Image enhancement optimization method based on multi-strategy improved sparrow search

Abstract

The invention provides an image enhancement optimization method based on multi-strategy improved sparrow search. The method comprises the following steps: performing normalization processing on a gray value of each pixel point of an original image; initializing a sparrow population and parameters according to an elite chaos reverse learning strategy; the method comprises the following steps: preprocessing an input image by adopting a dual gamma function BiGA; calculating the value of the evaluation function; updating the optimal fitness value of the sparrow population; when an iteration stop condition is not met, updating the position of the sparrow population by adopting a probabilistic following strategy, a self-adaptive local search strategy and a Cauchy Gaussian mutation strategy, and updating parameters of a dual gamma function BiGA; otherwise, ending iteration, and calculating an optimal parameter; processing the image, and outputting an enhanced image; according to the method, the visual effect of the image can be improved, edge details and textures can be richer, and the enhanced image can have higher contrast and better definition. The method is best in overall enhancement effect, and the obtained image is better in quality.

Description

Technical Field

[0001] The invention relates to an image enhancement optimization method based on multi-strategy improved sparrow search, belonging to the technical field of image enhancement. Background Art

[0002] With the development of new media technology, communication media such as text, voice, pictures, and videos are increasingly attracting public attention. Among them, images are the most intuitive way for people to understand the world. They will directly or indirectly change people's production and lifestyles and provide great convenience for people's lives. However, in the actual shooting process, due to the complexity of shooting conditions, especially under night conditions, video images are often low in brightness, large in noise, and low in contrast. It is difficult to achieve the ideal acquisition effect and obtain the desired information. These images that are disturbed by external factors during the shooting process are called low-quality images. These low-quality images will reduce people's perception of information and affect the role of images in traffic intelligence, network monitoring, military defense, and medical fields. Low-quality images are usually characterized by a serious lack of information, which not only damages social production interests, but also threatens social public safety and public security management. The most common low-quality images are low-light low-quality images. Due to the position, intensity, and color of the light source, the generation of low-light images occurs from time to time. Therefore, it is important to enhance the images formed under low-light conditions. This research not only occupies an important position in the field of image processing, but also provides a solid technical guarantee for the stable development of society.

[0003] In recent years, there have been extensive studies on low-light image enhancement methods at home and abroad. Traditional image enhancement methods are roughly divided into two categories: spatial domain image enhancement and frequency domain image enhancement. Specifically, they include tone mapping algorithm, histogram equalization algorithm, gamma correction algorithm, dark channel prior algorithm, Retinex decomposition-based algorithm, and fusion-based algorithm. The tone mapping algorithm is to apply a linear mapping function to increase the brightness of the image, which is the simplest and most direct way to enhance low-light images. Since this type of method is prone to problems such as oversaturation and loss of detail information, researchers have proposed using nonlinear mapping functions to adjust the image. Although it effectively avoids oversaturation of the image, it still does not consider local information. Histogram equalization is an algorithm that enhances the image from a global perspective. The problem of image detail loss caused by direct mapping is solved by expanding the pixel range of the image. However, because the algorithm does not take into account the spatial distribution of pixels, the enhanced image appears overexposed. In order to solve the above problems, many scholars have proposed different algorithms. For example, the adaptive histogram equalization algorithm has limited enhancement capabilities for low-light images and poor enhancement effects on some areas of the image. Wang Lijuan et al. proposed a weighted histogram equalization method, which solves the problem of detail loss by weighting. Jebadass et al. combined fuzzy set theory, first blurred the image, and then used the adaptive histogram equalization method with limited contrast to enhance the image, further improving the image quality. However, since this method directly stretches the histogram of the image, it will lead to over-stretching and then produce a fast effect that degrades the image quality. The gamma function is also a common method for image enhancement. For low-light images, gamma correction is simple and fast, but it is easy to produce color distortion and insufficient contrast. Therefore, Hou Lixia et al. proposed an adaptive gamma correction image enhancement algorithm, which improves the contrast of the image by multi-scale decomposition and combined with adaptive gamma correction. Then an improved adaptive gamma conversion image enhancement algorithm was proposed, which improved the gamma function and adaptively defined the relevant basic functions. However, gamma correction often causes color distortion of the image, is not sensitive enough to different scenes, and has poor effect. The Retinex algorithm is based on the theory of color consistency. Because it is more in line with the visual system of the human eye, a series of algorithms have emerged. First, the Retinex algorithm was proposed. This algorithm uses a path random method to calculate the relative brightness values ​​of adjacent pixels of a certain pixel within a certain range. Due to the random path selection, the results obtained are not very reliable. Pan et al. used a brightness enhancement function and an improved adaptive contrast enhancement method to obtain an enhanced image through multi-scale fusion. Yang Jie et al. proposed a low-light image enhancement algorithm that combines a fuzzy C-means clustering algorithm with the Retinex theory.Since clustering is to cluster the grayscale histogram of the image, the enhanced image has rich detail information. This method enhances the image by estimating the illumination component and eliminating the influence of the illumination component, but it is easy to cause image distortion and is not sensitive to noise, and the denoising effect is not good.

[0004] The above methods have certain effects on image enhancement, but they still have disadvantages such as low contrast, unclear detail processing and low clarity. Therefore, this application proposes a low-illumination image enhancement optimization method based on MISSA to make image enhancement more effective. Summary of the invention

[0005] The purpose of the present invention is to provide an image enhancement optimization method based on multi-strategy improved sparrow search to solve the problems of low contrast, loss of edge details and low clarity of low-illuminance images existing in the prior art.

[0006] The technical solution of the present invention is:

[0007] An image enhancement optimization method based on multi-strategy improved sparrow search includes the following steps:

[0008] S1, taking the low-light image as the input original image, and normalizing the grayscale value of each pixel of the original image;

[0009] S2, according to the elite chaos reverse learning strategy, initialize the parameters of the sparrow population and the bigamma function BiGA;

[0010] S3, using the dual gamma function BiGA to preprocess the input image;

[0011] S4, substituting each individual sparrow in the sparrow population into the bi-gamma function BiGA to enhance the image, and calculating the value of the evaluation function;

[0012] S5, updating the optimal fitness value of the sparrow population. If the fitness value of the updated sparrow individual is greater than the previous optimal fitness value of the sparrow individual, the updated position of the sparrow individual is regarded as the optimal individual; if the fitness value of the updated sparrow population is greater than the previous global optimal fitness value of the sparrow population, the updated position of the sparrow individual is regarded as the global optimal, and the fitness value of the updated sparrow population is taken as the global optimal fitness value gbest;

[0013] S6, determine whether the iteration stop condition is met, if not, adopt the probabilistic following strategy, adaptive local search strategy and Cauchy Gauss mutation strategy to update the position of the sparrow population, and update the parameters of the bigamma function BiGA, and go to step S3; if the iteration stop condition is met, end the iteration and go to the next step S7;

[0014] S7, calculating the optimal parameter value of the bi-gamma function BiGA according to the global optimal fitness value gbest of the sparrow population;

[0015] S8. After the bi-gamma function BiGA obtains the optimal parameter value, the image is processed, and finally the obtained image is denormalized to output the enhanced image.

[0016] Further, step S2 is specifically,

[0017] S21. Randomly initialize an initial sparrow population X with N individuals, and generate a cubic mapping sequence using the following formula:

[0018]

[0019] Among them, y c and c+1 is a cubic sequence;

[0020] S22, the cubic sequence y is transformed into c Mapped to sparrow individual x c Then we get the chaotic mapping population X c :

[0021]

[0022] Among them, ub and lb are the upper and lower bounds of the search space respectively;

[0023] S23, calculate the population X after chaotic mapping c The fitness value of

[0024] S24, find the population X after chaotic mapping through the following formula c The reverse solution population

[0025]

[0026] Among them, i d and j d They represent the minimum and maximum values ​​in the dth dimension of the current population, respectively, and m is the zoom factor of the lens;

[0027] S25. Calculate the reverse solution population The fitness value of

[0028] S26, population X after merging chaotic maps c and reverse solution population Sort the fitness values ​​of the merged sparrow individuals, and select the first N sparrow individuals with better fitness values ​​as the new initial sparrow population.

[0029] Furthermore, in step S4, the value of the evaluation function is calculated, specifically,

[0030] S41. Calculate the entropy value H of the image:

[0031]

[0032] Among them, p(i) represents the probability of a pixel with gray value i appearing in the image;

[0033] S42. Calculate the Sobel operator S for the edge content of the image:

[0034]

[0035] Wherein, N-edges(E) represents the number of edge pixels detected by the Sobel operator; T represents the number of all pixels in the image;

[0036] S43, calculating the grayscale standard deviation V;

[0037] S44, calculate the value F of the evaluation function:

[0038] F=μ 1 H+μ 2 S+μ 3 V

[0039] Among them, μ 1 , μ 2 and μ 3 are all constants and μ 1 +μ 2 +μ 3 =1.

[0040] Furthermore, in step S6, the position of the sparrow population is updated by adopting a probabilistic following strategy, an adaptive local search strategy and a Cauchy-Gauss mutation strategy, specifically,

[0041] S61, randomly select a set proportion of sparrow individuals in the sparrow population as scouts, sort the remaining individuals according to the fitness value, the first set proportion is used as discoverers, and the rest are all followers; use the sparrow search algorithm SSA to update the positions of discoverers, followers, and observers;

[0042] S62. Follower position update formula using probabilistic following strategy:

[0043]

[0044] in, Indicates the position of the cth sparrow individual in the search space at the n+1th iteration and the nth iteration; is the worst position of individual sparrows in the sparrow population at the nth iteration; is a position randomly selected in the sparrow population at the nth iteration, and r∈[1,N], r≠c; G is a random number that follows a normal distribution; R(0,1) is a uniformly distributed random number, f r is the fitness value of a randomly selected individual, f c is the fitness value of the individual in this round;

[0045] S63. After the probabilistic following strategy, the fitness value of each individual sparrow is calculated again, and the individual sparrows are re-sorted according to the fitness value. The individuals with high fitness values ​​are selected as discoverers and the individuals with low fitness values ​​are selected as followers according to the set ratio. An adaptive local search strategy is introduced, and the followers take the discoverer's position as the optimal solution. In the optimal solution Perform domain search to obtain the domain value of the optimal solution

[0046]

[0047] Where θ is the adaptive coefficient and D is the maximum number of iterations;

[0048] Finally, the domain value of the optimal solution With the optimal solution Compare the fitness values, select the sparrow individuals with high fitness values ​​as the new optimal solution, and then obtain a new sparrow population;

[0049] S64, based on the new sparrow population obtained after the adaptive search, the sparrow population is sorted according to the fitness value, and the sparrow individual with the highest fitness value is selected Perform Cauchy-Gauss mutation to obtain the position of the sparrow individual with the highest fitness value after mutation

[0050]

[0051] in, is the position of the sparrow with the highest fitness value after mutation; τ is the standard deviation of the Cauchy-Gaussian mutation strategy; Gauchy(0,τ 2 ) is a Cauchy distributed random variable; Gauss(0,τ 2 ) is a Gaussian distributed random variable; μ 1 and μ 2 is an adaptive parameter that is dynamically adjusted according to the number of iterations and μ 1 +μ 2 =1; as the number of iterations increases, μ 1 Gradually decreases and μ 2 Gradually increases; f(X B ) is the highest fitness value, f(X c) is the fitness value of the current sparrow individual;

[0052] Calculate the fitness value again and compare the sparrow individuals with the highest fitness value and the position of the sparrow with the highest fitness value after mutation Finally, the best position is selected to enter the next round of iteration.

[0053] Furthermore, in step S8, the image is processed, specifically,

[0054]

[0055] Among them, BiGA(x) is the a (x) and BiGA b (x) The double gamma function after weighted sum, x is the pixel value of the normalized image, and the weight BiGA a (x) is a convex function used to enhance the dark areas of the image, BiGA b (x) is a concave function used to suppress bright areas, γ is an adjustable variable used to control the degree of image enhancement.

[0056] The beneficial effects of the present invention are as follows: the image enhancement optimization method based on multi-strategy improved sparrow search can improve the visual effect of the image, make the edge details and texture richer, make the enhanced image have higher contrast and better clarity, and the enhanced image quality evaluation index is greatly improved. Compared with the existing method, the method has better overall enhancement effect and better image quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 It is a flowchart of an image enhancement optimization method for improving sparrow search based on multiple strategies according to an embodiment of the present invention;

[0058] Figure 2 1 is a comparison diagram of the original image in the embodiment, the existing methods including GA, PSO, SSA, and the image enhancement processing by the embodiment method, wherein (a) is the original image of 4 low-illumination images, (b) is the image result of the 4 low-illumination images after being enhanced by GA, (c) is the image result of the 4 low-illumination images after being enhanced by PSO, (d) is the image result of the 4 low-illumination images after being enhanced by SSA, and (e) is the image result of the 4 low-illumination images after being enhanced by the embodiment method. DETAILED DESCRIPTION

[0059] The preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0060] The embodiment provides an image enhancement optimization method based on multi-strategy improved sparrow search, such as Figure 1 , including the following steps,

[0061] S1. Taking the low-light image as the input original image, normalizing the grayscale value of each pixel of the original image.

[0062] S2. According to the elite chaos reverse learning strategy, the parameters of the sparrow population and the bigamma function BiGA are initialized.

[0063] In step S2, the sparrow population represents a numerical sequence, and the individuals of the sparrow population represent the numerical values ​​in the sequence. Then, the numerical values ​​in the sequence are selected and substituted into the bi-gamma function BiGA to pre-process the normalized image.

[0064] S21. Randomly initialize an initial sparrow population X with N individuals, and generate a cubic mapping sequence using the following formula:

[0065]

[0066] Among them, y c and c+1 is a cubic sequence;

[0067] S22, the cubic sequence y is transformed into c Mapped to sparrow individual x c Then we get the chaotic map population X c :

[0068]

[0069] Among them, ub and lb are the upper and lower bounds of the search space respectively;

[0070] S23, calculate the population X after chaotic mapping c The fitness value of

[0071] S24, find the population X after chaotic mapping through the following formula c The reverse solution population

[0072]

[0073] Among them, i d and j d They represent the minimum and maximum values ​​in the dth dimension of the current population, respectively, and m is the zoom factor of the lens;

[0074] In step S24, the reverse learning strategy based on lens imaging evaluates the feasible solution and its reverse solution to the problem, expands the search space, increases the probability of selecting a better solution, improves the quality of the initial solution, increases the possibility of the algorithm finding the optimal solution, and reduces the blindness of the algorithm during iterative optimization.

[0075] S25. Calculate the reverse solution population The fitness value of

[0076] S26, population X after merging chaotic maps c and reverse solution population Sort the fitness values ​​of the merged sparrow individuals, and select the first N sparrow individuals with better fitness values ​​as the new initial sparrow population.

[0077] In step S2, in order to expand the global exploration range while avoiding the uncertainty of the algorithm, an elite chaos reverse learning strategy is proposed and introduced into the initialization of the sparrow population. The chaotic solution and the reverse solution are combined through the elitist idea, and elite individuals are selected in a larger range to improve the diversity of the initial sparrow population and the quality of the individual sparrows. Because the chaos operator has the characteristics of randomness, regularity and non-repetitive traversal, the cubic chaos map with better uniformity is selected to initialize the population. The quality and distribution uniformity of the initial population can be enhanced, which is helpful for a more comprehensive search in the search space, and the purpose of improving the convergence accuracy and optimization performance of the algorithm can be achieved.

[0078] S3. Use BiGA function to preprocess the input image.

[0079] S4. Substituting each individual sparrow in the sparrow population into the bi-gamma function BiGA to enhance the image, and calculating the value of the evaluation function.

[0080] In step S4, the value of the evaluation function is calculated, specifically,

[0081] S41. Calculate the entropy value H of the image:

[0082]

[0083] Among them, p(i) represents the probability of a pixel with gray value i appearing in the image;

[0084] S42. Calculate the Sobel operator S for the edge content of the image:

[0085]

[0086] Wherein, N-edges(E) represents the number of edge pixels detected by the Sobel operator; T represents the number of all pixels in the image;

[0087] S43, calculating the grayscale standard deviation V;

[0088] S44, calculate the value F of the evaluation function:

[0089] F=μ 1 H+μ 2 S+μ 3 V

[0090] Among them, μ 1 , μ 2 and μ 3 are all constants and μ 1 +μ 2 +μ 3 =1.

[0091] In step S4, the values ​​in the initial sparrow population, i.e., the initial numerical sequence, are substituted into the dual gamma function BiGA to enhance the image. The design of the fitness function is particularly important in the image optimization method. In the evaluation function of the embodiment, the entropy value H of the image reflects the richness of the information contained in the image, and the Sobel operator S of the edge content of the image is calculated. The larger the value, the richer the edge information. The larger the grayscale standard deviation V value, the clearer the image texture, and the more obvious the difference seen during analysis and observation. The embodiment method can better improve the image quality by using the above-mentioned evaluation function to weigh the ability of the algorithm to find the best.

[0092] S5. Update the optimal fitness value of the sparrow population. If the fitness value of the updated sparrow individual is greater than the previous optimal fitness value of the sparrow individual, the updated position of the sparrow individual is regarded as the optimal individual. If the fitness value of the updated sparrow population is greater than the previous global optimal fitness value of the sparrow population, the updated position of the sparrow individual is regarded as the global optimal, and the fitness value of the updated sparrow population is used as the global optimal fitness value gbest.

[0093] S6. Determine whether the iteration stop condition is met. If not, adopt the probabilistic following strategy, adaptive local search strategy and Cauchy-Gauss mutation strategy to update the position of the sparrow population, and update the parameters of the bigamma function BiGA, and go to step S3; if the iteration stop condition is met, end the iteration and go to the next step S7.

[0094] In step S6, the position of the sparrow population is updated by adopting the probabilistic following strategy, the adaptive local search strategy and the Cauchy-Gauss mutation strategy, specifically,

[0095] S61, randomly select a set proportion of sparrow individuals in the sparrow population as scouts, and sort the remaining individuals according to their fitness values. The first set proportion are used as discoverers, and the rest are all followers;

[0096] S62: Use a probabilistic following strategy to update the position of the follower, specifically:

[0097]

[0098] in, Indicates the position of the cth sparrow individual in the search space at the n+1th iteration and the nth iteration; is the worst position of individual sparrows in the sparrow population at the nth iteration; is a position randomly selected in the sparrow population at the nth iteration, and r∈[1,N], r≠c; G is a random number that follows a normal distribution; R(0,1) is a uniformly distributed random number, f r is the fitness value of a randomly selected individual, f c is the fitness value of the individual in this round.

[0099] In order to improve the global optimization capability, the embodiment adopts the above probabilistic following strategy: the follower moves closer to the discoverer with a certain probability, ensuring convergence while taking into account the global search. The improved follower position update makes full use of the position information of the individuals in the previous round.

[0100] S63. After the probabilistic following strategy, the fitness value of each individual sparrow is calculated again, and the individual sparrows are re-sorted according to the fitness value. The individuals with high fitness values ​​are selected as discoverers and the individuals with low fitness values ​​are selected as followers according to the set ratio. An adaptive local search strategy is introduced, and the followers take the discoverer's position as the optimal solution. In the optimal solution Perform domain search to obtain the domain value of the optimal solution

[0101]

[0102] Where θ is the adaptive coefficient and D is the maximum number of iterations;

[0103] Finally, the domain value of the optimal solution With the optimal solution Compare the fitness values, select the sparrow individuals with high fitness values ​​as the new optimal solution, and then obtain a new sparrow population.

[0104] In step S63, by introducing adaptive local search, local search is performed with the position of the finder as the optimal position, thereby improving the ability of SSA to avoid local optimality.

[0105] S64, based on the new sparrow population obtained after the adaptive search, the sparrow population is sorted according to the fitness value, and the sparrow individual with the highest fitness value is selected Perform Cauchy-Gauss mutation to obtain the position of the sparrow individual with the highest fitness value after mutation

[0106]

[0107]

[0108] in, is the position of the sparrow with the highest fitness value after mutation; τ is the standard deviation of the Cauchy-Gaussian mutation strategy; Gauchy(0,τ 2 ) is a Cauchy distributed random variable; Gauss(0,τ 2 ) is a Gaussian distributed random variable; μ 1 and μ 2 is an adaptive parameter that is dynamically adjusted according to the number of iterations and μ 1 +μ 2 =1; as the number of iterations increases, μ 1 Gradually decreases and μ 2 Gradually increases; f(X B ) is the highest fitness value, f(X c ) is the fitness value of the current sparrow individual;

[0109] Calculate the fitness value again and compare the sparrow individuals with the highest fitness value and the position of the sparrow with the highest fitness value after mutation Finally, the best position is selected to enter the next round of iteration.

[0110] In step S64, since the algorithm is prone to fall into a local optimum in the later stage of SSA iteration, the Cauchy-Gauss mutation strategy is introduced to select the sparrow individual with the highest fitness value in the nth iteration. Mutate, compare the previous and next positions, and finally select the best position to enter the next round of iteration. The Cauchy-Gauss mutation strategy is used to improve the overall anti-stagnation ability of the algorithm and avoid the algorithm falling into the local optimum.

[0111] S7. Calculate the optimal parameter value of the bigamma function BiGA according to the global optimal fitness value gbest of the sparrow population.

[0112] S8. After the bi-gamma function BiGA obtains the optimal parameter value, the image is processed, and finally the obtained image is denormalized to output the enhanced image.

[0113] In step S8, the mathematical expression of the dual gamma function BiGA is as follows:

[0114]

[0115] Among them, BiGA(x) is the a (x) and BiGA b (x) The double gamma function after weighted sum, x is the pixel value of the normalized image, Convex function BiGA a (x) is used to enhance the dark areas of the image: γ is an adjustable variable used to control the degree of image enhancement and is usually set to 2.5; concave function BiGA b (x) is used to suppress bright areas: The adjustment function BiGA(x) of the dual gamma enhancement algorithm is given by BiGA a (x) and BiGA b (x) is obtained by taking the weighted average.

[0116] When using the dual gamma function to correct an image, the pixel grayscale value of the image is first normalized using the following formula: Secondly, the image is enhanced by the following formula: H 2 (x,y)=BiGA(H 1 (x, y)); Finally, the grayscale value range of the enhanced image is adjusted to [0, 255] using the following formula: H(x, y) = H 2 (x,y)·256, thus obtaining the output image.

[0117] In the image enhancement process, the addition of BiGA is conducive to maintaining the average brightness of the entire image. Using MISSA to optimize BiGA to obtain the optimal parameters can well enhance the information richness of the image. Therefore, the embodiment proposes a combination of MISSA and BiGA to achieve the purpose of enhancing the image. In the proposed search algorithm, it is necessary to determine the range of the parameter γ in the dual gamma function, and then Sparrow will find the optimal γ value according to the fitness function to improve the contrast of the image.

[0118] This image enhancement optimization method based on multi-strategy improved sparrow search can improve the visual effect of the image, make the edge details and texture richer, and make the enhanced image have higher contrast and better clarity. The enhanced image quality evaluation index has been greatly improved. Compared with the existing method, this method has better overall enhancement effect and better image quality.

[0119] This image enhancement optimization algorithm is based on the multi-strategy improved sparrow search algorithm (MISSA). Firstly, the traditional sparrow search algorithm (SSA) is improved and optimized by using multi-strategy, and the elite chaos reverse learning strategy is introduced to increase the diversity of the initial population of sparrows; the accuracy and speed of the algorithm convergence are enhanced by using the probabilistic following strategy and adaptive local search optimization to update the position of the follower; the Cauchy-Gauss mutation strategy is used to improve the overall anti-stagnation ability of the algorithm and avoid the algorithm from falling into the local optimum. Then, the improved algorithm is optimized with the bi-gamma function (BiGA) to obtain the optimal parameters, and MISSA is combined with BiGA and applied to low-light images to make image enhancement more effective. It can enhance the image contrast and enrich the image details while improving the image brightness, so that the enhanced image has the best visual effect.

[0120] This image enhancement optimization method based on multi-strategy improved sparrow search is aimed at the problem that in the sparrow search algorithm SSA, the initial sparrow individuals are randomly generated in the search space, and gradually gather during the iteration process, making it difficult to obtain a good population diversity and maintain it at a certain level, resulting in poor convergence performance of the algorithm and inconsistency between the global search capability and the local development performance. SSA is improved from four aspects: optimizing the initial population, probabilistic following strategy, adaptive local search, and Cauchy-Gauss mutation strategy, and a multi-strategy improved sparrow search algorithm MISSA is obtained.

[0121] This image enhancement optimization method based on multi-strategy improved sparrow search can improve the quality of low-light images and produce better visual effects while avoiding excessive image enhancement. It also enriches the edge detail effect while ensuring image quality. That is, the image enhanced by the embodiment method is closer to the natural image while ensuring image quality, and the enhancement effect is the best.

[0122] The pseudo code of this MISSA-based image enhancement optimization method is as follows:

[0123]

[0124] The image enhancement optimization method based on multi-strategy improved sparrow search in the embodiment is experimentally verified as follows:

[0125] In order to verify the effectiveness of the image enhancement optimization method based on multi-strategy improved sparrow search, four low-light images in the low-light pairing dataset (LOL) were selected as test images. Each image was processed using four methods, namely GA, PSO, SSA and embodiment method, and the enhancement effect of each image was compared and analyzed. All experiments were implemented in a computer with a 12th generation Intel(R) Core(TM) i5-12400 (2.5GHz) CPU, 16GB of running memory, and Windows 1064-bit operating system, and run in the MATLAB R2016b environment. In the experiment, all methods set the parameter population size to 30 and the maximum number of iterations to 100. The test results of the original image and the enhanced image after processing by each algorithm are listed here. The experimental results are shown in the figure. Figure 2 As shown:

[0126] Figure 2 (a) is the original image of 4 low-light images. Their common characteristics are that the overall image is dark and the detailed texture is not clear enough. Figure 2 (b), (c) and (d) are the image results obtained after enhancement using GA, PSO and SSA algorithms respectively. The above enhanced images are compared with the images obtained after MISSA enhancement in the embodiment. Figure 2 (e) for comparison. Figure 2 The image in (b) is dark overall, indicating that GA does not significantly improve the brightness and contrast. The image is still dark overall and the visual effect is not obvious. Figure 2 The image ratio in (c) Figure 2 The image in (b) is slightly brighter, which shows that PSO is slightly better than GA. Although it improves the overall brightness of the image, the image clarity is still low and contains less information. Figure 2 The image in (d) is brightened overall, but the noise still exists. This proves that SSA improves the image brightness significantly, but the image contrast is not improved, making the entire image appear gray. Moreover, the edge details are not clear enough, and the visual effect is still poor. In contrast, Figure 2 The image in (e) is not only smooth, but also can restore texture details without overexposure. This proves that the algorithm in this paper not only improves the brightness of the image, but also enhances the contrast of the image. It enriches the detailed information of the image and has the best visual effect.

[0127] In order to evaluate the image enhancement effects of various algorithms more objectively and accurately, four commonly used image quality evaluation indicators were selected, namely, average gradient, standard deviation, information entropy and spatial frequency.

[0128] Average gradient (AG) is an indicator to measure the clarity of an image, and can characterize the ability of image detail contrast and texture change. The larger its value, the higher the image clarity. Standard deviation (SD) indicates the discrete degree of pixel grayscale value, reflecting the richness of image details. The larger its value, the better the image contrast. Information entropy (IE) reflects the richness of image information. The larger the entropy value, the richer the image information. Spatial frequency (SF) measures the activity of the image in the spatial domain, and can characterize the clarity of image details and the degree of spatial change of the image. The larger its value, the richer the edges and textures. The image enhancement metric data of the four low-light images are shown in Table 1-4:

[0129] Table 1 Evaluation index data of House image enhancement results

[0130]

[0131] Table 2 Evaluation index data of Doll image enhancement results

[0132]

[0133] Table 3 Evaluation index data of the enhanced results of the corridor image

[0134]

[0135] Table 4 Evaluation index data of door image enhancement results

[0136]

[0137] As can be seen from Tables 1-4, the objective evaluation results of low-light image enhancement performed by different algorithms are compared. It can be concluded that the embodiment method is significantly better than other algorithms in terms of SD value. The SD value of the proposed algorithm is 10% higher than that of PSO or SSA, which shows that the method proposed in the embodiment handles the image details more clearly. The data results obtained by different algorithms are also ranked on SF and AG, and the values ​​of these two evaluation indicators are 20% higher than PSO. This shows that the image processed by the embodiment method is not over-enhanced, reducing the occurrence of image distortion. This also shows that the edge details and texture of the image are clearer. Although the IE value obtained by the embodiment method is not the best, it also improves the information richness of the image as a whole. This is mainly because the images generated by other algorithms will generate noise, which will increase the amount of information in the generated image, proving that it cannot effectively improve the information richness of the image.

[0138] In summary, after 30 experimental tests on images in the LOL public dataset, the average data obtained verified that the embodiment method can improve the quality of low-light images while avoiding excessive enhancement of the image. It produces a better visual effect and enriches the edge detail effect while ensuring the image quality. The enhanced image has a higher contrast and better clarity. That is, the image enhanced by the embodiment method is closer to the natural image while ensuring the image quality, and the enhancement effect is the best.

[0139] This image enhancement optimization method based on multi-strategy improved sparrow search aims at the problems of poor visual effect, low recognition rate and blurred edges of low-light images. The traditional SSA is improved through elite chaos reverse learning strategy, probabilistic following strategy, adaptive local search and Cauchy-Gauss mutation strategy. Then, MISSA is used to optimize the gamma value in BiGA. Finally, BiGA with the best gamma value is applied to enhance low-light images, which can improve the contrast and clarity of low-light images, enrich the edge and texture details of the image, and greatly improve the visual effect, which is conducive to human eye observation and computer recognition analysis. Finally, the simulation experimental results show that the image quality evaluation index after the algorithm is enhanced has been greatly improved, the visual effect of the image has been improved, and the edge details and texture are richer. Compared with other algorithms, this method has the best overall enhancement effect and obtains better image quality.

[0140] Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An image enhancement optimization method based on multi-strategy improved sparrow search, characterized in that: The following steps are included: S1, taking the low-light image as the input original image, and normalizing the grayscale value of each pixel of the original image; S2. Initialize the parameters of the sparrow population and the bigamma function BiGA according to the elite chaos reverse learning strategy; S3, using the dual gamma function BiGA to preprocess the input image; S4, substituting each individual sparrow in the sparrow population into the bi-gamma function BiGA to enhance the image, and calculating the value of the evaluation function; S5, updating the optimal fitness value of the sparrow population. If the fitness value of the updated sparrow individual is greater than the previous optimal fitness value of the sparrow individual, the updated position of the sparrow individual is regarded as the optimal individual. If the fitness value of the updated sparrow population is greater than the global optimal fitness value of the previous sparrow population, the updated individual position of the sparrow is regarded as the global optimal, and the fitness value of the updated sparrow population is taken as the global optimal fitness value gbest; S6, determine whether the iteration stop condition is met, if not, adopt the probabilistic following strategy, adaptive local search strategy and Cauchy Gauss mutation strategy to update the position of the sparrow population, and update the parameters of the bigamma function BiGA, and go to step S3; if the iteration stop condition is met, end the iteration and go to the next step S7; S7, calculating the optimal parameter value of the bi-gamma function BiGA according to the global optimal fitness value gbest of the sparrow population; S8. After the bi-gamma function BiGA obtains the optimal parameter value, the image is processed, and finally the obtained image is denormalized to output the enhanced image.

2. The image enhancement optimization method based on multi-strategy improved sparrow search as claimed in claim 1, characterized in that: Step S2 is specifically, S21. Randomly initialize an initial sparrow population X with N individuals, and generate a cubic mapping sequence using the following formula: Among them, y c and c+1 is a cubic sequence; S22, the cubic sequence y is transformed into c Mapped to sparrow individual x c Then we get the chaotic mapping population X c : Among them, ub and lb are the upper and lower bounds of the search space respectively; S23, calculate the population X after chaotic mapping c The fitness value of S24, find the population X after chaotic mapping through the following formula c The reverse solution population Among them, i d and j d They represent the minimum and maximum values ​​in the dth dimension of the current population, respectively, and m is the zoom factor of the lens; S25. Calculate the reverse solution population The fitness value of S26, population X after merging chaotic maps c and reverse solution population Sort the fitness values ​​of the merged sparrow individuals, and select the first N sparrow individuals with better fitness values ​​as the new initial sparrow population.

3. The image enhancement optimization method based on multi-strategy improved sparrow search as claimed in claim 1, characterized in that: In step S4, the value of the evaluation function is calculated, specifically, S41. Calculate the entropy value H of the image: Among them, p(i) represents the probability of a pixel with gray value i appearing in the image; S42. Calculate the Sobel operator S for the edge content of the image: Wherein, N-edges(E) represents the number of edge pixels detected by the Sobel operator; T represents the number of all pixels in the image; S43, calculating the grayscale standard deviation V; S44, calculate the value F of the evaluation function: F=μ1H+μ2S+μ3V Among them, μ1, μ2 and μ3 are all constants and μ1+μ2+μ3=1.

4. The image enhancement optimization method based on multi-strategy improved sparrow search according to any one of claims 1 to 3, characterized in that: In step S6, the position of the sparrow population is updated using a probabilistic following strategy, an adaptive local search strategy, and a Cauchy-Gauss mutation strategy, specifically, S61, randomly select a set proportion of sparrow individuals in the sparrow population as scouts, and sort the remaining individuals according to their fitness values. The first set proportion are used as discoverers, and the rest are all followers; S62. Follower position update formula using probabilistic following strategy: in, Indicates the position of the cth sparrow individual in the search space at the n+1th iteration and the nth iteration; is the worst position of individual sparrows in the sparrow population at the nth iteration; is a position randomly selected in the sparrow population at the nth iteration, and r∈[1,N], r≠c; G is a random number that follows a normal distribution; R(0,1) is a uniformly distributed random number, f r is the fitness value of a randomly selected individual, f c is the fitness value of the individual in this round; S63. After the probabilistic following strategy, the fitness value of each individual sparrow is calculated again, and the individual sparrows are re-sorted according to the fitness value. The individuals with high fitness values ​​are selected as discoverers and the individuals with low fitness values ​​are selected as followers according to the set ratio. An adaptive local search strategy is introduced, and the followers take the discoverer's position as the optimal solution. In the optimal solution Perform domain search to obtain the domain value of the optimal solution Among them, θ is the adaptive coefficient, D is the maximum number of iterations; Finally, the domain value of the optimal solution With the optimal solution Compare the fitness values, select the sparrow individuals with high fitness values ​​as the new optimal solution, and then obtain a new sparrow population; S64, based on the new sparrow population obtained after the adaptive search, the sparrow population is sorted according to the fitness value, and the sparrow individual with the highest fitness value is selected Perform Cauchy-Gauss mutation to obtain the position of the sparrow individual with the highest fitness value after mutation in, is the position of the sparrow with the highest fitness value after mutation; τ is the standard deviation of the Cauchy-Gaussian mutation strategy; Gauchy(0,τ 2 ) is a Cauchy distributed random variable; Gauss(0,τ 2 ) is a Gaussian distribution random variable; μ1 and μ2 are adaptive parameters that are dynamically adjusted according to the number of iterations and μ1+μ2=1; as the number of iterations increases, μ1 gradually decreases and μ2 gradually increases; f(X B ) is the highest fitness value, f(X c ) is the fitness value of the current sparrow individual; Calculate the fitness value again and compare the sparrow individuals with the highest fitness value and the position of the sparrow with the highest fitness value after mutation Finally, the best position is selected to enter the next round of iteration.

5. The image enhancement optimization method based on multi-strategy improved sparrow search according to any one of claims 1 to 3, characterized in that: In step S8, the image is processed, specifically, Among them, BiGA(x) is the a (x) and BiGA b (x) The double gamma function after weighted sum, x is the pixel value of the normalized image, and the weight BiGA a (x) is a convex function used to enhance the dark areas of the image, BiGA b (x) is a concave function used to suppress bright areas, γ is an adjustable variable used to control the degree of image enhancement.

Images