Image quality evaluation method based on scene content and visual perception thereof

Abstract

The invention discloses an image quality evaluation method based on scene content and visual perception thereof, relates to the technical field of image communication, and provides an image quality evaluation method and model in combination with real scene content characteristics of an image and visual characteristics of masking, contrast sensitivity, brightness perception nonlinearity and comfort of a human visual system. In the method, firstly, sensed image information is divided into basic visual information, image texture and complexity sensing visual information, image definition and sensing visual information; respectively calculating the quality scores of the three parts of visual information by adopting the mode of defining the image quality by using the contrast ratio, and weighting and synthesizing the quality scores of the three parts of visual information; and finally, processing the quality scores of all the sub-blocks by adopting a global average pooling method to obtain an image quality evaluation score. According to the method, visual characteristics are fully utilized, a proper amount of effective image characteristics are extracted, and it is effectively guaranteed that the quality evaluation model has good precision, generalization and complexity.

Description

Technical Field

[0001] This invention relates to the field of image communication technology, and specifically to an image quality evaluation method based on scene content and its visual perception. Background Technology

[0002] Currently, images have permeated all aspects of people's lives, and people expect to obtain excellent terminal image display effects and service quality. However, in reality, network instability, different encoding and decoding modes, and uncertainties in transmission conditions often damage images, directly reducing user experience and service quality. Objective Image Quality Assessment (IQA) aims to design a mathematical calculation model that can automatically, quickly, and accurately assess the quality of image visual information, enabling it to play a more important role in image communication systems. This model not only automatically assesses image quality but also tracks and monitors it in real time, providing technical support for image transmission and quality assurance.

[0003] In recent years, extensive research has been conducted in the field of IQA, with two main approaches: (1) in the spatial domain of the image, constructing an IQA model based on the pixel differences between the original and distorted images; and (2) in the frequency domain, extracting various image features and constructing an IQA model based on the effective feature differences of the image, combined with the characteristics of the Human Visual System (HVS) and mathematical models. Based on these approaches, the typical models proposed so far include: mean squared error (MSE), peak signal-to-noise ratio (PSNR), visual signal-to-noise ratio (VSNR), structural similarity index (SSIM), and feature similarity index (FSIM). However, existing IQA metrics primarily focus on the impact of transmission distortion on image quality, rarely considering the real-world scene content and visual perception information of the image. This significantly reduces evaluation accuracy, and a discrepancy remains between the IQA score and the Mean Opinion Score (MOS). Furthermore, some IQA methods and models extract more image features to improve accuracy, making the IQA models more complex. Simultaneously, the increased focus on more image features limits their generalization performance. Therefore, despite the numerous IQA methods proposed, they still fall far short of practical needs. Thus, there is a need to design a simple, convenient, and effective IQA method and model with good generalization performance and high accuracy, conforming to HVS visual perception. Summary of the Invention

[0004] The purpose of this invention is to fully utilize the characteristics of the human visual system, extract appropriate and effective image features, and design and propose a simple, convenient, and effective image quality evaluation method and model that has good generalization performance and high accuracy and conforms to HVS visual perception.

[0005] This invention adopts the following technical solution: an image quality evaluation method based on scene content and its visual perception, comprising the following steps: S1. Construction of Basic Visual Information of the Image. The image is divided into sub-blocks of 8 pixels × 8 pixels. For each sub-block, the image is processed using a nonlinear model of human visual system (HVS) brightness perception to obtain an intensity-perceived image. Since the perceived intensity information is affected by local contrast and HVS sensitivity, the HVS sensitivity value and local average contrast of each point in the image are calculated, and these are used to weight the intensity-perceived image.

[0006] The impact of S2 and HVS perceptual characteristics on basic visual information of images. Considering the influence of HVS perceptual comfort on image quality assessment (IQA), and the contribution and influence of brightness on image presentation quality, a perceptual comfort model and a model of the influence of brightness on image presentation are used to process the weighted results. The processed data are then synthesized to obtain basic information from instinctively perceiving image (BIIP).

[0007] S3. Image Texture Analysis. The texture distribution of the image scene content is calculated using the gray-level gradient co-occurrence matrix, and the image texture features are quantified based on its statistical data.

[0008] S4. Image Texture Information Entropy and its Calculation. Using Huffman entropy coding, the definition and calculation method of image texture information entropy are proposed, and the image texture complexity is calculated.

[0009] S5. Texture Complexity Aware Quantization Calculation. The HVS complexity awareness model is used to process the image texture information entropy. The sum of the processing result and the quantized texture feature value is used as the visual information from perceiving the texture and its complexity, IPTC.

[0010] S6. Image Sharpness Calculation. Image sharpness is described and calculated using sharpness, signal-to-noise ratio (SNR), high-frequency component ratio (PHFC), and resolution. Then, the sharpness is graded and measured, and the result is used as the visual information from perceiving the sharpness of the image (VIPC) in HVS.

[0011] S7. Contrast-Based Image Quality Assessment Method. Combining the characteristics of human contrast perception and based on the definition of contrast, an image quality assessment (IQA) calculation model is proposed.

[0012] S8. Image perceptual information component quality evaluation. The proposed image quality calculation method is used to evaluate the BIIP, IPTC, and VIPC of the image respectively to obtain its IQA score. The BIIP, IPTC, and VIPC evaluation scores of the image are used as image quality components. The three scores are weighted according to a certain coefficient, and the weighted result is used as the IQA score of the image sub-block.

[0013] S9. Image Quality Assessment Score. For each sub-block, the above method is applied to obtain a score for each sub-block. Global average pooling is then used to process the IQA scores of all sub-blocks to obtain the overall IQA score of the image. This leads to the proposal of an IQA method based on image scene content and its visual perception, denoted as MCVP.

[0014] Compared with the prior art, the present invention, employing the above technical solution, has the following technical effects: 1. This invention combines the contrast sensitivity, brightness perception nonlinearity, visual perception comfort, and texture complexity perception characteristics of the human visual system, and proposes an image quality evaluation method and model based on image scene content information and features such as image brightness, chroma, texture, sharpness, and local contrast.

[0015] 2. The method proposed in this invention fully utilizes visual characteristics to extract appropriate and effective image features, effectively ensuring the accuracy, generalization, and complexity of the proposed image quality assessment model. The PLCC value of the proposed model can reach a maximum of 0.9683 and a minimum of 0.8969 in six databases (LIVE, TID2013, CSIQ, IVC, SPAQ, KonIQ-10k), and the weighted average of the six databases can reach 0.9171. Compared with the existing IQA model's PSNR (Peak Signal-to-Noise Ratio) and SSIM (Structural Similarity Index), based on the IQA results in three databases, the accuracy parameters PLCC and SROCC values ​​of the proposed model are improved by an average of 26.58% and 22.71%, and 10.66% and 13.67%, respectively. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the overall implementation of the present invention; Figure 2 These are the test results of the IQA model proposed in this embodiment of the invention in six databases; Figure 3 This invention compares the complexity and accuracy of the proposed model with 12 existing IQA models based on computation time and PLCC value in the embodiments of the present invention. Detailed Implementation

[0017] The present invention will be further described below with reference to specific embodiments and accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and should not be construed as limiting the scope of protection of the present invention.

[0018] To achieve the above objectives, this invention proposes an image quality evaluation method based on scene content and visual perception. Combining the characteristics of real-world scene content and the visual characteristics of the human visual system, such as masking, contrast sensitivity, nonlinearity of brightness perception, and comfort, a method and model for image quality evaluation are proposed. In this method, the perceived image information is first divided into basic visual information, image texture and complexity-perceived visual information, and image sharpness and its perceived visual information. Then, based on a visual perception model, the three parts of information are quantitatively described and calculated using image brightness, chromaticity, texture information entropy, local contrast, and sharpness. Next, the proposed contrast-based image quality definition model is used to calculate the quality scores of the three parts of visual information, and these scores are then weighted and synthesized. Finally, a global average pooling method is used to process the quality scores of all sub-blocks to obtain the image quality evaluation score. Example

[0019] An image quality assessment method based on scene content and its visual perception, specifically including: S1. Construction of basic visual information of the image: The image is divided into sub-blocks of size 8 pixels × 8 pixels. For each sub-block, the image is processed using the HVS brightness perception nonlinear model to obtain the intensity perception map. The HVS sensitivity value and local average contrast of each point in the image are calculated and weighted onto the intensity perception map.

[0020] The intensity nonlinear sensing model processing, local contrast, and HVS sensitivity calculations are as follows.

[0021] (1) Intensity nonlinear sensing model processing: Since the brightness and color intensity of HVS-sensing images change non-linearly, an HVS intensity-sensing non-linear model is used to process the image to obtain intensity-sensing image I. P The HVS intensity-sensing model is shown in equation (1): ; In the formula, k R , k G , k B It is a constant for color mixing in RGB images; I R I G I B These are the luminance and chromaticity intensities of the R, G, and B component images, respectively; I P_R I P_G and I P_B It is the intensity information of the perceived image; x , y () represents the spatial coordinates of the image.

[0022] (2) Calculation of local contrast: Local contrast is a key factor in HVS image recognition. When viewing an image, each point on the image is influenced by multiple neighboring pixels. Therefore, in the R, G, and B components, the contrast between the point and its eight nearest neighbors, and its average value C, are calculated respectively. A Contrast ratio is calculated using equation (2): ; In the formula, m and n It represents the number of rows and columns in the image.

[0023] (3) HVS sensitivity calculation: The reciprocal of the HVS contrast perception threshold is generally denoted as the HVS sensitivity. The HVS contrast sensitivity function (CSF) is used to calculate the HVS contrast perception threshold, i.e., equation (3).

[0024] ; In the formula, a , b and c For parameters, a It is obtained by substituting brightness and angular frequency into the calculation. b Obtained through calculation of brightness values; L To observe the average brightness of the target, that is, the average brightness of a local sub-block of the image, w It displays the angle size in degrees; f θ Let be the angular frequency, and its calculation method is described below.

[0025] The image is transformed using Discrete Cosine Transform (DCT). The coordinates of any point in the transform domain spectrogram are (…). f x , f y The value represents the spatial frequency; its inverse DCT (IDCT) forms a spatial image, displayed as a grating pattern. This grating pattern is the same size as the original spatial image, and the number of periods of the stripes on the grating is ( f x 2 + f y 2 ) 1 / 2 Therefore, any image can be considered as a combination of many such grating patterns of different frequencies. Combining this with the spatial HVS characteristics, observing any point on an image is equivalent to observing the corresponding grating pattern at that point. f θ =(( f x 2 + f y 2 ) 1 / 2 ) / i , i It is the spatial visual angle of the grating. The angular frequency of each point on the image can be obtained from this. Substituting it into equation (3) will give the HVS sensitivity value of each point on the image.

[0026] By combining the calculation of local contrast and HVS sensitivity, the basic visual information I obtained from the HVS-sensing image is... BPCS ( x , y Calculate according to formula (4): ; In the formula, x 1 and x 2 is a constant, I BP This is the calculation result for the sub-blocks, the average local contrast. C A and HVS sensitivity S Normalization is required.

[0027] The impact of S2 and HVS perception characteristics on basic visual information of images: The results obtained in S1 were processed using the perceptual comfort model and the model of brightness affecting image presentation, respectively. The processed data were then synthesized to obtain the basic visual perception information BIIP of the image. (1) The contribution and impact of brightness on image rendering effect: Generally, the higher the image brightness, the better the image display effect and quality. Considering the influence of local brightness and different R, G, B ratios and color rendering effects, the calculation method for the influence and contribution of image brightness and chromaticity intensity on image quality is as shown in Equation (5): ; in, l 1. l 2 and l 3 represents the scaling factors for the R, G, and B components, respectively, and their values ​​are consistent with the color mixing coefficients, being 0.299, 0.587, and 0.114. , and These are the average brightness values ​​of the sub-blocks in the three-component image, respectively; Equation (5) is the calculation result of each sub-block in the image, and it is normalized. Then, the average value of the entire image is calculated according to the number of sub-blocks. The average value is used as the quality factor for the influence and contribution of IQA, denoted as F. B .

[0028] (2) The contribution and impact of visual perceived comfort on IQA: F C The contribution and influencing factor of visual perceived comfort to IQA are represented by Equation (6): ; in, α and oh These are empirical values ​​derived from the analysis of experimental data. It is the average brightness of any sub-block.

[0029] The processed data is synthesized to obtain the basic information of image visual perception, BIIP, which is calculated as shown in equation (7): ; In the formula, β i ( i=1, 2, 3, 4) are parameters.

[0030] S3. Image texture analysis: The texture distribution of the image scene content is calculated using the gray-level gradient co-occurrence matrix, and the image texture features are quantified based on its statistical data; the calculation method is as shown in equation (8): ; In the formula, H is a gray-level gradient co-occurrence matrix. gray i Grayscale degree j For gradient, m × n This represents the number of pixels in each image, and the gradient is divided into 32 levels. T( i , j A matrix is ​​used to describe the texture features of an image, including grayscale gradient distribution and size.

[0031] S4. Image texture information entropy and its calculation: The image texture complexity is calculated using the Huffman entropy coding method; the calculation method is as shown in equation (9): .

[0032] S5. Texture Complexity Aware Quantization Calculation: The HVS complexity awareness model is used to process the image texture information entropy. The sum of the processing result and the quantized texture feature value is used as the visual information IPTC for perceiving the image texture and its complexity. The calculation method is as shown in equation (10). For ease of description, the following will be used: Abbreviated as : ; In the formula, K 1. K 2 and K 3 represents the three empirical parameter values ​​of this piecewise model, obtained by fitting experimental data.

[0033] The result after processing is summed with the quantized texture feature value as the visual information IPTC of the perceived image texture and its complexity, and its calculation is as shown in equation (11): ; in, or It is a parameter, and also an experimental value; in this invention, it is set to 0.372. T B The texture feature value of each sub-block is calculated using equation (8), T BP This is because it takes into account the results of HVS perception.

[0034] S6. Image sharpness calculation: Image sharpness is described and calculated using sharpness, signal-to-noise ratio, high-frequency component ratio, and resolution; then, the sharpness is graded and measured, and the result is used as the visual information VIPC for HVS-perceived image sharpness. The calculation method is as shown in equation (12): ; In the formula, m × n This represents the number of pixels in an image. dx This represents the distance increment (i.e., pixel interval). d I represents the magnitude of grayscale change. For signal-to-noise ratio (SNR), the type and magnitude of noise in an image are determined by the image's mean, variance, kurtosis, and skewness. For percentage of high-frequency components (PHFC), the more high-frequency components an image contains, the sharper it appears; conversely, the fewer high-frequency components, the blurrier it appears. Therefore, PHFC is measured by the proportion of high-frequency components. f s This represents spatial frequency. Image resolution is typically a fixed value and is considered a constant. K r .parameter c 1. c 2. c 3 and c 4 is an empirical value. In practice, each factor is normalized and then substituted into equation (12) for calculation.

[0035] Subsequently, each value of each sub-block in the entire image is normalized and averaged; the result is considered the image sharpness. The calculated values ​​are then graded to measure the visual information of image sharpness, thereby simulating the perception and understanding of the human visual system (HVS). C SSPR VIPC serves as visual information for HVS to perceive image sharpness.

[0036] S7. Contrast-based image quality calculation method: Combining the characteristics of human contrast perception, based on the definition of contrast, an image quality evaluation calculation model is proposed; the calculation method is as shown in equation (13): ; In the formula, f 1 and f 2 represents the real scene content information and visual perception information of the original image and the distorted image, respectively.

[0037] S8. Image perceptual information component quality evaluation: The BIIP, IPTC and VIPC of the image are evaluated using the aforementioned steps to obtain their IQA scores. The BIIP, IPTC and VIPC evaluation scores of the image are used as image quality components. The three scores are weighted and the weighted result is used as the IQA score of the image sub-block. The sum of BIIP, IPTC, and VIPC is taken as the visual information of the entire image. Using the IQA calculation method in S7, i.e., equation (13), the quality scores of BIIP, IPTC, and VIPC are calculated respectively and denoted as Scores. BIIP_B Scores IPTC_ B and Scores VIPC_B These three scores are then weighted according to a certain coefficient, and the weighted result is used as the IQA score of the image sub-block, denoted as MCVP. B Its calculation is as shown in equation (14): ; In the formula, m 1. m 2 and m 3 is a weighting factor, which in this invention takes values ​​of 0.591, 0.183 and 0.226, respectively, and are derived from the training results of experimental data.

[0038] S9, Image Quality Assessment Score: For each sub-block, the above method is used to process it and obtain the score of each sub-block. The global average pooling method is used to process the IQA scores of all sub-blocks to obtain the IQA score of the image, which is denoted as MCVP. Its calculation is as shown in equation (15): .

[0039] To illustrate the performance of the IQA model proposed in this invention, its accuracy, generalization performance, and complexity are tested and analyzed below, mainly including: 1) model accuracy and generalization performance testing and analysis, and 2) model complexity analysis. Details are as follows.

[0040] 1) Model accuracy and generalization performance testing and analysis Objectively, the accuracy of IQA models is generally described by the correlation between subjective and objective IQA scores, specifically the correlation between the subjective opinion score (MOS / DMOS) of the distorted image and the evaluation score calculated by the proposed model. The main correlation parameters include the Pearson Linear Correlation Coefficient (PLCC), the Spearman Rank Order Correlation Coefficient (SROCC), the Root Mean Square Error (RMSE), and the Outlier Ratio (OR). Higher PLCC and SROCC values, and lower RMSE and OR values, indicate better consistency between subjective and objective evaluations, and thus better model performance; conversely, lower values ​​indicate lower consistency. Subjectively, comparing scatter plots of subjective and objective IQA scores reveals a smaller dispersion of points, indicating better correlation and model performance. Generalization performance is reflected by testing the model's accuracy on different databases.

[0041] Using the methods and models of this invention, images from six open-source databases (TID2013, LIVE, CSIQ, Koniq10k, SPAQ, and IVC) were tested and analyzed. A total of 6430 images were selected from the six databases (3000, 779, 866, 800, 800, and 185 images respectively). An objective quality evaluation score was calculated for each image. Combining the subjective IQA scores MOS / DMOS from the open-source databases, four correlation parameters between the subjective and objective IQA scores were calculated; simultaneously, a scatter plot of the subjective and objective IQA scores was plotted. The results are shown below. Figure 2 The curves in the figure are all the results of fitting using the Logistic function (5 parameters).

[0042] Test results using the method of this invention Figure 2 From the results, (1) the PLCC value of the correlation parameter reflecting the model performance in the six image databases can reach a maximum of 0.9683 and a minimum of 0.8969, indicating that the proposed model has high accuracy; (2) for each of the six databases, the overall evaluation effect is good and the accuracy is high, and the weighted PLCC accuracy of the six databases reaches 0.9171; this indicates that the proposed model not only has high accuracy, but also good generalization performance; (3) combined with the effect of the scatter plot, that is, subjectively judging from the degree of dispersion of the points, the degree of dispersion of the points in the six scatter plots is low, indicating that the subjective and objective IQA scores have good consistency. Combining the four consistency parameter values ​​and the scatter plot, it is shown that the model has good accuracy and generalization performance, and the model evaluation effect is good.

[0043] To illustrate the advantages of the IQA model proposed in this invention, IQA results from three international open-source image databases (LIVE, TID2013, and CSIQ) were used. The subjective IQA scores MOS / DMOS provided in these databases were used to calculate the consistency correlation parameters PLCC and SROCC. Based on these values, the proposed IQA model was compared and analyzed with 15 IQA models proposed in recent years or classic models. The results are shown in Table 1.

[0044] Table 1. Performance comparison of the proposed model with 15 existing models based on IQA results from three databases for PLCC and SROCC.

[0045] Comparative analysis of the experimental results of each IQA model in Table 1 shows that: (1) Compared with 15 existing IQA models, the proposed model shows higher accuracy in all three databases, with the difference being 7.81% and 8.86% higher in the weighted average PLCC and SROCC values ​​in the three databases compared with the 15 models; (2) PSNR and SSIM are currently widely used IQA models, and the PLCC and SROCC values ​​of the proposed model are 26.58% and 22.71% higher than these two models, and 10.66% and 13.67% higher, respectively; (3) PEL and DeepSKLD are IQA models based on deep learning proposed in recent years, and their accuracy parameter values ​​are very close to those of the proposed model in terms of PLCC and SROCC results. In summary, the proposed model shows both high accuracy performance and good generalization performance.

[0046] 2) Model complexity analysis To illustrate the complexity of the proposed IQA model MCVP and provide a reference for its application, the complexity of the proposed model is analyzed and compared with the complexity of 12 existing IQA models (i.e., PSNR, SSIM, PSNRHVS, GMSD, VSI, MSIM, FSIMc, VSNR, DISTS, PEL, DeepSKLD, and LPIPS models).

[0047] To quantitatively describe the complexity of an IQA model, the average runtime for evaluating each image is typically used as a metric. Therefore, 200 images from three databases were randomly selected, and Matlab was used in a 64-bit Windows 10 PC with an x64 processor to evaluate them, obtaining the average runtime for each image. Based on the weighted PLCC values ​​of the proposed IQA model from the three databases, the results (computation time and PLCC values) were compared with those of 12 existing IQA models, yielding a combined analysis of accuracy and complexity. Figure 3 In the figure, the computation time of the model is taken as a relative value of PSNR.

[0048] Comparative analysis of model accuracy and computation time Figure 3 The results show that: (1) The complexity of the proposed model is lower than that of the eight existing IQA models, and its weighted PLCC and SROCC values ​​in the three databases are significantly higher than those of the eight existing models (except DeepSKLD). (2) For the four existing IQA models PSNR, SSIM, GMSD and PEL, the complexity of the proposed model is significantly higher than that of PSNR and GMSD, and closer to that of SSIM and PEL. However, in terms of accuracy, whether in the results of each of the three databases or in its weighted results, it is significantly better than PSNR and SSIM, and slightly higher than GMSD. The results show that the proposed model exhibits lower complexity and higher accuracy. Considering accuracy, complexity and generalization performance, the overall benefit of the proposed model is better than that of the 12 existing IQA models.

[0049] In summary, the method and model proposed in this invention have good accuracy and generalization performance, while having low complexity, and can effectively achieve image quality assessment.

[0050] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for image quality assessment based on scene content and its visual perception, characterized by, Comprise: S1, image basic visual information construction: the image is divided into sub-blocks according to 8 pixels x 8 pixels size, for each sub-block, the image is processed by using HVS brightness perception nonlinear model, the intensity perception image is obtained, the HVS sensitivity value and local average contrast of each point in the image are calculated, and the intensity perception image is weighted; S2, the influence of HVS perception characteristics on image basic visual information: the weighted results are processed by using the perception comfort model and the model of brightness affecting image presentation effect respectively, the processed data is synthesized, and the basic information BIIP of image visual perception is obtained; S3, image texture analysis: the texture distribution of image scene content is calculated by using gray level gradient co-occurrence matrix, and the image texture features are quantified according to the statistical data; S4, image texture information entropy and its calculation: the image texture complexity is calculated by using Huffman entropy coding method; S5, texture complexity perception quantization calculation: the image texture information entropy is processed by using HVS complexity perception model, and the sum of the processing results and the quantized texture feature value is taken as the visual information IPTC of perceived image texture and its complexity; S6, image definition calculation: the definition of image is described by using sharpness, signal-to-noise ratio, high frequency component proportion and resolution, and is calculated; then, the definition is graded and measured, and the result is taken as the visual information VIPC of HVS perceived image definition; S7, image quality calculation method based on contrast: an image quality evaluation calculation model is proposed based on the definition of contrast and the characteristics of human contrast perception; S8, image perception information component quality evaluation: the BIIP, IPTC and VIPC of the image are evaluated respectively by using the foregoing steps, the BIIP, IPTC and VIPC evaluation scores of the image are taken as the image quality components, the three scores are weighted, and the weighted results are taken as the IQA score of the image sub-block; S9, image quality evaluation score: for each sub-block, the score of each sub-block is obtained by processing according to the foregoing method, the IQA scores of all sub-blocks are processed by using global average pooling method, and the IQA score of the image is obtained.

2. The method of image quality assessment based on scene content and its visual perception according to claim 1, characterized in that, In step S1, the intensity nonlinear perception model processing, local contrast and HVS sensitivity calculation are as follows: (1) Intensity nonlinear perception model processing, as formula (1): ； wherein k R , k G , k B is a constant for RGB image color mixing; I R , I G , I B are the intensity of luminance and chrominance of R, G, B component images, respectively. I P_R I P_G and I P_B It is the intensity information of the perceived image; x , y () represents the spatial coordinates of the image; (2) Local contrast calculation: In R, G and B components, respectively calculate the contrast between this point and the surrounding nearest 8 points and their average value C A , using equation (2): ； wherein m and n are the number of image rows and columns; (3) HVS sensitivity calculation: The reciprocal of HVS contrast threshold is taken as HVS sensitivity, and HVS contrast sensitivity function CSF is used to calculate HVS contrast threshold, that is, formula (3): ； In the formula, a , b and c are parameters, a calculated by substituting the luminance and angular frequency, b obtained by luminance value calculation; L is the average luminance of the observation target, i.e., the average luminance of the local sub-block of the image, w is the display angle size in degrees; f θ is the angular frequency, f θ = ( f x 2 + f y 2 ) 1 / 2 ) / θ , θ is the spatial viewing angle of the observation grating, the coordinate ( f x , f y ) value represents the spatial frequency; The basic visual information I obtained by the HVS perceptual image is calculated in combination with the local contrast and HVS sensitivity BPCS ( x , y ) is calculated according to formula (4): ； wherein ξ 1 and ξ 2 are constants, I BP is the result of the calculation on the sub-block, local contrast average C A and HVS sensitivity S requires normalization.

3. The method of evaluating image quality based on scene content and its visual perception according to claim 1, characterized in that, In step S2: (1) The calculation method of the contribution and influence of image brightness and chrominance intensity on image quality is as formula (5): ； wherein, λ 1、 λ 2 and λ 3 are the proportional coefficients of R, G and B components, respectively, whose values are consistent with the color mixing coefficients, 0.299, 0.587 and 0.114, respectively; 、 and are the average luminance values of the sub-blocks in the three-component image, respectively; formula (5) is the calculation result of each sub-block in the image, and is normalized, and then the average value of the entire image is calculated according to the number of sub-blocks, and the average value is taken as the quality factor of the influence and contribution of IQA, denoted as F B ; (2) F C The representative visual perception comfort degree contributes to the IQA and the influence factor, and the calculation method is as formula (6): ； wherein, α and ω are empirical values based on experimental data analysis, is the average luminance of any sub-block; The processed data is synthesized, and the basic information BIIP of image visual perception is obtained, and the calculation is as formula (7): ； wherein β i ( i =1, 2, 3, 4) are parameters.

4. The method of claim 1, wherein the image quality assessment is based on scene content and its visual perception. In step S3, the calculation method is as formula (8): ； where H is a gray level gradient co-occurrence matrix, gray i is the gray level, grad j is the gradient, m × n denotes the number of pixels in each image, and the gradient is divided into 32 levels; T( i , j ) is a matrix for describing the texture features of the image, including the gray level gradient distribution and size.

5. The method of evaluating image quality based on scene content and its visual perception according to claim 4, characterized in that, In step S4, the calculation method is as formula (9): 。 6. The method of evaluating image quality based on scene content and its visual perception according to claim 5, characterized in that, In step S5, the image texture information entropy is processed by using HVS complexity perception model, and the calculation method is as formula (10): ； K 1、 K 2 and K 3 are the three empirical parameter values of this segmented model, respectively, obtained by fitting experimental data; After processing, the sum of the quantitative texture feature values is taken as the visual information IPTC of the perceived image texture and its complexity, and its calculation is as shown in formula (11): ； wherein η is a parameter; T B is calculated using equation (8) for each sub-block, T BP is the result of which takes into account the HVS perception.

7. The method of evaluating image quality based on scene content and its visual perception according to claim 1, wherein, In step S6, the calculation method is as shown in formula (12): ； wherein m x n represents the number of pixels per image, dx represents the distance increment, d I represents the amplitude of the grey level variation; f s represents the spatial frequency; the resolution of the image is a fixed value, considered as a constant K r ; the parameters γ 1, γ 2, γ 3 and γ 4 are empirical values; Then, each value of each sub-block in the whole image is normalized and the average value is taken, and the result is regarded as the sharpness of the image; and the calculated values are ranked to measure the visual information of the image sharpness, so as to simulate the perception and understanding effect of the HVS, and the calculation result C SSPR as the visual information VIPC perceived by the HVS to the image sharpness.

8. The method of evaluating image quality based on scene content and its visual perception according to claim 1, wherein, In step S7, the calculation method is as shown in formula (13): ； In the formula, f 1 and f 2 represent the real scene content information of the original image and the distorted image and their visual perception information, respectively.

9. The method of evaluating image quality based on scene content and its visual perception according to claim 8, characterized in that, In step S8, the IQA score of the image sub-block is denoted as MCVP B which is computed as in equation (14): ； wherein μ 1, μ 2 and μ 3 are weighting factors.

10. The method of evaluating image quality based on scene content and its visual perception according to claim 9, characterized in that, In step S9, the IQA score of the image is recorded as MCVP, and its calculation is as shown in formula (15): 。

Images