Multi-dimensional image quality evaluation method based on improved image information entropy

Abstract

This invention provides a multi-dimensional image quality assessment method based on improved image information entropy, relating to the field of image quality analysis technology. The method includes: constructing a spatial information function of image pixels and improving the image information entropy function using this function to obtain a spatial value entropy function; constructing an image quality assessment function including spatial value entropy, variance, and grayscale median; performing grayscale processing on the image to be evaluated, calculating the spatial value entropy, variance, and grayscale median of the grayscale image; using a particle swarm optimization algorithm to perform bidirectional adaptive calculation on the image quality assessment function to obtain its maximum and minimum values; calculating the average value based on the maximum and minimum values ​​of the image quality assessment function and normalizing the average value; and determining the image quality category based on the classification interval where the normalized image quality assessment function value falls.

Description

Technical Field

[0001] This invention relates to the field of image quality analysis technology, and in particular to a multi-dimensional image quality evaluation method based on improved image information entropy. Background Technology

[0002] With the widespread use of mobile phones and other electronic devices, images appear anytime and anywhere in daily life. Image distortion and quality degradation are inevitable during the generation, compression, and transmission processes, therefore, research on image quality assessment methods has significant application value.

[0003] Image quality assessment methods aim to automatically evaluate the objective image quality that conforms to human subjective quality judgment. These methods can be categorized into subjective assessment, full-reference assessment, partial-reference assessment, and no-reference assessment, with no-reference assessment having the greatest practical value. In the field of no-reference assessment, researchers have proposed different technical approaches to improve the accuracy of image quality assessment, but each method has its own limitations: distortion-based methods mainly extract distortion features such as blur and noise for distortion calculation, but real-world images often exhibit multiple distortions, making it difficult for traditional distortion calculation methods to accurately identify and evaluate them; salient region-based methods focus too much on salient objects as perceived by the human eye, ignoring the impact of background on image quality, leading to biased evaluation results; depth model-based methods are currently the mainstream no-reference assessment methods. These methods are highly practical and applicable to most real-world scenarios, but their drawbacks include strong data dependence, complex model structure, and high computational cost.

[0004] Therefore, no-reference quality assessment methods need to be evaluated from the perspective of multi-dimensional and multi-feature (spatial, frequency, and temporal) fusion, and to evaluate image quality at a "fine-grained" level, strengthening the evaluation of image quality with subtle differences. Summary of the Invention

[0005] To address the problems existing in the prior art, this invention proposes a multi-dimensional image quality evaluation method based on improved image information entropy. This method comprehensively considers the information content, contrast, sharpness, and brightness of the image to construct an image quality evaluation function, and uses the particle swarm optimization algorithm to automatically set the parameters in the image quality evaluation function, thereby achieving a better image quality evaluation effect.

[0006] This invention provides a multi-dimensional image quality assessment method based on improved image information entropy, comprising: A spatial information content function for image pixels is constructed, and the image information entropy function is improved using this function to obtain the spatial value entropy function of the image, as shown in the following formula: .

[0007] .

[0008] in, This represents the spatial information of an image pixel, where i represents the gray value of the image pixel, j represents the average gray value of the neighboring region of the image pixel, and p... i This represents the probability of a pixel with grayscale value i appearing in the image. This represents the maximum gray level of the image. H represents the minimum gray level of an image. E Represents the spatial entropy of an image.

[0009] An image quality evaluation function is constructed, which includes spatial entropy, variance, and gray-level median, as shown in the following formula: .

[0010] Among them, H E H represents the spatial entropy of an image. σ H represents the variance of the image. M x1 represents the median gray level of the image, and x2, x3 represent the spatial entropy, variance, and evaluation coefficients corresponding to the median gray level.

[0011] The image to be evaluated is processed into grayscale, and the spatial entropy, variance, and grayscale median of the grayscale image are calculated.

[0012] The image quality evaluation function is calculated bidirectionally and adaptively using the particle swarm optimization algorithm. The average of the maximum and minimum values ​​of the image quality evaluation function during the solution process is selected as the final image quality evaluation function value; the formula is as follows: .

[0013] in, This represents the average value of the image quality evaluation function. This represents the minimum value of the image quality evaluation function. This represents the maximum value of the image quality evaluation function.

[0014] The image quality evaluation function H is normalized using the following formula: .

[0015] Among them, H N This represents the normalized image quality evaluation function value.

[0016] The image quality category is determined based on the classification interval where the normalized image quality evaluation function value falls.

[0017] Optionally, the step of using the particle swarm optimization algorithm to perform bidirectional adaptive calculation of the image quality evaluation function includes: Spatial value entropy, variance, and gray median are used as X, Y, and Z axes respectively to form a three-dimensional attribute space. In this space, a position corresponds to a value of the evaluation coefficients x1, x2, and x3 in the image quality evaluation function, and each coefficient takes a value between [0,1]. The basic particle swarm optimization (PSO) algorithm model is simplified using the selected parameter set. The simplified PSO algorithm model is as follows: .

[0018] .

[0019] in, This represents the velocity of particle i in the j-th dimension at generation t. This represents the velocity of particle i in the j-th dimension at generation t+1. This represents the position of particle i in the j-th dimension at generation t. This represents the new position of particle i in the j-th dimension at generation t+1. This represents the optimal position of particle i in the j-th dimension at generation t. This represents the optimal position of the particle swarm in the j-th dimension at generation t.

[0020] The particle moves simultaneously along the X, Y, and Z directions in a three-dimensional attribute space. A bidirectional adaptive value selection approach is used to find the optimal solutions for the evaluation coefficients x1, x2, and x3 of the image quality evaluation function. The specific process is as follows: Set the particle swarm size, search space range, and maximum number of iterations, and randomly generate the initial particle positions and velocities.

[0021] Calculate the image quality evaluation function value for each particle in the initial particle swarm, and select the largest image quality evaluation function value as the initial maximum H value of the particle swarm and the smallest image quality evaluation function value as the initial minimum H value of the particle swarm.

[0022] Throughout the bidirectional adaptive solution process, particles whose image quality evaluation function value is closest to the initial maximum H value of the particle swarm undergo forward iteration until the maximum value of the image quality evaluation function is obtained, while particles whose image quality evaluation function value is closest to the initial minimum H value of the particle swarm undergo backward iteration until the minimum value of the image quality evaluation function is obtained.

[0023] During the bidirectional iteration process, the particle velocity and position are updated simultaneously in three directions using a simplified particle swarm optimization model.

[0024] For particles undergoing forward iteration, the image quality evaluation function value corresponding to each particle after each iteration is calculated using the particle's velocity and position after each iteration. If the image quality evaluation function value of a particle after iteration is greater than the particle's historical maximum value or the particle swarm's maximum H value, then the image quality evaluation function value of the particle after iteration is taken as the particle's new historical maximum value or the new particle swarm's maximum H value, and the evaluation coefficients x1, x2, and x3 corresponding to the new particle swarm's maximum H value are recorded. Otherwise, the historical particle swarm's maximum H value is retained until the maximum number of iterations is reached, at which point the iteration ends, and the maximum value of the image quality evaluation function is obtained.

[0025] For particles undergoing reverse iteration, the image quality evaluation function value corresponding to each particle after each iteration is calculated using the particle's velocity and position after each iteration. If the image quality evaluation function value of a particle after iteration is less than the particle's historical minimum or the particle swarm's minimum H value, then the image quality evaluation function value of the particle after iteration is taken as the particle's new historical minimum or the new particle swarm's minimum H value, and the evaluation coefficients x1, x2, and x3 corresponding to the new particle swarm's minimum H value are recorded. Otherwise, the historical particle swarm's minimum H value is retained until the maximum number of iterations is reached, at which point the iteration ends, and the minimum value of the image quality evaluation function is obtained. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 This is a flowchart illustrating a multi-dimensional image quality evaluation method based on improved image information entropy, as provided in an embodiment of the present invention.

[0028] Figure 2 This is a three-dimensional attribute space structure diagram.

[0029] Figure 3 This is a flowchart of the bidirectional adaptive computation process. Detailed Implementation

[0030] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] 1. Existing image quality evaluation functions and their characteristics.

[0032] Existing image quality assessment functions include numerous "multi-dimensional" functions. These functions primarily follow three different implementation paths. The first involves extracting image features from multiple spatial or frequency channels for evaluation, such as evaluating different resolutions of a display image, and then fusing them together to achieve multi-channel analysis. This method is computationally complex, lacks versatility, and is unsuitable for certain types of distortion. The second method comprehensively simulates multiple aspects of human visual perception, evaluating image quality from complementary feature dimensions such as contrast, structure, gradient, and natural scene statistics, and then fusing them together to achieve multi-feature, multi-dimensional fusion evaluation. This type of evaluation is the mainstream technique for no-reference evaluation. This method is computationally complex, highly dependent on the dataset, cannot consider high-order semantic features of the image, and the weight allocation for each feature dimension lacks theoretical guidance, relying more on "experience" for setting. The third method starts from different attributes and different evaluation purposes, providing more granular evaluation information, such as attribute values ​​for "blur" and "noise," to achieve multi-task, multi-attribute fusion evaluation. In this type of method, the definition of attributes is somewhat subjective, lacks objective standards, and some attributes are highly coupled.

[0033] To more comprehensively and intuitively evaluate image quality, we consider the image's grayscale features, contrast features, and statistical attributes such as entropy and brightness. Taking into account various evaluation factors, we construct an image quality evaluation function that includes spatial entropy, variance, and grayscale median from the perspectives of image features and attributes.

[0034] 2. Existing methods for calculating information entropy and their shortcomings.

[0035] Currently, image entropy calculation is mainly based on spatial domain pixel statistics, i.e., spatial domain gray-level entropy, which is the most basic and commonly used calculation method. Spatial domain gray-level entropy is a global, single statistic, only considering the probability of the image gray-level value i appearing in the image, thus losing all spatial structure information. To better reflect the spatial structure information of the image, local entropy, two-dimensional entropy, etc., have emerged. Local entropy calculates gray-level entropy within each window, reflecting texture details and local structure, but the computational cost increases significantly. In addition, the choice of window size is very important. Two-dimensional entropy combines the spatial relationships of pixels, has a strong ability to describe texture, has higher computational complexity, and is more sensitive to noise.

[0036] Local entropy and two-dimensional entropy combine the spatial relationships of pixels, considering not only the grayscale probability of a pixel but also the grayscale probability of pixels in adjacent regions, thus capturing texture information more effectively. However, each pixel in an image contains not only grayscale probability but also its corresponding grayscale value. Furthermore, the spatial structure information of an image includes not only the spatial relationships of pixels but also the spatial information of pixels, i.e., the grayscale values ​​of pixels in adjacent regions. Therefore, to comprehensively consider the spatial structure information of an image and fully utilize its grayscale and spatial information, improvements are made to the spatial domain grayscale entropy and two-dimensional entropy. First, a spatial information quantity function for image pixels is constructed. This spatial information quantity is then used to replace the grayscale probability to calculate the information entropy, which is called the Space Information Value Entropy (SIVE).

[0037] 3. Existing particle swarm optimization algorithms and their characteristics.

[0038] Particle Swarm Optimization (PSO) is a swarm intelligence algorithm with various forms, including simple PSO, improved PSO, multi-objective PSO, and hybrid PSO algorithms, with simple PSO being the foundation. It combines inertia weights, individual cognition, and social learning to balance exploration and exploitation, making it the most widely used classic form suitable for continuous function optimization and parameter optimization problems. The algorithm is simple in principle, easy to implement, and requires relatively few parameter adjustments. Particle flight is parallel, making it suitable for distributed computing. However, for complex multimodal functions, it is prone to getting trapped in local optima.

[0039] To ensure the accuracy of the image quality assessment function results, it is necessary to find the maximum and minimum values ​​of the function for data normalization. Finding the maximum and minimum values ​​falls under the category of optimal function problem. To simultaneously find the maximum and minimum values, a multi-objective PSO algorithm can be used, but this yields only a solution set, from which the optimal value needs to be selected, and the algorithm is relatively complex. To simplify and ensure a reliable solution, a simpler PSO algorithm can be run twice, separately finding the maximum and minimum values, without information sharing between the two runs.

[0040] Based on two simple PSO runs, the algorithm's execution process is improved. During algorithm initialization, the particle swarm is classified using the distance between particles. One class undergoes forward iteration to find the maximum value of its particle swarm, while the other class undergoes backward iteration to find the minimum value of its particle swarm. This achieves bidirectional adaptive calculation of the image quality evaluation function, obtaining its maximum and minimum values. The implementation process of forward and backward iteration is the same as that of a single simple PSO run.

[0041] In order to more comprehensively evaluate image quality, this invention evaluates image quality from multiple dimensions, including image information content, contrast, sharpness, and brightness.

[0042] The amount of information in an image is generally measured by its entropy. The higher the entropy value, the more information the image contains, the more complex the details, and the higher the image quality or the more noise it contains.

[0043] Image contrast refers to the degree of brightness and darkness of image pixels. Stronger contrast means the image provides more detail and information, resulting in a clearer image. Image contrast is generally evaluated using the image's variance. A larger variance indicates a more uniform distribution of gray values ​​among the pixels, generally resulting in higher contrast, better image quality, and a clearer image. Conversely, a smaller variance indicates a more concentrated distribution of gray values ​​among the pixels, generally resulting in lower contrast, poorer image quality, and a blurrier image.

[0044] The median grayscale value of an image reflects its "typical" brightness level. The median grayscale value is the middle value when all the grayscale values ​​of the pixels in the image are arranged in ascending order. A median grayscale value that is too low may indicate that the image is too dark, while a median grayscale value that is too high may indicate that it is overexposed. Therefore, the median grayscale value can be used to assess the exposure of an image.

[0045] This invention evaluates image quality from multiple dimensions, including information content, contrast, sharpness, and brightness. It constructs an image quality evaluation function and uses a particle swarm optimization algorithm to perform bidirectional adaptive calculations on each parameter in the quality evaluation function. The result of the image quality evaluation function is then normalized, and the image quality can be classified based on the normalized result of the image quality evaluation function.

[0046] Based on the above-described inventive concept, embodiments of the present invention provide a multi-dimensional image quality evaluation method based on improved image information entropy, such as... Figure 1 As shown, it includes: 1. Improve the information entropy function of images.

[0047] For a grayscale image, the information entropy function of the image is defined as: .

[0048] .

[0049] .

[0050] Where the gray level ranges from [0, L], M is the total number of pixels in the image, and m i p is the number of pixels with a given grayscale value i. i Let be the probability of a pixel with gray value i appearing in the image, and H be the information entropy function of the image.

[0051] Because image entropy does not consider pixel spatial information and grayscale values, we improve image entropy by including not only grayscale probabilities but also corresponding grayscale values. Furthermore, we consider pixel spatial information, i.e., the pixel information of adjacent regions. Therefore, when calculating image entropy, we construct a function to measure the spatial information of image pixels, in order to fully utilize the image's grayscale and spatial information. .

[0052] in, This represents the spatial information of an image pixel, where i represents the gray value of the image pixel, j represents the average gray value of the neighboring region of the image pixel, and p... i This represents the probability of a pixel with grayscale value i appearing in the image.

[0053] Using the spatial information of image pixels instead of grayscale probability to calculate image information entropy, the resulting information entropy is called spatial information value entropy (SIVE). .

[0054] in, This represents the maximum gray level of the image. H represents the minimum gray level of an image. E Represents the spatial entropy of an image.

[0055] 2. Construct an image quality evaluation function.

[0056] Spatial entropy is used to measure the information content of an image, image variance is used to measure image contrast and sharpness, and image median grayscale value is used to measure image brightness. An image quality evaluation function is constructed, comprising spatial entropy, variance, and median grayscale value, as shown in the following formula: .

[0057] Among them, H E H represents the spatial entropy of an image. σ H represents the variance of the image. M The gray median of the image is represented by x1, x2, and x3, which represent the spatial entropy, variance, and evaluation coefficients corresponding to the gray median, respectively, and are obtained adaptively through the particle swarm optimization algorithm.

[0058] 3. Bidirectional adaptive calculation of image quality evaluation function based on particle swarm optimization algorithm.

[0059] The image to be evaluated is processed into grayscale, and the spatial entropy, variance, and grayscale median of the grayscale image are calculated to calculate the image quality evaluation function value.

[0060] Using spatial entropy, variance, and grayscale median as the X, Y, and Z axes respectively, a three-dimensional attribute space is formed. A position in this three-dimensional attribute space corresponds to a value of the evaluation coefficients x1, x2, and x3 in the image quality evaluation function, with a value range of [0,1]. The values ​​of x1, x2, and x3 cannot all be 0 simultaneously. The structure of the three-dimensional attribute space is as follows: Figure 2 As shown.

[0061] There are many models of particle swarm optimization (PSO), the most basic of which is the Simple PSO Model: .

[0062] .

[0063] Where j represents the j-th dimension of the particle, i represents particle i, and t represents the t-th generation, it contains two control parameters c1 and c2, which can be regarded as acceleration constants. c1 reflects the influence of the best position (Pbest) remembered by the particle during flight on the particle's flight speed, and c2 reflects the influence of the best position (Gbest) remembered by the entire particle swarm on the particle's flight speed. r1~U(0,1) and r2~U(0,1) are two independent random functions. This represents the velocity of particle i in the j-th dimension at generation t. This represents the velocity of particle i in the j-th dimension at generation t+1. This represents the position of particle i in the j-th dimension at generation t. This represents the new position of particle i in the j-th dimension at generation t+1. This represents the optimal position of particle i in the j-th dimension at generation t. This represents the optimal position of the particle swarm in the j-th dimension at generation t.

[0064] Because the sum of parameters c2 and c1 should ideally be close to 4, and the algorithm prioritizes the influence of the swarm on individual particles during execution, the parameters are chosen as c1 = 1, c2 = 3, and r1 = r2 = 0.5. The basic particle swarm optimization (PSO) model is simplified using this set of parameters, as follows: .

[0065] .

[0066] The particle moves simultaneously along the X, Y, and Z directions in a three-dimensional attribute space. A bidirectional adaptive value selection approach is used to find the optimal solutions for the evaluation coefficients x1, x2, and x3 of the image quality evaluation function, such as... Figure 3 As shown, the specific process is as follows: Set the particle swarm size, search space range, and maximum number of iterations, and randomly generate the initial particle position and velocity. During initialization, you can choose some special positions such as (1,0,0), (0,1,0), and (0,0,1), and then select some other positions as the initial particles, but you cannot choose the position (0,0,0) to avoid the H value being 0.

[0067] Calculate the image quality evaluation function value for each particle in the initial particle swarm, and select the largest image quality evaluation function value as the initial maximum H value of the particle swarm and the smallest image quality evaluation function value as the initial minimum H value of the particle swarm.

[0068] The difference between the image quality evaluation function value corresponding to each particle and the initial maximum H value and initial minimum H value of the particle swarm is calculated, and the magnitude of the two differences is compared. If the difference is smaller than the initial maximum H value of the particle swarm, it means that the particle is closer to the initial maximum H value, and forward iteration is required to find the maximum value of the particle's evaluation function; conversely, if the difference is larger than the initial minimum H value, it means that the particle is closer to the initial minimum H value, and backward iteration is required to find the minimum value of the particle's evaluation function. Therefore, the particle swarm is divided into two, with one half undergoing forward iteration and the other half undergoing backward iteration, which are carried out simultaneously to achieve bidirectional adaptive computation.

[0069] During the bidirectional iteration process, the particle velocity and position are updated simultaneously in three directions using a simplified particle swarm optimization model.

[0070] For particles undergoing forward iteration, the image quality evaluation function value corresponding to each particle after each iteration is calculated using the particle's velocity and position after each iteration. If the image quality evaluation function value of a particle after iteration is greater than the particle's historical maximum value or the particle swarm's maximum H value, then the image quality evaluation function value of the particle after iteration is taken as the particle's new historical maximum value or the new particle swarm's maximum H value, and the evaluation coefficients x1, x2, and x3 corresponding to the new particle swarm's maximum H value are recorded. Otherwise, the historical particle swarm's maximum H value is retained until the maximum number of iterations is reached, at which point the iteration ends, and the maximum value of the image quality evaluation function is obtained.

[0071] For particles undergoing reverse iteration, the image quality evaluation function value corresponding to each particle after each iteration is calculated using the particle's velocity and position after each iteration. If the image quality evaluation function value of a particle after iteration is less than the particle's historical minimum or the particle swarm's minimum H value, then the image quality evaluation function value of the particle after iteration is taken as the particle's new historical minimum or the new particle swarm's minimum H value, and the evaluation coefficients x1, x2, and x3 corresponding to the new particle swarm's minimum H value are recorded. Otherwise, the historical particle swarm's minimum H value is retained until the maximum number of iterations is reached, at which point the iteration ends, and the minimum value of the image quality evaluation function is obtained.

[0072] To classify image quality within a precise range, the image quality evaluation function value is normalized to a range of 0 to 1. The average of the maximum and minimum values ​​of the image quality evaluation function during the solution process is chosen as the final image quality evaluation function value; the formula is as follows: .

[0073] in, This represents the average value of the image quality evaluation function. This represents the minimum value of the image quality evaluation function. This represents the maximum value of the image quality evaluation function.

[0074] The image quality evaluation function H is normalized using the following formula: .

[0075] Among them, H N This represents the normalized image quality evaluation function value.

[0076] 4. Image quality classification.

[0077] A high image spatial entropy does not always indicate better quality; it may indicate noise. A high image variance can sometimes lead to excessive contrast. A high median gray value may indicate overexposure. Therefore, a higher value in the image quality evaluation function does not necessarily mean better image quality.

[0078] Two thresholds, α and β, are set. When the normalized image quality evaluation function value is in the range [0, α], the image quality is considered poor; when it is in the range [α, β], the image quality is considered good; and when it is in the range [β, 1], the image is considered to have noise or be overexposed. Therefore, images can be classified according to the range of their quality evaluation function values: poor quality, good quality, and noisy / overexposed images.

[0079] By following the steps above, the quality of the image can be evaluated and calculated, and the image quality category can be determined based on the image quality evaluation function value.

[0080] The present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A multi-dimensional image quality evaluation method based on improved image information entropy, characterized in that, include: A spatial information content function for image pixels is constructed, and the image information entropy function is improved using this function to obtain the spatial value entropy function of the image, as shown in the following formula: ； ； in, This represents the spatial information of an image pixel, where i represents the gray value of the image pixel, j represents the average gray value of the neighboring region of the image pixel, and p... i This represents the probability of a pixel with grayscale value i appearing in the image. This represents the maximum gray level of the image. H represents the minimum gray level of an image. E Represents the spatial entropy of an image; An image quality evaluation function is constructed, which includes spatial entropy, variance, and gray-level median, as shown in the following formula: ； Among them, H E H represents the spatial entropy of an image. σ H represents the variance of the image. M The median gray level of the image is represented by x1, x2, and x3, which represent the spatial entropy, variance, and evaluation coefficients corresponding to the median gray level, respectively. The image to be evaluated is processed into grayscale, and the spatial entropy, variance, and grayscale median of the grayscale image are calculated. The image quality evaluation function is calculated bidirectionally and adaptively using the particle swarm optimization algorithm. The average of the maximum and minimum values ​​of the image quality evaluation function during the solution process is selected as the final image quality evaluation function value; the formula is as follows: ； in, This represents the average value of the image quality evaluation function. This represents the minimum value of the image quality evaluation function. This represents the maximum value of the image quality evaluation function; The image quality evaluation function H is normalized using the following formula: ； Among them, H N This represents the normalized image quality evaluation function value; The image quality category is determined based on the classification interval where the normalized image quality evaluation function value falls.

2. The multi-dimensional image quality evaluation method based on improved image information entropy according to claim 1, characterized in that, The method of using particle swarm optimization to perform bidirectional adaptive calculation of the image quality evaluation function includes: Spatial value entropy, variance, and gray median are used as X, Y, and Z axes respectively to form a three-dimensional attribute space. In this space, a position corresponds to a value of the evaluation coefficients x1, x2, and x3 in the image quality evaluation function, and each coefficient takes a value between [0,1]. The basic particle swarm optimization (PSO) algorithm model is simplified using the selected parameter set. The simplified PSO algorithm model is as follows: ； ； in, This represents the velocity of particle i in the j-th dimension at generation t. This represents the velocity of particle i in the j-th dimension at generation t+1. This represents the position of particle i in the j-th dimension at generation t. This represents the new position of particle i in the j-th dimension at generation t+1. This represents the optimal position of particle i in the j-th dimension at generation t. This represents the optimal position of the particle swarm in the j-th dimension at generation t; The particle moves simultaneously along the X, Y, and Z directions in a three-dimensional attribute space. A bidirectional adaptive value selection approach is used to find the optimal solutions for the evaluation coefficients x1, x2, and x3 of the image quality evaluation function. The specific process is as follows: Set the particle swarm size, search space range, and maximum number of iterations, and randomly generate the initial particle positions and velocities; Calculate the image quality evaluation function value for each particle in the initial particle swarm, and select the largest image quality evaluation function value as the initial maximum H value of the particle swarm and the smallest image quality evaluation function value as the initial minimum H value of the particle swarm. Throughout the bidirectional adaptive solution process, the particles whose image quality evaluation function value is closest to the initial maximum H value of the particle swarm are subjected to forward iteration until the maximum value of the image quality evaluation function value is obtained, while the particles whose image quality evaluation function value is closest to the initial minimum H value of the particle swarm are subjected to reverse iteration until the minimum value of the image quality evaluation function value is obtained. During the bidirectional iteration process, the simplified particle swarm algorithm model is used to update the velocity and position of the particles simultaneously in three directions. For particles undergoing forward iteration, the image quality evaluation function value corresponding to each particle after each iteration is calculated using the particle's velocity and position after each iteration. If the image quality evaluation function value of a particle after iteration is greater than the particle's historical maximum value or the particle swarm's maximum H value, then the image quality evaluation function value of the particle after iteration is taken as the particle's new historical maximum value or the new particle swarm's maximum H value, and the evaluation coefficients x1, x2, and x3 corresponding to the new particle swarm's maximum H value are recorded. Otherwise, the historical particle swarm's maximum H value is retained until the maximum number of iterations is reached, at which point the iteration ends, and the maximum value of the image quality evaluation function is obtained. For particles undergoing reverse iteration, the image quality evaluation function value corresponding to each particle after each iteration is calculated using the particle's velocity and position after each iteration. If the image quality evaluation function value of a particle after iteration is less than the particle's historical minimum or the particle swarm's minimum H value, then the image quality evaluation function value of the particle after iteration is taken as the particle's new historical minimum or the new particle swarm's minimum H value, and the evaluation coefficients x1, x2, and x3 corresponding to the new particle swarm's minimum H value are recorded. Otherwise, the historical particle swarm's minimum H value is retained until the maximum number of iterations is reached, at which point the iteration ends, and the minimum value of the image quality evaluation function is obtained.

Images