CN103124347A - Method for guiding multi-view video coding quantization process by visual perception characteristics - Google Patents

Abstract

The present invention relates to a method for guiding the coding and quantization process by using visual perception characteristics. The operation steps of this method are as follows: (1) read the luminance value of each frame of the input video sequence, and set up a just discernible distortion threshold model in the frequency domain, (2) each frame of the input video sequence passes through the intra-viewpoint and the inter-viewpoint Prediction, (3) perform discrete cosine transform on the residual data, (4) dynamically adjust the quantization step size of each macroblock in the current frame, (5) dynamically adjust the Lagrangian parameters in the rate-distortion optimization process, (6 ) performs entropy encoding on the quantized data to form a code stream and transmit it through the network. The present invention improves the video compression efficiency under the condition that the subjective quality remains basically unchanged, and is more suitable for transmission in the network.

Description

A method to guide the quantization process of multi-view video coding by using the characteristics of visual perception

technical field

The invention relates to the technical field of multi-viewpoint video coding and decoding, in particular to a method for guiding the multi-viewpoint video coding and quantization process by using visual perception characteristics, which is suitable for high-definition 3D video signal coding and decoding.

Background technique

With the development of the times, people have higher and higher requirements for audio-visual experience, and are not satisfied with the existing single-view two-dimensional video. People have higher and higher requirements for stereoscopic experience, and stereoscopic perception from a fixed angle can be experienced from any angle, thus giving birth to the development of multi-viewpoint coding technology. However, the data required by multi-viewpoints is greatly increased, and how to effectively improve the video compression efficiency has become a research hotspot. At present, video compression technology mainly focuses on three aspects of removing spatial redundancy, temporal redundancy and statistical redundancy. Although video experts have launched a new generation of video compression coding technology (HEVC), it is expected that the video compression efficiency will be doubled on the basis of H.264. However, due to the characteristics of the human visual system (HVS), there is perceptual redundancy that has not been removed. With the gradual deepening of the research on the visual characteristics of the human eye, some video workers have proposed a Just Noticeable Distortion (JND) model that removes the redundancy of the human eye. That is, measure the size of perceptual redundancy according to the obtained JND threshold, and when the change value is lower than this threshold, it will not be perceived by human eyes.

At present, the research on JND is mainly divided into two categories: pixel-domain JND and frequency-domain JND models. Among them, the JND model proposed in the literature [1] is a classic pixel domain model, which studies the characteristics of brightness masking, texture masking and time domain masking respectively. The frequency-domain JND model proposed in [2] not only studies the first three characteristics, but also studies the sensitivity of the human eye to different frequency bands, which makes the frequency-domain JND model more in line with the visual characteristics of the human eye.

Aiming at the JND model proposed in the literature [2], it is a relatively complete DCT domain JND model at present. In addition to including the brightness masking feature and texture masking feature of the pixel, it also adds the effect of the spatial sensitivity function. The spatial sensitivity function reflects the bandpass characteristics of the human eye, and removes the frequency components that the human eye cannot perceive to achieve the purpose of removing the frequency redundancy of the human eye. In the time-domain masking effect, the smooth eyeball movement effect is included, which includes not only the magnitude of the motion amplitude, but also the direction information of the motion. Some researchers combined it with multi-view video and applied it to the residual DCT transform (discrete cosine transform), which greatly improved the compression efficiency. However, it is not used in other coding processes such as quantization, so it is not thorough enough to remove visual redundancy.

The JND model established in the literature [3], although proposed to use the JND model to guide the quantization process. However, the JND model established by it is in the pixel domain, and lacks the process of removing the frequency redundancy of the human eye, resulting in inaccurate guidance and quantization. Secondly, the subjective quality is guaranteed for the JND model, and the quantization value only needs to be adjusted in places where the human eye is not sensitive, while the quantization value of other areas remains unchanged. Finally, while adjusting the quantization parameters, the Lagrangian parameters are correspondingly adjusted.

The patent application of the present invention proposes for the first time that the DCT domain JND model is applied to the quantization process of multi-viewpoint video coding, so as to further improve the video compression efficiency while keeping the subjective quality unchanged.

Literature [1]: X. Yang, W. Lin, and Z. Lu, “Motion-compensated residue preprocessing in video coding based on just-noticeable-distortion profile,” IEEE Trans. Circuits Syst. Video Technol., vol. 15 , no. 6, pp. 742–752, 2005.

Literature [2]: Zhenyu Wei and King N. Ngan., "Spatio-Temporal Just Noticeable Distortion Profile for Gray Scale Image/Video in DCT Domain." IEEE transactions on circuits and systems for video technology. VOL. 19, NO. 3 , March 2009.

Literature [3]: Z. Chen and C. Guillemot, "Perceptually friendly H.26 /AVC video coding based on foveated just noticeable distortion model," IEEE Trans. Circuits Syst. Video Technol., vol. 20, no. 6, pp. 806–819, Jun. 2010.

Contents of the invention

The purpose of the present invention is to provide a method for guiding the quantization process of multi-viewpoint video coding by using visual perception characteristics to solve the defects in the prior art. The method uses the frequency domain JND model to guide multi-viewpoint video coding while ensuring that the subjective quality of the video remains unchanged. During the viewpoint quantization process, the quantization step size is increased in areas that are not sensitive to human eyes, and the video compression efficiency is improved. While adjusting the step size, the Lagrangian parameter of the rate-distortion optimization function is dynamically adjusted, so that the coding efficiency is further improved.

To achieve the above object, the present invention adopts the following technical solutions:

A method of using visual perception characteristics to guide the quantization process of multi-viewpoint video coding, characterized in that the operation steps are as follows:

(1) Read the brightness value of each frame of the input video sequence, and establish a just discernable distortion threshold model in the frequency domain,

(2) Each frame of the input video sequence undergoes intra-viewpoint and inter-viewpoint prediction,

(3) Discrete cosine transform (DCT transform) is performed on the residual data,

(4) Dynamically adjust the quantization step size of each macroblock in the current frame,

(5) Lagrangian parameters in the dynamic adjustment rate-distortion optimization process,

(6) Entropy encoding is performed on the quantized data to form a code stream for transmission through the network.

Compared with the prior art, the method of using visual perception characteristics to guide the quantization process of multi-viewpoint video encoding in the present invention has the following obvious outstanding substantive features and significant technological progress:

1) The multi-view video coding method can reduce the coding bit rate through the subroutine quantization in the coding process while ensuring the quality of the reconstructed video. In the experiment, the maximum bit rate can be reduced to 12.35%;

2) This multi-viewpoint video coding method uses the average subjective score difference while ensuring the quality of the reconstructed video. When the subjective score difference is close to 0, it means that the subjective quality of the two methods is closer, and the average subjective score of this method is closer to 0. The score difference is 0.03, so the subjective quality of the present invention is equivalent to the subjective quality of multi-viewpoint video codec JMVC code;

3) The multi-viewpoint video encoding method does not add a particularly complicated encoding process, and improves video encoding and compression efficiency with less complexity.

Description of drawings

FIG. 1 is a functional block diagram of the method for guiding the quantization process of multi-viewpoint video coding by using visual perception characteristics in the present invention.

Figure 2 is a block diagram of a just discernible distortion model in the frequency domain.

3 is a block diagram of intra/inter-view prediction.

Fig. 4 is a block diagram of DCT transformation.

Fig. 5 is a block diagram of dynamically adjusting the quantization step size.

6 is a block diagram of dynamically adjusting Lagrange parameters in a rate-distortion cost function.

Figure 7 is a block diagram of an entropy encoded output.

Fig. 8a is the reconstructed image of the 15th frame image of the 0th viewpoint of the video sequence ballroom using the JMVC original coding method.

Fig. 8b is the reconstructed image of the 15th frame image of the 0th viewpoint of the video sequence ballroom using the method of the present invention.

Figure 9 is the comparison result of the video sequence ballroom using the JMVC original encoding method and the method of the present invention under different QP and different viewpoints, the bit rate, PSNR value, and the difference in subjective quality evaluation score (DMOS) of the reconstructed video

Fig. 10a is the reconstructed image of the 35th frame image of the first viewpoint of the video sequence race1 using the JMVC original coding method.

Figure 10b is the reconstructed image of the 35th frame image of the first viewpoint of the video sequence race1 using the method of the present invention

Fig. 11 is the comparison result of video sequence race1 using JMVC original coding method and the method of the present invention under different QP and different viewpoints, bit rate, PSNR value, reconstructed video subjective quality evaluation score difference (DMOS)

Fig. 12a is the reconstructed image of the 45th frame image of the second viewpoint of the video sequence Crowd using the JMVC original encoding method.

Fig. 12b is the reconstructed image of the 45th frame image of the second viewpoint of the video sequence Crowd using the method of the present invention.

Figure 13 shows the comparison results of video sequence Crowd using the JMVC original coding method and the method of the present invention under different QP and different viewpoints, bit rate, PSNR value, and reconstructed video average subjective score difference (DMOS).

Detailed ways

Below in conjunction with accompanying drawing, preferred embodiment of the present invention is described in further detail:

Embodiment one:

In this embodiment, the method of using visual perception characteristics to guide the quantization process of multi-viewpoint video coding, as shown in FIG. 1, includes the following steps:

(1) Read the brightness value of each frame of the input video sequence, and establish a just discernable distortion threshold model in the frequency domain,

(2) Each frame of the input video sequence undergoes intra-viewpoint and inter-viewpoint prediction,

(3) Discrete cosine transform is performed on the residual data,

(4) Dynamically adjust the quantization step size of each macroblock in the current frame,

(5) Lagrangian parameters in the dynamic adjustment rate-distortion optimization process,

(6) Entropy encoding is performed on the quantized data to form a code stream for transmission through the network.

Embodiment 2: This embodiment is basically the same as Embodiment 1, and the special features are as follows:

The establishment of the frequency domain JND model in the above step (1) includes four models, see Figure 2:

其基本的JND阈值可表示为： (1-1) The spatial contrast sensitivity function model is based on the bandpass characteristic curve of the human eye, for a specific spatial frequency

Its basic JND threshold can be expressed as:

的计算公式为： spatial frequency

The calculation formula is:

和表示DCT变换块的坐标位置，

为DCT变换块的维数，

和

表示水平和垂直的视角，一般认为水平视角等于垂直视角，其表达为： in,

and Indicates the coordinate position of the DCT transform block,

is the dimension of the DCT transform block,

and

Indicates the horizontal and vertical viewing angles. It is generally believed that the horizontal viewing angle is equal to the vertical viewing angle, which is expressed as:

Since the visual sensitivity of the human eye is directional, it is more sensitive to horizontal and vertical directions, and less sensitive to other directions. Add the modulation factor of the direction to get:

为DCT系数向量所代表的频率的角度，

为DCT系数归一化因子表达式为：

is the angle of the frequency represented by the DCT coefficient vector,

The normalization factor expression for the DCT coefficient is:

形成最终的空间灵敏度函数的调制因子为： Finally add the control parameters

The modulation factor forming the final spatial sensitivity function is:

为0.6，

为1.33，

为0.11，为0.18；对于4×4块尺寸的DCT编码格式，

为0.6，

为0.8，

为0.035，

为0.008。 In the process of multi-view coding, the parameters are different due to the existence of 8×8 and 4×4 DCT transforms. In the experiment, for the DCT coding format with 8×8 block size,

is 0.6,

is 1.33,

is 0.11, is 0.18; for the DCT coding format of 4×4 block size,

is 0.6,

is 0.8,

is 0.035,

is 0.008.

(1-2) The brightness masking effect model is based on experiments. The human visual perception sensitivity is more sensitive in the middle gray value area than in the darker and brighter background areas. Finally, the brightness masking effect curve is fitted, and its expression for:

是当前编码块的平均亮度值。 in

is the average luminance value of the current coded block.

(1-3) The texture masking effect model divides the image into three areas according to the texture of the image: boundary area, smooth area and texture area. The human eye becomes less sensitive to it in turn. The canny operator is usually used to separate the various regions of the image.

The edge pixel density obtained by using the canny operator is as follows:

是块的边缘像素总数，由Canny边缘检测器获得。 in,

is the total number of edge pixels of the block, obtained by the Canny edge detector.

Use edge pixel density The image block is divided into flat area, texture area and edge area, and the basis formula for image block classification is as follows:

For textured areas, the eye is not sensitive to distortion in the low frequency part, but the high frequency part is properly preserved. Therefore, the estimated factor for contrast masking is:

）是DCT系数标号。 in(

) is the DCT coefficient label.

Due to the overlapping effect of the spatial contrast sensitivity function effect and the brightness effect, the final masking effect factor is obtained as:

表示输入视频序列的第

帧，

为DCT系数，

为空间对比度灵敏度函数的阈值，

为亮度掩盖效应特性调制因子。 in,

Indicates the first step of the input video sequence

frame,

is the DCT coefficient,

is the threshold of the spatial contrast sensitivity function,

Modulation factor for the brightness masking effect characteristic.

(1-4) The time-contrast sensitivity function model is based on the experimentally measured modulation factor of the time-domain masking effect:

表示空间频率。时间频率

其一般计算公式如下： in, represents the time frequency,

represents the spatial frequency. time frequency

Its general calculation formula is as follows:

分别为空间频率的水平和垂直分量，

为视网膜上物体运动的速度。

are the horizontal and vertical components of the spatial frequency, respectively,

is the velocity of the object moving on the retina.

的计算式为：

The calculation formula is:

和

表示像素水平和垂直的视角，

为DCT变换维数，

和

表示DCT变换块的坐标位置。 in,

and

Indicates the horizontal and vertical viewing angle of the pixel,

For the DCT transformation dimension,

and

Indicates the coordinate position of the DCT transform block.

计算方法如下： speed of images on retina

The calculation method is as follows:

是平滑跟踪眼球移动效应增益，实验中取0.98。

表示物体在图像平面的速度，

表示由于漂移运动引起的最小的眼球移动速度，其经验值为0.15.deg/s。

是和眼睛跳跃运动相对应的眼球的最大速度，通常取80deg/s，

是视频序列的帧率。

是每个块的运动矢量，是像素的视角。 in,

is the smooth tracking eye movement effect gain, which is 0.98 in the experiment.

Indicates the velocity of the object in the image plane,

Indicates the minimum eye movement speed caused by drifting motion, and its empirical value is 0.15.deg/s.

is the maximum speed of the eyeball corresponding to the eye jumping movement, usually 80deg/s,

is the frame rate of the video sequence.

is the motion vector of each block, is the viewing angle of the pixel.

(1-5) The weighted product of the four factors constitutes the just discernible distortion threshold of the current coded frame, and its expression is:

为空间对比度灵敏度函数的阈值，

为亮度掩盖效应调制因子，

为掩盖效应调制因子，

为时域掩盖调制因子。 in,

is the threshold of the spatial contrast sensitivity function,

is the brightness masking effect modulation factor,

is the masking effect modulation factor,

Modulation factor for the time-domain mask.

Above-mentioned step (2) is to carry out inter/intra-viewpoint prediction to input video sequence, referring to Fig. 3, its specific steps are as follows:

(2-1) Intra-view inter/intra prediction is to remove the temporal redundancy of the current frame through intra-view inter prediction, and remove the spatial redundancy of the current frame through intra-view intra prediction. The prediction method with the smallest rate-distortion optimization function is selected in intra-frame prediction and inter-frame prediction. The expression of the rate-distortion optimization function is:

为失真信号，

为不同编码模式下编码的比特数，

是调整后的拉格朗日参数。 in

is a distorted signal,

is the number of bits encoded in different encoding modes,

is the adjusted Lagrange parameter.

(2-2) Inter-viewpoint prediction is performed because this method encodes multiple viewpoints, and predicts the current frame through corresponding frames between viewpoints, which can remove redundant information between viewpoints.

(2-3) Compare the encoding cost between views and within views, select the best prediction method in intra-view prediction and compare it with the prediction method between views, and choose the prediction method with the smallest rate-distortion optimization cost function as the best prediction method . Fully consider the redundant characteristics between views and within views, and select the appropriate prediction method to further improve the video compression efficiency.

Above-mentioned step (3) carries out discrete cosine transform to residual data, referring to Fig. 4, its concrete steps are as follows:

七种情况，前四种归结为

变换块，后三种为变换块。 (3-1) Judgment of the coding block size, in the multi-view coding method, the coding block size has

Of the seven cases, the first four boil down to

transform block, the last three are Transform blocks.

DCT变换，对于

变换块采用

DCT变换。 (3-2) The corresponding DCT transformation, for The transform block uses

DCT transform, for

The transform block uses

DCT transformation.

Above-mentioned step (4) dynamically adjusts the quantization step size of each macroblock in the current frame, referring to Fig. 5, its specific steps are as follows:

(4-1) Calculate the average JND value of the current frame through the established JND model, and the average JND threshold is:

和

分别表示图像帧的高度和宽度，

表示当前帧的恰可辨失真阈值，

表示像素的坐标。 in,

and

represent the height and width of the image frame, respectively,

Indicates the just discernible distortion threshold of the current frame,

Represents the coordinates of a pixel.

(4-2) The JND mean value of the current macroblock, the average JND threshold value of the Mth macroblock is expressed as:

(4-3) Dynamically adjust the quantization step size of the current macroblock. The just discernible distortion threshold reflects the sensitivity of the human eye to each part of an image, so each part can be dynamically adjusted according to the difference of the just discernible distortion threshold. The quantization step size of the macroblock. For places where the human eye is not sensitive, the quantization step size should be increased appropriately, otherwise, the quantization value will remain unchanged. The proposed quantization parameter tuning is:

是编码框架原有的步长，

为调节因子，其表达式由下式给出： in,

is the original step size of the encoding frame,

is the adjustment factor whose expression is given by:

。 in,

.

The above step (5) dynamically adjusts the Lagrangian parameters in the rate-distortion optimization process, see Figure 6, and its specific operation steps are as follows:

(5-1) Calculate and compare the average JND value of the current frame and the average JND value of the current coded macroblock to provide a basis for the weighting of the Lagrangian parameters in the next step.

为： (5-2) Adjust the Langrange parameters. The quantization parameters have been adjusted before, and the distortion value and bit rate in the Lagrangian rate-distortion optimization will change. At this time, the original Lagrangian parameters will be used. value, it cannot be guaranteed to be the optimal solution. At the same time, corresponding to the weighted Lagrangian parameters, the cost function can be optimized again, and the adjusted

for:

表示多视点编码方法内生成的量化参数，

表示第个宏块调整后的量化参数值。 in,

denotes the quantization parameter generated within the multi-view coding method,

Indicates the first The adjusted quantization parameter value of each macroblock.

(5-3) Substituting the adjusted Lagrangian parameters into the rate-distortion optimization cost function, the expression is as follows:

为失真信号，

为不同编码模式下编码的比特数，

是调整后的拉格朗日参数。这样使得在量化参数改变的同时，相应改变拉格朗日参数，使得率失真优化函数依然得到最优解。 in

is a distorted signal,

is the number of bits encoded in different encoding modes,

is the adjusted Lagrange parameter. In this way, when the quantization parameter is changed, the Lagrangian parameter is correspondingly changed, so that the rate-distortion optimization function still obtains an optimal solution.

The above step (6) entropy-encodes the quantized data to form a code stream for transmission through the network, see Figure 7, the specific steps are as follows:

(6-1) Entropy encoding is performed on the quantized data, so that the quantized data can be most effectively represented by the binary code stream, and the statistical redundancy of the quantized data is removed.

(6-2) Transmit the code stream formed by entropy coding through the network to realize video transmission. The encoding method processed by visual perception characteristics can better adapt to network transmission because of its small bandwidth occupation.

A large number of simulation experiments are carried out below to evaluate the performance of the multi-view video coding method proposed in this paper using visual characteristics. On a PC configured with Intel Pentium 4 CPU 3.00GHz, 512M Internal Memory, Intel 8254G Express Chipset Family, and Windows XP Operation System, the first 48 frames of multi-viewpoint video sequences ballroom, race1, and crowd were encoded and decoded, among which BASIC QP was set to 20 , 24, 28, 32. The multi-view video codec reference software JMVC was selected as the experimental platform, the codec prediction structure was selected as HHI-IBBBP, and the inter-viewpoint prediction method was bi-directional prediction.

The experimental results of the video sequence ballroom are shown in Figures 8a-8b and Figure 9. Figure 8a is the reconstructed image of the video sequence ballroom in the case of the quantization parameter QP=24, the 15th frame image of the 0th viewpoint using the JMVC original coding method, and the PSNR of the reconstructed video image is 40.31dB. Fig. 8b is the reconstructed video image of the 15th frame image of the 0th viewpoint using the method of the present invention in the case of the quantization parameter QP=24 of the video sequence ballroom, and the PSNR of the reconstructed video image is 40.10dB. Figure 9 shows the video sequence ballroom using JMVC original encoding and the two methods of the present invention, in the case of different QP and different viewpoints, bit rate, PSNR value, bit rate saving percentage, reconstructed video subjective quality evaluation score difference (DMOS), average Statistical result of bit rate saving percentage. It can be seen that, under different QPs for the video sequence ballroom, the encoding rate using the method of the present invention is 7.47% to 9.16% lower than that using the original encoding method of JMVC, and the subjective video quality of the original encoding method of JMVC and the method of the present invention are The evaluation score difference is 0.03-0.07, and it can be considered that the subjective quality remains unchanged.

The experimental results of the video sequence race1 are shown in Figures 10a-10b and Figure 11. Figure 10a is the reconstructed video image of the 25th frame image of the first viewpoint using the JMVC original coding method in the case of the video sequence race1 with the quantization parameter QP=24, and the PSNR of the reconstructed video image is 41.15dB. Figure 10b shows the reconstructed video image of the 36th frame image of the first viewpoint using the JMVC original coding method in the case of the video sequence race1 with the quantization parameter QP=24, and the PSNR of the reconstructed video image is 40.51dB. Figure 11 shows the video sequence race1 using JMVC original encoding and the two methods of the present invention, in the case of different QP and different viewpoints, the bit rate, PSNR value, bit rate saving percentage, reconstructed video subjective quality evaluation score difference (DMOS), average Statistical result of bit rate saving percentage. It can be seen that, under different QPs for the video sequence race1, the coding rate using the method of the present invention is 10.77% to 12.35% lower than the coding rate using the JMVC original coding method, and the subjective video quality of the JMVC original coding method and the inventive method The evaluation score difference is 0.06-0.09, and it can be considered that the subjective quality remains unchanged.

The experimental results of video sequence crowd are shown in Figures 12a-12b and Figure 13. Fig. 12a is the reconstructed video image of the 45th frame image of the second viewpoint using the JMVC original coding method in the case of the video sequence crowd with the quantization parameter QP=35, and the PSNR of the reconstructed video image is 33.77dB. Figure 12b shows the reconstructed video image of the 45th frame image of the second viewpoint using the JMVC original coding method when the quantization parameter QP=35 of the video sequence crowd, and the PSNR of the reconstructed video image is 33.12dB. Figure 13 shows the video sequence crowd using JMVC original encoding and the two methods of the present invention, in the case of different QP and different viewpoints, bit rate, PSNR value, bit rate saving percentage, reconstructed video subjective quality evaluation score difference (DMOS), average Statistical result of bit rate saving percentage. It can be seen that the video sequence crowd is under different QPs, and the encoding rate using the method of the present invention is 8.95% to 9.83% lower than that using the JMVC original encoding method. The subjective video quality of the JMVC original encoding method and the inventive method The evaluation score difference is 0.03 to 0.08, and it can be considered that the subjective quality remains unchanged.

Combining the above charts, it can be seen that the present invention establishes a JND model in the DCT domain and applies it to the quantization process and rate-distortion optimization process of the multi-viewpoint video coding framework. Viewpoint video coding bit rate, which improves the compression efficiency of multi-viewpoint video coding.

Claims (7)

1. A method utilizing visual perception characteristics to guide the multi-viewpoint video coding quantization process, characterized in that the steps of operation are as follows: (1) Read the brightness value of each frame of the input video sequence, and establish a just discernable distortion threshold model in the frequency domain, (2) Each frame of the input video sequence undergoes intra-viewpoint and inter-viewpoint prediction, (3) Discrete cosine transform is performed on the residual data, (4) Dynamically adjust the quantization step size of each macroblock in the current frame, (5) Lagrangian parameters in the dynamic adjustment rate-distortion optimization process, (6) Entropy encoding is performed on the quantized data to form a code stream for transmission through the network. 2. the method for utilizing visual perception characteristics according to claim 1 to guide the multi-view coding quantization process, characterized in that said step (1) reads the size of the brightness value of each frame of the input video sequence, and establishes the accuracy of the frequency domain The steps to identify the distortion threshold model are as follows: ① Calculate the spatial sensitivity factors of 4x4 and 8x8DCT transformation according to the dimension of DCT transformation

, whose formula is:

where s is the control parameter,

is the angle of the frequency represented by the DCT coefficient vector,

is the DCT coefficient normalization factor,

is the spatial frequency, the parameters r, a, b, and c vary according to the size of the DCT transform: for a DCT coding format of 8×8 block size, is 0.6,

is 1.33,

is 0.11, is 0.18; for the DCT coding format of 4×4 block size,

is 0.6,

is 0.8,

is 0.035, is 0.008; ② According to the experimental results, the brightness masking effect of the human eye under different background brightness conditions The curves are represented as follows:

in, is the average pixel value of the current coding block; ③ Use the edge detector to detect the texture characteristics of the current coding block, and find the texture masking factor

, whose expression is as follows:

in,

Represents the horizontal and vertical coordinate coefficients of the transformation block,

Denotes the contrast mask estimator, is the spatial sensitivity factor,

is the DCT transform coefficient of the nth coding block of the current frame; ④ According to the speed of object movement in each frame of the video sequence, the time-domain masking effect factor is experimentally measured

The expression is: in,

is the spatial frequency,

is the time frequency; ⑤ The weighted product of the four factors obtained in steps ①~④ constitutes the just discernable distortion threshold of the current coded frame. 3. the method for utilizing visual perception characteristics according to claim 1 to guide the multi-view coding quantization process, characterized in that each frame of said step (2) input video sequence is through the operation steps of prediction in the viewpoint and between the viewpoints as follows: ① Perform inter-frame and intra-frame prediction within the viewpoint, compare the predicted value with the current frame to be encoded, and select an encoding method with a lower encoding cost; ② For inter-view prediction, the current encoded frame of the current view is predicted based on the corresponding frame of the reference view, and the predicted value is compared with the corresponding frame of the reference view to obtain the encoding cost of inter-view prediction; ③ Compare the coding cost between views and within the view, and choose the prediction mode with smaller coding cost. 4. The method for utilizing visual perception characteristics to guide the multi-view coding quantization process according to claim 1, wherein said step (3) is characterized in that the operation steps of discrete cosine transform of residual data are as follows: ① Judgment on the size of the coding block, when the length of any side of the coding block is less than 8, it is classified as a 4x4 transformation block, otherwise, it is an 8x8 transformation block; ② When it is a 4x4 transformation block, select 4x4 DCT transformation, and when it is an 8x8 transformation block, select 8x8DCT transformation. 5. the method for utilizing visual perception characteristics to guide the multi-view coding quantization process according to claim 1, characterized in that said step (4) dynamically adjusts the operation steps of the quantization step size of each macroblock in the current frame as follows: ① Calculate the average value of the just discernible distortion threshold of the current frame; ② Calculate the average value of the just discernible distortion threshold of the current coded macroblock; ③ Compare the mean value of the just discernible distortion threshold of each frame with the mean value of the just discernable distortion threshold of the current macroblock, and dynamically adjust the quantization step of the current macroblock. The expression of the adjusted quantization step is as follows:

in, Indicates the original quantization step size of the encoding frame,

Indicates the mean value of the just discernible distortion threshold of the current macroblock,

Indicates the mean value of the just discernable distortion threshold for the current frame, is the adjustment factor. 6. The method according to claim 1 in which visual perception characteristics are used to guide the multi-view coding quantization process, wherein the operation steps of the step (5) dynamically adjusting the Lagrangian parameters in the rate-distortion optimization process are as follows: ① Compare the mean value of the just discernible distortion threshold of each frame with the mean value of the just discernable distortion threshold of the current macroblock; ② Adjust the Lagrangian parameters, the expression of the adjusted Lagrangian parameters is:

in is the adjustment factor,

is the adjusted quantization step size,

Indicates the original quantization step size of the encoding frame,

Indicates the mean value of the just discernible distortion threshold of the current macroblock, Indicates the mean value of the just discernable distortion threshold of the current frame; ③ The optimization of the encoding cost function dynamically adjusts the Lagrangian parameters, so that the rate-distortion optimization function can regain the optimal solution when the quantization step size is changed; the expression is:

in

is a distorted signal,

is the number of bits encoded in different encoding modes,

is the adjusted Lagrange parameter. 7. The method according to claim 1 utilizing visual perception characteristics to guide the multi-view coding quantization process, characterized in that said step (6) carries out entropy coding to the quantized data, and the operation steps of forming a code stream for transmission through the network are as follows: ① The quantized data is entropy encoded, so that the quantized data forms a binary code stream; ② The encoded code stream is transmitted through the network.

Images