CN103475881B - The image JND threshold value computational methods of view-based access control model attention mechanism in DCT domain - Google Patents

Abstract

A visual attention mechanism based image JND threshold calculation method in the DCT domain. The present invention proposes two schemes combining saliency and block classification, one is to combine the visual attention masking factor of a single point with the masking factor of block classification in a point-to-point manner, and the other is to use each block The average saliency of represents the saliency of the whole block, and then combines the visual attention masking factor based on each block and the masking factor of the block classification in a block-to-block manner. The traditional JND threshold is modulated using the value calculated by the integrated contrast masking function, resulting in a more accurate JND threshold. Both methods can effectively improve the accuracy of the JND threshold, thus making the JND threshold more closely match the human visual system. The model realized by the image JND threshold calculation method proposed by the present invention can accommodate more noise, and in terms of PSNR, the average model can be improved by 0.54DB.

Description

Image JND Threshold Calculation Method Based on Visual Attention Mechanism in DCT Domain

technical field

The invention relates to the technical field of image/video coding.

technical background

Traditional image/video coding technology mainly compresses and codes spatial redundancy, time domain redundancy, and statistical redundancy, but rarely considers the characteristics and psychological effects of the human visual system, so a large amount of visual redundant data is encoded and transmitted , in order to further improve the efficiency of coding, researchers have started research on removing visual redundancy. At present, an effective method to represent visual redundancy is the least detectable distortion model based on psychology and physiology, referred to as the JND model, which can also be called the just detectable distortion model, that is, the changes that the human eye cannot perceive, due to various human eyes. The shielding effect means that the human eye can only perceive noise above a certain threshold, which is the just detectable distortion of the human eye, representing the visual redundancy in the image. The JND model is often used to guide the perceptual coding and processing of images or videos, such as preprocessing, adaptive quantization, stream control, motion estimation, etc.

The existing perceivable distortion (JND) models can be roughly divided into two categories: the first category is the pixel-domain JND model, and its basic principles are mostly modeled by characterizing the brightness adaptive effect and the texture masking effect, such as literature 1 (see X.Yang, W.Lin, Z.Lu, E.P.Ong, and S.Yao, "Just-noticeable-distortion profile with nonlinear additivity model for perceptual masking color images", IEEETrans.Circuits Syst.Video Technol., vol.15, no.6, pp742-752, Jun.2005) proposed a color image JND model based on airspace, but because it cannot integrate the contrast sensitivity function (ContrastSensitive Function, CSF) well, there is no way for this type of model Get an accurate JND value, often used as a quick way to calculate the JND threshold.

The second type of JND model is the subband JND model, which is calculated in the transform domain, such as DCT domain, wavelet domain, CONTOURLET domain and so on. Since most image/video coding standards are based on the DCT domain (such as JPEG, H.261/3/4, MPEG-1/2/4), the JND model based on the DCT domain has attracted the attention of many researchers, such as Document 2 (see Z.Wei and K.N.Ngan, "Spatial just noticeable distortion profile for image inDCT domain," In Proc.IEEE Int.Conf.Multimeda and Expo, pp.925-928, 2008.) combined with the brightness of the image from Adaptive characteristics, spatial contrast effect and contrast masking effect based on block classification, but this model does not consider the influence of the human eye's visual attention mechanism on the JND model, so the calculation accuracy needs to be further improved.

Contents of the invention

On the basis of the WEI model, the present invention combines the visual attention mechanism to propose a new JND model modeling method for images in the DCT domain, and designs a comprehensive modulation function and brightness by comprehensively considering the visual attention effect and contrast masking effect A method for modulating spatial contrast-sensitive functions together with adaptive effects.

For this reason, the present invention provides technical scheme implementation steps as:

A DCT domain-based image perceivable distortion calculation method adopts the following technical solution, including the following steps:

Step S1: Perform 8x8 DCT transformation on the selected image, and transform it from the spatial domain to the DCT domain.

Step S2: In the DCT domain, calculate the basic perceivable distortion degree JND value obtained according to the product of the spatial contrast effect threshold and the brightness adaptive modulation factor.

Step S3: using the canny edge detector to divide the image into smooth blocks, edge blocks and texture blocks to obtain a contrast masking factor based on the block structure.

Step S4: Use the visual attention model to perform saliency detection on the image to obtain a saliency map of the image.

Step S5: Segment the image according to the saliency map obtained in step S4, divide the image into salient regions and non-salient regions, and then obtain the contrast masking factor based on the visual attention mechanism based on the saliency value of each point.

Or step S5 is: first divide the image into salient areas and non-salient areas, then divide the salient map obtained in step S4 into blocks, replace the salient value of the entire block with the average value of the salient values of each block, and based on each block The saliency value of is obtained as a contrast masking factor based on the visual attention mechanism.

Step S6: Combining the segmentation results of salient areas and non-salient areas of the image obtained in step S5 with the block classification results obtained in step S3, the image is divided into more detailed blocks, and the contrast masking factor based on the visual attention mechanism and The contrast masking factors based on the block structure are combined according to a linear relationship to obtain a comprehensive contrast masking modulation function,

Step S7: Modulate the modulation function value calculated in step S6 to the basic JND threshold calculated in step S2 to obtain a final JND threshold.

The key technical points reflected in the above technical solutions:

1. In view of the problem that the traditional image just perceptible distortion model does not consider the visual attention mechanism, the present invention proposes two modeling algorithms for the image just perceptible distortion model based on the visual attention mechanism in the DCT domain. One is, by calculating The visual attention modulation factor based on image saliency, and the pixel-based visual attention masking factor and the block structure-based contrast masking factor are combined to calculate a comprehensive contrast masking function, which is based on the traditional spatial contrast effect and The JND threshold of the brightness adaptive effect is modulated; the second is, by calculating the saliency map of the image, and replacing the saliency factor of the whole block with the saliency average value of each block, and then establishing a block-based visual attention masking factor and The block-structure-based contrast masking factors are combined to obtain a synthetic contrast masking function that modulates the underlying JND threshold. Both methods can effectively improve the accuracy of the JND threshold, thus making the JND threshold more closely match the human visual system.

2. The present invention proposes the concept of visual attention masking effect, and simulates the human eye's attention to image pixels through visual salience, thereby establishing a visual attention masking effect factor based on visual salience.

3. The present invention proposes two schemes combining saliency and block classification, one is to combine the masking factor of visual attention of a single point and the masking factor of block classification in a point-to-point manner, and the other is to use each The average saliency of each block represents the saliency of the whole block, and then combines the visual attention masking factor based on each block and the masking factor of the block classification in a block-to-block manner.

4. By combining visual saliency and block classification, images are segmented more finely and accurately.

5. Based on the saliency and block structure characteristics of each block, by setting different saliency modulation factors and block structure modulation factors, the visual attention masking factor and the block classification contrast masking factor are combined to obtain a comprehensive contrast masking function .

The beneficial effect of the method of the present invention is that the traditional JND threshold is modulated by using the value calculated by the comprehensive contrast masking function, and finally a more accurate JND threshold is obtained. Under the premise of ensuring the same visual subjective quality, the model realized by the image JND threshold calculation method proposed by the present invention can accommodate more noise.

Description of drawings

Fig. 1 is a block diagram of an image perceivable distortion model based on a visual attention mechanism in the DCT domain of the present invention.

Fig. 2 is the Airplane image of the example of the present invention.

Fig. 3 is an image after block classification of the Airplane image of the example of the present invention.

Fig. 4 is the image after the saliency classification of the Airplane image of the example of the present invention.

Fig. 5 is the detailed block result of the Airplane image of the example of the present invention considering the saliency and the block structure comprehensively.

FIG. 6 is a flowchart of a method for calculating an image JND threshold based on a visual attention mechanism in the DCT domain of the present invention.

detailed description

Below in conjunction with accompanying drawing, the present invention will be further described with specific example:

The example that the present invention provides adopts MATLAB7 as the emulation experiment platform, uses the bmp grayscale image Airplane of 512 * 512 as the selected test image, describes this example in detail below in conjunction with each step:

Step (1), select the bmp grayscale image of 512 * 512 as the image of input test, carry out the DCT transformation of 8 * 8 to it, transform it into DCT domain by space domain;

In step (2), in the DCT domain, the perceivable distortion JND value is obtained by calculating the product of the basic threshold value of the spatial contrast and the adaptive modulation factor of the brightness, and the calculation formula is as follows:

T _JND (n,i,j)=T _Basic (n,i,j)×F _lum (n) (1)

Among them, T _Basic (n,i,j) represents the spatial contrast sensitivity threshold, F _lum (n) represents the brightness adaptive adjustment factor, n is the index of the DCT block, W is the width of the image, H is the height of the image, i, j are the indices of the coefficients in the DCT block, 1≤i≤64, 1≤j≤64. The general DCT block is 8x8 in size, then the test image is divided into A DCT block, n takes a value between 1 and 4096, and i and j take a value between 1 and 64.

The basic threshold T _Basic (n, i, j) in the above formula (l) is calculated by the following method:

The frequencies of the DCT subbands can be expressed as follows:

${\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; N</span>}}\sqrt{{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; \&theta;</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; \&theta;</span>}}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>$

Wherein θ _x = θ _y = 2·arctan(γ/2·l) is the horizontal and vertical viewing angle of a pixel, l is the viewing distance of the image, in this invention, l is 3 times of the image width, and γ represents the One pixel display width or length. From the above, the spatial contrast sensitivity threshold of the DCT block is:

In the formula (3): s represents the spatial set effect, gets 0.25 in this invention, Represents the tilt effect, where r=0.6, Ф _i and Ф _j are DCT normalization coefficients, Represents the orientation angle of the corresponding DCT coefficient:

${\mathrm{<span\; class="notranslate">\; \&Phi;</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}\sqrt{\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; N</span>}}& \mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}\\ \sqrt{\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; N</span>}}& \mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ></span>\mathrm{<span\; class="notranslate">\; 0</span>}\end{array}\right.<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 4</span>}<span\; class="notranslate">\; )</span>$

Parameters a=1.33, b=0.11, c=0.005.

The brightness adaptive modulation factor F _lum (n) in the above formula (1) is calculated by the following method:

${\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; lu</span>}\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 60</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; 150</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; 60</span>}\\ \mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; 60</span>}<span\; class="notranslate">\; <</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; <</span>\mathrm{<span\; class="notranslate">\; 170</span>}\\ <span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 170</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; 425</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; \&Greater\; Equal;</span>\mathrm{<span\; class="notranslate">\; 170</span>}\end{array}\right.<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 6</span>}<span\; class="notranslate">\; )</span>$

In formula (6), I represents the average brightness value of each 8×8 block.

Then calculate according to the formula (1) to obtain the traditional JND threshold.

In step (3), block classification is performed on the image, and a contrast masking factor F _contrast (n,i,j) based on block classification is calculated.

Classify all 8×8 DCT blocks in the image into three categories: texture, flat and edge (as shown in Figure 2); use the canny operator to detect and calculate the edge information in the image, by calculating each 8×8 The edge pixel density ρ _edge in the block is used to classify the block, and the specific process is as follows:

ρ _edge ＝(∑ _8X8 edge)/N ² (7)

Wherein, ρ _edge represents the total number of edge pixels obtained by the canny operator in each 8×8 block; N is the DCT block size, and N=8 is taken in this example;

The block classification methods are:

According to the 8×8 DCT block type, the inter-block masking effect factor is:

Considering the masking effect between adjacent subbands, the traditional block classification-based contrast masking function is expressed as follows:

Step (4), performing saliency detection on the image to obtain a saliency map of the image.

In this example, the spectral residual method is used to detect the saliency of the image:

Among them, P(f) and R(f) represent the phase spectrum and amplitude spectrum after Fourier transform of the image respectively, Represents the inverse Fourier transform, g(σ) represents the Gaussian filter, σ is the filter window size, in this example, σ is 0.32 times the image width. S(x) represents the saliency map of each image.

Step (5), if option 1 is selected, refer to (a) description, if option 2 is selected, refer to (b) description:

(a) Based on the saliency map of the image, calculate the saliency contrast masking factor of the image:

F _vs (n,i,j)=μ·(S _max -S(n,i,j)) (12)

The larger the saliency value of the image, the higher the attention of the human eye to this point, where S _max represents the maximum value after normalization of the saliency map, and S(n,i,j) represents each Each point in the block corresponds to the saliency of position (i,j). μ represents the modulation factor, and μ=1.0 is taken in this example.

(b) Image-based saliency map, where the average saliency of each block is used to represent the saliency of the whole block. And based on the saliency of each block, calculate the saliency contrast masking factor of the image:

F _vs (n)＝μ·(S _max (n)-S(n)) (13)

Among them, S _max (n) represents the maximum value of the saliency of all blocks, and S(n) represents the saliency of each block. μ represents the modulation factor, and μ=1.0 is taken in this example. F _vs (n) represents the saliency masking factor for each block.

In step (6), the images are classified in more detail from both the perspective of block structure and the perspective of saliency.

First, according to the saliency of the image, the image is divided into salient areas and non-salient areas (as shown in Figure 3):

where T represents the segmentation threshold, in this example T is the mean value of the saliency map.

Combining the results of saliency segmentation and block classification, the image is classified more finely, as shown in Figure 4.

Step (7), according to different block structures, set different salience adjustment factors and block structure adjustment factors, so as to calculate the comprehensive modulation function:

${\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; no</span>}\mathrm{<span\; class="notranslate">\; t</span>}\mathrm{<span\; class="notranslate">\; r</span>}\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; the\; s</span>}\mathrm{<span\; class="notranslate">\; t</span>}}^{\mathrm{<span\; class="notranslate">\; v</span>}\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; \&alpha;</span>}<span\; class="notranslate">\; \&times;</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; v</span>}\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&beta;</span>}<span\; class="notranslate">\; \&times;</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; no</span>}\mathrm{<span\; class="notranslate">\; t</span>}\mathrm{<span\; class="notranslate">\; r</span>}\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; the\; s</span>}\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 15</span>}<span\; class="notranslate">\; )</span>$

In formula (15), if scheme 1 is adopted, F _vs represents the saliency masking factor calculated using formula (12); if scheme 2 is adopted, F _vs represents the saliency masking factor calculated using formula (13). F _contrast (n, i, j) represents the contrast masking factor based on the block structure calculated by formula (10). The values of saliency modulation factor α and block structure adjustment factor β are based on different block structures:

Step (8), based on the JND basic threshold calculated in step (2), use step (7) to calculate and obtain a comprehensive modulation function, and modulate the JND basic threshold:

${\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; J</span>}\mathrm{<span\; class="notranslate">\; N</span>}\mathrm{<span\; class="notranslate">\; D.</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; the\; s</span>}\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&times;</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; l</span>}\mathrm{<span\; class="notranslate">\; u</span>}\mathrm{<span\; class="notranslate">\; m</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&times;</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; no</span>}\mathrm{<span\; class="notranslate">\; t</span>}\mathrm{<span\; class="notranslate">\; r</span>}\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; the\; s</span>}\mathrm{<span\; class="notranslate">\; t</span>}}^{\mathrm{<span\; class="notranslate">\; v</span>}\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 17</span>}<span\; class="notranslate">\; )</span>$

The JND threshold of the image is calculated by combining all the above steps. The threshold comprehensively considers the spatial contrast effect, the brightness adaptive effect, the block classification contrast masking effect and the visual attention mechanism, so the threshold is more consistent with the human visual system and more accurate. .

Innovation point of the present invention:

The idea of applying image saliency to improve the just perceptible distortion model of image is proposed.

A visual attention masking factor based on image saliency is proposed, which describes the degree of human eye's attention to the image.

Considering the salient feature and block structure feature of the image comprehensively, the image can be classified more accurately and meticulously.

Based on the comprehensive consideration of the salient characteristics and block structure characteristics of different blocks, a comprehensive modulation function is constructed to modulate the traditional JND model, and the JND threshold value that is more consistent with the visual system is calculated.

Claims (1)

1. An image perceptible distortion calculation method based on the DCT domain, comprising the following steps: Step S1: Perform 8x8 DCT transformation on the selected image, and transform it from the spatial domain to the DCT domain; Step S2: In the DCT domain, calculate the basic perceptible distortion degree JND value obtained according to the product of the spatial contrast effect threshold and the brightness adaptive modulation factor; Step S3: using the canny edge detector to divide the image into smooth blocks, edge blocks and texture blocks to obtain a contrast masking factor based on the block structure; Step S4: using the visual attention model to perform saliency detection on the image to obtain a saliency map of the image; Step S5: Segment the image according to the saliency map obtained in step S4, divide the image into salient areas and non-salient areas, and then obtain a contrast masking factor based on the visual attention mechanism based on the salient value of each point; Or step S5 is: first divide the image into salient areas and non-salient areas, then divide the salient map obtained in step S4 into blocks, replace the salient value of the entire block with the average value of the salient values of each block, and based on each block The saliency value of is obtained as a contrast masking factor based on the visual attention mechanism; Step S6: Combining the segmentation results of salient areas and non-salient areas of the image obtained in step S5 with the block classification results obtained in step S3, the image is divided into more detailed blocks, and the contrast masking factor based on the visual attention mechanism and The contrast masking factors based on the block structure are combined according to a linear relationship to obtain a comprehensive contrast masking modulation function; Step S7: Modulate the JND basic threshold calculated in step S2 with the contrast masking modulation function value calculated in step S6 to obtain a final JND threshold.