CN109978834A - A kind of screen picture quality evaluating method based on color and textural characteristics - Google Patents

Abstract

The invention relates to a screen image quality evaluation method based on color and texture features, which is characterized in that: first, a parameter model is designed to describe the features of color information in the screen image, and a statistical model is used to extract the quality-related features to evaluate the screen image spatial color continuity. Second, local texture features are used to describe the spatial texture continuity of screen images. Finally, visual features are mapped to subjective scores using support vector regression as the mapping function. Experimental results show that the proposed method shows good performance in predicting the visual quality of screen images compared with the performance of existing algorithms without reference screen images, and even outperforms some algorithms with full reference screen images.

Description

A Screen Image Quality Evaluation Method Based on Color and Texture Features

technical field

The invention designs a screen image quality evaluation method based on color and texture features, belonging to the technical field of multimedia, in particular to the technical field of digital image and digital video processing.

Background technique

With the rapid development of wireless networks, digital images are no longer just about capturing natural scenes with DSLR cameras. In fact, digital images contain a variety of image content, such as natural scene content, computer-generated graphics, hand-painted, special signs, and images synthesized by software. The images mentioned above are collectively referred to as screen images. Screen images are composed of natural scenes, graphics, tables, words, and special symbols generated by computers, and contain information that is different from natural images.

In recent years, mean square error and peak signal-to-noise ratio have been widely used for visual quality evaluation of naturally distorted images. However, the quantitative measures of these traditional methods are not consistent with human ratings, because these methods only consider the difference between the corresponding pixels of the reference image and the distorted image, and do not take into account the characteristics of the human visual system. To solve these problems, many studies take human visual characteristics into consideration and design advanced visual quality evaluation methods. These existing methods are all for evaluating the quality of natural images, and they can often achieve better results in evaluating the quality of natural images, but their performance on the screen image database is much lower than that of natural images, so they cannot effectively measure distortion. The quality of the screen image is evaluated.

Due to the difference of image content, the quality evaluation of screen images by the quality evaluation method of natural distorted images may ignore some important information, which will lead to inaccurate perceived quality of screen images. A superior screen image quality evaluation algorithm can take into account the rich information including graphics, edges and textures and give the correct quality evaluation score. Therefore, how to effectively evaluate the visual quality of screen images is very important.

Generally speaking, subjective quality assessment is the most effective and reliable, which reflects the human eye's intuitive judgment on the quality of screen images. However, this method is time-consuming and labor-intensive, so subjective quality assessment cannot be applied in practical applications. middle. Therefore, it is worth studying to design an effective and robust no-reference quality evaluation method for screen images based on the inherent characteristics of screen images.

The purpose of proposing an algorithm that can automatically evaluate screen images is to:

(1) A non-referenced screen image quality evaluation algorithm can be designed to be embedded in any processing link containing screen images to supervise the quality of this link and provide timely feedback, which is beneficial to subsequent processing.

(2) Through an efficient screen image quality evaluation method, it can play a guiding role in the generation and display of screen images, so that the screen images are more in line with the perception of the human visual system.

Therefore, an accurate and effective screen image quality evaluation algorithm will greatly promote the development of screen images.

SUMMARY OF THE INVENTION

In order to improve the performance of the objective quality evaluation method of screen images and for the specific distortion types and contents of screen images, a new screen image quality evaluation model based on spatial continuity is proposed. The adopted visual features include color statistical features based on spatial color continuity and texture statistical features based on spatial texture continuity.

The process of the present invention is shown in Figure 1, and the concrete process is as follows:

Step 1: Use zero-order and first-order invariant-based color descriptors to extract color statistical features of screen images to evaluate the spatial color continuity of screen images.

Step 2: Extract the texture statistical features of the screen image by using the local ternary mode operation to describe the spatial texture continuity of the screen image.

Step 3: Using the machine learning method, map the extracted features to the image quality score through support vector regression, and use the learned features to map the scores as the standard for screen image quality evaluation.

Three commonly used criteria are used to evaluate the algorithm's accuracy in predicting screen image quality. The evaluation criteria include: Pearson Linear Correlation Coefficient PLCC (Pearson Linear Correlation Coefficient) for evaluating the accuracy of prediction; Spearman Rank-order Correlation Coefficient SRCC (Spearman Rank-order Correlation Coefficient) for evaluating the monotonicity of prediction; The Root Mean Squared Error (RMSE) of the correlation of subjective scores. In general, higher PLCC and SRCC, and lower RMSE values indicate better predictive performance of the algorithmic model. In order to verify the performance of the algorithm proposed in the present invention, the proposed algorithm is compared with the existing full-reference and no-reference image quality evaluation methods on the databases SIQAD and SCID. In order to eliminate the evaluation bias, each algorithm was repeatedly evaluated 1000 times with the above criteria, and the median of the results of each criterion was taken as the final algorithm result. The experimental results show that the no-reference quality evaluation model for screen images proposed by the present invention is significantly better than other current models.

A quality evaluation method for screen images, the feature extraction includes the following steps:

A. Using color invariant descriptors to extract color statistical features in screen images and evaluate the spatial color continuity of screen images.

B. The local ternary mode operation operator and logarithmic derivative are used to extract local texture statistical features to describe the spatial texture continuity of the screen image.

C. Using support vector regression as the mapping function to map image perceptual features to subjective scores.

2 . The quality evaluation method for screen images according to claim 1 , wherein the included visual features include: color statistical features and texture statistical features. 3 .

Further, the visual features include color statistical features. Color statistical features are obtained through zero-order color invariance descriptors and first-order color invariance descriptors.

Further, color statistical features are extracted through two zero-order color invariance descriptors and two first-order color invariance descriptors, and the specific steps are:

A. Use zero-order color invariance descriptors saturation φ and chroma ψ to extract color statistical features. Saturation can be defined as:

Among them, R, G, B represent the color of the red, green and blue channels of the image, respectively. Taking into account the spatial correlation, the image φ(i, j)' after removing redundant information is calculated, and the calculation of the saturation difference in the four directions between adjacent pixels is shown in the following formula:

φ _h (i,j)′=φ(i,j+1)-φ(i,j) (2)

φ _v (i,j)′=φ(i+1,j)-φ(i,j) (3)

φ _d1 (i,j)′=φ(i+1,j+1)-φ(i,j) (4)

φ _d2 (i,j)′=φ(i+1,j-1)-φ(i,j) (5)

Where h represents the horizontal direction, v represents the vertical direction, d1 represents the main diagonal direction, and d2 represents the secondary diagonal direction. φ(i, j) represents the saturation of the pixel in the i-th row and the j-th column.

Since the intensity value distribution of screen images is not like the Gaussian or Gaussian-like distribution _obeyed by natural images, it is determined by performing Histogram statistics and binning to extract color features c ₁ ∈ {φ′ _h , φ′ _v , φ′ _d1 , φ′ _d2 }. Use the following model to calculate a statistical histogram of relative saturation:

where m is the histogram block index m∈[1,10], and d represents four directions, namely d∈{h, v, d ₁ , d ₂ }, H and W are the height and width of the image, respectively, And i∈{1,2,…,H}, j∈{1,2,…,W}.

The luminosity and reflection invariance of the image is measured based on the chromaticity descriptor. First, the image is converted from the RGB color space to the OC space. The conversion formula is: Among them, O1 and O2 represent the color components that do not change with photometric and geometric transformations, and O3 is the luminance component. Chroma can be defined as:

Similarly, the chromaticity difference between adjacent pixels is used to extract the features based on the chromaticity of the image. The calculation formula of the chromaticity difference in the four directions is as follows:

ψ _h (i,j)′=Γ(ψ(i,j+1),ψ(i,j)) (9)

ψ _v (i, j)′=Γ(ψ(i+1,j),ψ(i,j)) (10)

ψ _d1 (i, j)′=Γ(ψ(i+1, j+1), ψ(i, j)) (11)

ψ _d2 (i, j)′=Γ(ψ(i+1,j-1),ψ(i,j)) (12)

Among them, ψ(i,j) represents the chromaticity value on the screen image space (i,j), the chromaticity difference _ψd (i,j)′ represents the color angle difference between adjacent pixels, and d∈{h , v, d ₁ , d ₂ }, the color angle difference operation Γ can be defined as:

Similar to the saturation descriptor, histogram statistics are performed on the chrominance difference map c ₂ ∈{ψ′ _h ,ψ′ _v ,ψ′ _d1 ,ψ′ _d2 }, and further extract the corresponding color features.

B. The first-order color invariance descriptor is used to extract color statistical features for vertex angle θ and spherical angle δ. Opposite angles can be defined as:

in and represent the Gaussian partial derivatives of O1 and O2, respectively, and the Gaussian function with parameter σ is defined as: Therefore, it can be defined where R _x , G _x and B _x are the first-order Gaussian partial derivatives of the R, G, and B channels of the color image, respectively. Similar to equations (9)-(12), the diagonal angles can be calculated in four directions, namely θ _h (i, j)′, θ _v (i, j)′, θ _d1 (i, j)′ , θ _d2 (i,j)′. and the corresponding statistical histogram is calculated c ₃ ∈ {θ′ _h , θ′ _v , θ′ _d1 , θ′ _d2 }, and further extract color features according to the histogram.

The last color invariant descriptor spherical angle can be defined as:

in is defined as and represents the dot product operation, in addition, can be defined as:

in γ=R ² +G ² , similarly, the spherical angles in four directions can be calculated, namely δ _h (i,j)′,δ _v (i,j)′,δ _d1 (i,j)′,δ _d2 (i,j)′, and further calculate the corresponding statistical histogram c ₄ ∈{δ′ _h ,δ′ _v ,δ′ _d1 ,δ′ _d2 }, thereby extracting the corresponding color statistical features according to the histogram.

Finally, a total of 320 color features are extracted from the two scales of the distorted screen image through the above four color invariance descriptors, and these features constitute the color feature vector of perceptual quality. The calculation formula is as follows:

in, is the connection operation, Feature vector representing the input to the SVR model.

Further, the texture statistical features are extracted through the local ternary mode operation operator, and the specific steps are:

Because the local ternary mode operation can extract important information sensitive to image distortion, the distorted screen image is firstly divided into two channels, the top mode (u) and the bottom mode (l), according to the local three-value mode operation. The corresponding statistical histogram can be calculated according to formulas (6) and (7). Use the statistical histogram to extract the corresponding texture statistical features respectively, namely and and m∈[1, 10]. According to the existing research, the logarithmic derivative of the image can well represent the texture information of the image, and the derivative can be defined as the difference of the intensity values of adjacent pixels in four directions. Therefore, the logarithmic derivative of the top mode and bottom mode of the screen image can also be used to extract the corresponding texture features. The updated modes u(d, j)' and l(d, j)' are obtained by logarithmic operation on the top mode and bottom mode of the image, and the calculation formula is as follows:

u(i,j)′=log[u(d,j)+C] (18)

l(i,j)′=log[l(i,j)+C] (19)

where C is a small constant. To further calculate the spatial correlation between adjacent pixels in the top mode, the calculation formula is as follows:

Similarly, the spatial correlations between adjacent pixels of the bottom pattern are also calculated, and then the corresponding statistical histograms are further calculated based on these calculated correlation maps, i.e. and Thereby, the corresponding texture statistical features are extracted according to the histogram.

Finally, a total of 200 color features are extracted from the two scales of the distorted screen image through the above-mentioned top mode and bottom mode, and these features constitute the texture feature vector of perceptual quality. The calculation formula is as follows:

in, is the connection operation, Feature vector representing the input to the SVR model.

Description of drawings

FIG. 1 is an algorithm flow frame diagram of the present invention.

Detailed ways

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some, but not all, embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present invention.

Wherein, the technical features, abbreviations/abbreviations, symbols, etc. involved in this document are explained, defined/illustrated on the basis of the common knowledge/common understanding of those skilled in the art.

In order to improve the performance of the objective quality evaluation method of screen images and for the specific distortion types and contents of screen images, a new screen image quality evaluation model based on spatial continuity is proposed. The adopted visual features include color statistical features based on spatial color continuity and texture statistical features based on spatial texture continuity.

The process of the present invention is shown in Figure 1, and the concrete process is as follows:

Step 1: Use zero-order and first-order invariant-based color descriptors to extract color statistical features of screen images to evaluate the spatial color continuity of screen images.

Step 2: Extract the texture statistical features of the screen image by using the local ternary mode operation to describe the spatial texture continuity of the screen image.

Step 3: Using the machine learning method, map the extracted features to the image quality score through support vector regression, and use the learned features to map the scores as the standard for screen image quality evaluation.

Three commonly used criteria are used to evaluate the algorithm's accuracy in predicting screen image quality. The evaluation criteria include: Pearson Linear Correlation Coefficient PLCC (Pearson Linear Correlation Coefficient) for evaluating the accuracy of prediction; Spearman Rank-order Correlation Coefficient SRCC (Spearman Rank-order Correlation Coefficient) for evaluating the monotonicity of prediction; The Root Mean Squared Error (RMSE) of the correlation of subjective scores. In general, higher PLCC and SRCC, and lower RMSE values indicate better predictive performance of the algorithmic model. In order to verify the performance of the algorithm proposed in the present invention, the proposed algorithm is compared with the existing full-reference and no-reference image quality evaluation methods on the databases SIQAD and SCID. In order to eliminate the evaluation bias, each algorithm was repeatedly evaluated 1000 times with the above criteria, and the median of the results of each criterion was taken as the final algorithm result. The experimental results show that the no-reference quality evaluation model for screen images proposed by the present invention is significantly better than other current models.

The concrete operations of each part of the present invention are as follows:

Extraction of color statistical features:

Color statistics are extracted from zero- and first-order invariant color descriptors based on spatial color continuity.

For the zero-order invariant color descriptor, it represents the zero-order derivative information of the image color channel. First, the corresponding color statistical features are extracted through the color saturation invariance descriptor. Saturation is the degree to which a pure color is diluted by white light, which is an important part of the color representation of the image. Generally, high saturation means the image has better visual quality. For a given screen image I = {R, G, B}, the color saturation is calculated as follows:

Usually, the change of the saturation value of adjacent pixels may lead to the generation of visual distortion. Therefore, in order to eliminate the spatial redundancy and predict the spatial correlation, the horizontal direction (h), vertical direction (v), main diagonal The relative saturation of color between adjacent pixels in the line direction (d1) and the sub-diagonal direction (d2). Its calculation formula is as follows:

φ _h (i,j)′=φ(i,j+1)-φ(i,j) (2)

φv(i, _j )′=φ(i+1,j)-φ(i,j) (3)

φ _d1 (i, j)′=φ(i+1, j+1)-φ(i, j) (4)

φ _d2 (i, j)′=φ(i+1, j-1)-φ(i, j) (5)

On this basis, take the absolute value of φ _d (i,j)′,d∈(h,v,d1,d2), and express the color feature in the form of a histogram. The number of blocks of the histogram is 10, and the extracted color statistical features are c ₁ ∈ {φ′ _h , φ′ _v , φ′ _d1 , φ′ _d2 }. The corresponding histogram calculation is as follows:

where m is the histogram block index m∈[1,10], and d represents four directions, namely d∈{h,v,d ₁ ,d ₂ }, H and W are the height and width of the image, respectively, And i∈{1,2,…,H}, j∈{1,2,…,W}.

Secondly, the corresponding color statistical features are extracted through the color chromaticity invariance descriptor. Chroma is an important factor affecting the color perception of an image. Convert an image from RGB color space to OC space: Among them, O1 and O2 represent the color components that do not change with photometric and geometric transformations, and O3 is the luminance component. Chroma is calculated as follows:

Similarly, the chromaticity difference between adjacent pixels is used to extract features based on image chromaticity, horizontal (h), vertical (v), main diagonal (d1) and sub-diagonal (d2) The formula for calculating the chromaticity difference in the four directions is as follows:

ψ _h (i, j)′=Γ(ψ(i, j+1), ψ(i, j)) (9)

ψ _v (i, j)′=Γ(ψ(i+1, j), ψ(i, j)) (10)

ψ _d1 (i, j)′=Γ(ψ(i+1, j+1), ψ(i, j)) (11)

ψ _d2 (i, j)′=Γ(ψ(i+1, j-1),ψ(i,j)) (12)

Among them, ψ(i,j) represents the chromaticity value on the screen image space (i,j), _ψd (i,j)′ represents the color chromaticity difference between adjacent pixels, and d∈{h,v , d ₁ , d ₂ }, the color chromaticity difference operation Γ can be defined as:

Similar to the saturation descriptor, histogram statistics are performed on the chrominance difference map c ₂ ∈{ψ′ _h ,ψ′ _v ,ψ′ _d1 ,ψ′ _d2 }, and further extract the corresponding color features.

For the first-order invariant color descriptor, it represents the first-order derivative information of the image color channel. Corresponding color statistical features are extracted using color pair vertex descriptors. The vertex angle is defined as follows:

in and represent the Gaussian partial derivatives of O1 and O2, respectively, and the Gaussian function with parameter σ is defined as: Therefore, it can be defined where R _x , G _x and B _x are the first-order Gaussian partial derivatives of the R, G, and B channels of the color image, respectively. Similar to formulas (9)-(12), the difference of the opposite angles is calculated in four directions, namely θ _h (i,j)′, θ _v (i, j)′, θ _d1 (i, j)′, θ _d2 (i,j)′. And the corresponding statistical histogram is calculated as c ₃ ∈{θ′ _h , θ′ _v , θ′ _d1 , θ′ _d2 }, and further extract color statistical features according to the histogram.

The last first-order invariant color descriptor spherical angle can be defined as:

in is defined as and represents the dot product operation, in addition, can be defined as:

in γ=R ² +G ² , similarly, the spherical angle difference in four directions can be calculated, namely δ _h (i,j)′,δ _v (i,j)′,δ _d1 (i,j)′, δ _d2 (i,j)′, and further calculate the corresponding statistical histogram c ₄ ∈{δ′ _h ,δ′ _v ,δ′ _d1 ,δ′ _d2 }, so as to extract the corresponding color statistical features according to the histogram.

Finally, a total of 320 color features are extracted from the two scales of the distorted screen image through the above four color invariance descriptors, and these features constitute the color statistical feature vector of the perceptual quality of the screen image, and the vector is defined as follows:

in, is the connection operation, Feature vector representing the input to the SVR model.

Extraction of texture statistical features:

Local features are extracted by local ternary mode operations based on spatial texture continuity.

The local ternary mode operation can extract important information sensitive to image distortion, so first, according to the local ternary mode operation, the distorted screen image is divided into two channels: the top mode (u) and the bottom mode (l). Equations (6) and (7) can calculate corresponding statistical histograms. Use the statistical histogram to extract the corresponding texture statistical features respectively, namely and and m∈[1,10]. In addition, existing research shows that the logarithmic derivative of the image can well represent the texture information of the image, and the derivative can be defined as the difference in the intensity values of adjacent pixels in four directions. Therefore, the corresponding texture features can also be extracted for the logarithmic derivatives of the top mode and bottom mode of the screen image. The updated modes u(i,j)' and l(i,j)' are obtained by logarithmic operation on the top mode and bottom mode of the image, and the calculation formula is as follows:

u(i,j)′=log[u(i,j)+C] (18)

l(i,j)′=log[l(i,j)+C] (19)

where C is a small constant. To further calculate the spatial correlation between adjacent pixels in the top mode, the calculation formula is as follows:

Similarly, the spatial correlation between adjacent pixels of the bottom pattern is calculated, and then the corresponding statistical histogram is further calculated according to these calculated correlation maps, namely and Thereby, the corresponding texture statistical features are extracted according to the histogram.

Finally, according to the above-mentioned top-bottom initial mode and update mode, a total of 200 color features are extracted from the two scales of the distorted screen image, and these features constitute the texture feature vector of perceptual quality, and the vector is defined as follows:

in, is the connection operation, Feature vector representing the input to the SVR model.

Calculation of screen image quality:

For distorted screen images, 520-dimensional features are obtained through the above operations, including 320-dimensional color statistical features and 200-dimensional texture statistical features. The color statistical features include 160-dimensional color features extracted based on zero-order invariant descriptors and 160-dimensional color features extracted based on first-order invariant descriptors. The texture statistical features include 40-dimensional texture features based on the top mode and bottom mode of the local three-value mode, and 160-dimensional texture features based on the top update mode and bottom update mode of the local three-value mode. In the present invention, the machine learning method SVR is used as the mapping function of the final quality score. In the experiment, the screen images in the screen image database are randomly divided into training set and test set 1000 times, of which 80% are training data sets and the remaining 20% are test data sets.

In addition, in order to evaluate the fit of the quality predicted by the proposed model and the subjective scores, the logistic regression function shown below was used to map the predicted quality scores to the common space, and non-linear correlations were eliminated.

Among them, f(x) represents the prediction quality score after nonlinear logistic regression, and {β ₁ , β ₂ , β ₃ , β ₄ , β ₅ } are set parameters.

Table 1: The performance comparison of the present invention and other different full reference quality evaluation method models in databases SIQAD and SCID:

Table 1 compares the experimental results of different image full reference quality evaluation methods. From these experimental results, the proposed screen image quality evaluation method is more correlated with subjective evaluation.

Table 2: The performance comparison of the present invention and other different non-reference quality evaluation method models in databases SIQAD and SCID:

Table 2 compares the experimental results of different image no-reference quality evaluation methods. From these experimental results, the proposed screen image quality evaluation method is more correlated with subjective evaluation.

Table 3: Performance comparison using different features in databases SIQAD and SCID:

Table 3 is a comparison of experimental results using different features to predict screen image quality.

The above-mentioned embodiment is an illustration of the present invention, not a limitation of the present invention. It can be understood that various changes, modifications, replacements and modifications can be made to these embodiments without departing from the principle and spirit of the present invention. The protection scope of the present invention It is defined by the appended claims and their equivalents.

Claims (5)

1. a screen image quality evaluation method based on color and texture features, its feature extraction comprises the following steps: A. Use the color invariance descriptor to extract the color statistical features in the screen image, and evaluate the spatial color continuity of the screen image; B. Use local ternary mode operation operator and logarithmic derivative to extract local texture statistical features to describe the spatial texture continuity of screen images; C. Using support vector regression as the mapping function to map image perceptual features to subjective scores. 2 . The method for evaluating the quality of screen images based on color and texture features according to claim 1 , wherein the visual features included include: color statistical features and texture statistical features. 3 . 3. a kind of screen image quality evaluation method based on color and texture feature according to claim 2, is characterized in that: described visual feature comprises color statistics feature, and color statistics feature is by zero-order color invariance descriptor and first-order color invariant descriptor Color invariance descriptors are obtained. 4. A color and texture feature-based screen image quality evaluation method according to claim 3, characterized in that: color statistics are extracted by two kinds of zero-order color invariance descriptors and two kinds of first-order color invariant descriptors features, the specific steps are: A. Using zero-order color invariance descriptors saturation φ and chroma ψ to extract color statistical features, saturation is defined as: Among them, R, G, B represent the colors of the red, green and blue channels of the image respectively; considering the spatial correlation, calculate the image φ(i, j)' after removing redundant information, and the four adjacent pixels The directional saturation difference is calculated as follows: φ _h (i, j)′=φ(i, j+1)-φ(i, j) (2) φ _v (i, j)′=φ(i+1, j)-φ(i, j) (3) φ _d1 (i, j)′=φ(i+1, j+1)-φ(i, j) (4) φ _d2 (i, j)′=φ(i+1, j-1)-φ(i, j) (5) Where h represents the horizontal direction, v represents the vertical direction, d1 represents the main diagonal direction, d2 represents the secondary diagonal direction; φ(i, j) represents the saturation of the pixel in the i-th row and the j-th column; Color features are extracted by histogram statistics of the absolute values of φd(i,j)′(d∈{h,v, _d1 ,d2}) and binning c ₁ ∈ {φ′ _h , φ′ _v , φ′ _d1 , φ′ _d2 }, the statistical histogram of relative saturation is calculated using the following model: where m is the histogram block index m∈[1, 10], and d represents four directions, namely d∈{h, v, d ₁ , d ₂ }, H and W are the height and width of the image, respectively, and i∈{1,2,…,H},j∈{1,2,…,W}; The luminosity and reflection invariance of the image is measured based on the chromaticity descriptor. First, the image is converted from the RGB color space to the OC space. The conversion formula is: where O1 and O2 represent color components that do not change with photometric and geometric transformations, and O3 is the luminance component; chromaticity is defined as: Similarly, the chromaticity difference between adjacent pixels is used to extract features based on the chromaticity of the image. The calculation formula of the chromaticity difference in the four directions is as follows: ψ _h (i, j)′=Γ(ψ(i, j+1), ψ(i, j)) (9) ψ _v (i, j)′=Γ(ψ(i+1, j), ψ(i, j)) (10) ψ _d1 (i, j)′=Γ(ψ(i+1, j+1), ψ(i, j)) (11) ψ _d2 (i, j)′=Γ(ψ(i+1, j-1), ψ(i, j)) (12) Among them, ψ(i, j) represents the chromaticity value on the screen image space (i, j), the chromaticity difference ψ _d (i, j)′ represents the color angle difference between adjacent pixels, and d∈{h , v, d ₁ , d ₂ }, the color angle difference operation Γ can be defined as: Similar to the saturation descriptor, histogram statistics are performed on the chrominance difference map c ₂ ∈{ψ′ _h , ψ′ _v , ψ′ _d1 , ψ′ _d2 }, and further extract the corresponding color features; B. The first-order color invariance descriptor is used to extract color statistical features on the vertex angle θ and spherical angle δ, and the vertex angle is defined as: in and represent the Gaussian partial derivatives of O1 and O2, respectively, and the Gaussian function with parameter σ is defined as: definition where R _x , G _x and B _x are the first-order Gaussian partial derivatives of the R, G, and B channels of the color image, respectively; the vertex angle can be calculated in four directions, namely θ _h (i, j)′, θ _v (i, j)′, θ _d1 (i, j)′, θ _d2 (i, j)′; and the corresponding statistical histograms are calculated c ₃ ∈{θ′ _h , θ′ _v , θ′ _d1 , θ′ _d2 }, and further extract color features according to the histogram; The last color invariant descriptor spherical angle can be defined as: in is defined as and represents the dot product operation, in addition, can be defined as: in Similarly, calculate the spherical angles in the four directions, namely δ _h (i, j)', δ _v (i, j)', δ _d1 (i, j)', δ _d2 (i, j)', and Further calculate the corresponding statistical histogram c ₄ ∈{δ′ _h , δ′ _v , δ′ _d1 , δ′ _d2 }, so as to extract the corresponding color statistical features according to the histogram; Finally, a total of 320 color features are extracted from the two scales of the distorted screen image through the above four color invariance descriptors, and these features constitute the color feature vector of perceptual quality. The calculation formula is as follows: in, is the connection operation, Feature vector representing the input to the SVR model. 5. a kind of screen image quality evaluation method based on color and texture feature according to claim 3, is characterized in that: extract texture statistical feature by local ternary mode operation operator, and its concrete steps are: Because the local ternary mode operation can extract important information sensitive to image distortion, the distorted screen image is firstly divided into two channels, the top mode (u) and the bottom mode (l), according to the local three-value mode operation. According to formulas (6) and (7), the corresponding statistical histogram can be calculated; where m is the histogram block index m∈[1, 10], and d represents four directions, namely d∈{h, v, d ₁ , d ₂ }, H and W are the height and width of the image, respectively, and i∈{1,2,…,H},j∈{1,2,…,W}; Use the statistical histogram to extract the corresponding texture statistical features respectively, namely and and m∈[1, 10]. According to the existing research, the texture information of the image can be well represented by the logarithmic derivative of the image, which can be defined as the difference in intensity values in four directions of adjacent pixels; The logarithmic derivative of the mode is used to extract the corresponding texture features, and the updated modes u(i, j)' and l(i, j)' are obtained by performing logarithmic operations on the top mode and bottom mode of the image. The calculation formula is as follows : u(i,j)′=log[u(i,j)+C] (18) l(i,j)′=log[l(i,j)+C] (19) Among them, C is a constant; to further calculate the spatial correlation between adjacent pixels in the top mode, the calculation formula is as follows: Similarly, the spatial correlations between adjacent pixels of the bottom pattern are also calculated, and then the corresponding statistical histograms are further calculated based on these calculated correlation maps, i.e. and Thereby, the corresponding texture statistical features are extracted according to the histogram; Finally, a total of 200 color features are extracted from the two scales of the distorted screen image through the above-mentioned top mode and bottom mode, and these features constitute the texture feature vector of perceptual quality. The calculation formula is as follows: in, is the connection operation, Feature vector representing the input to the SVR model.