _SThe brightness value matrix be P _S, reference block template brightness value matrix is X _S, reference block template brightness value matrix X _STransposed matrix be X _S ^T,

Step 1, with reference block template X _SIn all pixels and reference block P _SCorrelation normalization form a series of candidate's weight matrix B _α:

$B_{\mathrm{\α}}=(\frac{l_{\mathrm{\α},0}}{\underset{i=0}{\overset{m-1}{\mathrm{\Σ}}}{l}_{\mathrm{\α},i}},\frac{l_{\mathrm{\α},1}}{\underset{i=0}{\overset{m-1}{\mathrm{\Σ}}}{l}_{\mathrm{\α},i}},...,\frac{l_{\mathrm{\α},m-1}}{\underset{i=0}{\overset{m-1}{\mathrm{\Σ}}}{l}_{\mathrm{\α},i}})$

Reference block template X _SIn m pixel altogether, each pixel on the angle [alpha] direction to reference block P _SCorrelation be expressed as l respectively _{α, 0}, l _{α, 1}..., l _{α, m-1}, m is the natural number greater than 1,

Wherein, α is reference block template X _SIn pixel can with reference block P _SThe angle that intersects is the value of Discrete Change, and the Discrete Change amount of α is 5 °～20 °, 0 °≤α≤180 °,

Step 2, obtain reference block P _SWeight matrix B=B _j, wherein, parameter j is determined by following formula:

$j=\arg \underset{\mathrm{\α}}{\min }|P_S-B_{\mathrm{\α}}{X_S}^{T}E|,$

Step 3, according to B (X ₀ ^T-X _S ^T) E acquisition current block P ₀With reference block P _SBetween luminance difference, to current block P ₀Carry out luminance compensation, wherein E is 1 complete 1 matrix for value, and its exponent number and reference block P _SThe brightness value matrix be P _SExponent number identical.

2. a kind of image brightness compensation method that is used for image encoding according to claim 1 is characterized in that current block P ₀Carry out luminance compensation, compensation back current block P ₀Brightness value matrix P ₀For:

P ₀＝P _S+B(X ₀ ^T-X _S ^T)E。

3. a kind of image brightness compensation method that is used for image encoding according to claim 1 is characterized in that, is divided into 10 °～18 ° mutually between adjacent two discrete points of angle [alpha].

4. a kind of image brightness compensation method that is used for image encoding according to claim 1 is characterized in that, is divided into 15 ° mutually between adjacent two discrete points of angle [alpha].

5. a kind of image brightness compensation method that is used for image encoding according to claim 1 is characterized in that, is divided into 18 ° mutually between adjacent two discrete points of angle [alpha].

Images