Image segmentation optimization method, system, device and medium

Abstract

The invention discloses an image segmentation optimization method, system, equipment and medium, and relates to the technical field of computer optimization algorithms, and the method comprises the steps: constructing a two-dimensional histogram for an original gray level image, employing a two-dimensional Renyi entropy as a target function for evaluating the segmentation quality of the two-dimensional histogram, finding a group of optimal threshold combinations, and carrying out the segmentation of the two-dimensional histogram through the optimal threshold combinations; the problem of searching the optimal threshold value combination is converted into an optimization problem to be solved, the found optimal threshold value combination is used for segmenting the original image, and a final segmentation result image is generated; when an optimization problem is solved, operating individuals in the population by adopting a differential variation strategy so as to generate a first group of candidate solutions, and executing a Rime optimization mechanism and a randomized step size covariance matrix adaptive mechanism in parallel on each individual in the population so as to generate a second group of candidate solutions; and updating the population according to a greedy selection mechanism. The algorithm premature convergence can be effectively prevented, and the method can be applied to complex optimization tasks such as image segmentation and engineering design.

Description

Technical Field

[0001] This invention relates to the field of computer optimization algorithm technology, and in particular to an optimization method, system, device and medium for image segmentation. Background Technology

[0002] Intelligent optimization algorithms are computational methods that solve complex optimization problems by simulating natural phenomena. These algorithms are frequently used to solve practical engineering problems, with one typical application being multi-level threshold image segmentation. Multi-level threshold segmentation divides image pixels into multiple categories by determining a set of optimal grayscale thresholds. The core challenge of this technique lies in the fact that finding the optimal combination of thresholds is an NP-hard problem; the search space grows exponentially with the number of thresholds.

[0003] To address these issues, various intelligent optimization algorithms have been developed. For example, the Rime Optimization Algorithm (RIME) simulates the formation of rime ice; its mechanism is simple and converges quickly, but its balance between exploration and utilization is not flexible enough, leading to premature convergence when dealing with complex high-dimensional problems. Differential Evolution (DE) is known for its simple structure and strong global search capabilities, but its local search accuracy needs improvement in high-dimensional problems. Another powerful technique is the Covariance Matrix Adaptation Evolution Strategy (CMA-ES), which adaptively adjusts the covariance matrix, resulting in extremely high local search efficiency. However, the step-size adaptation mechanism of CMA-ES is tightly coupled with the covariance matrix evolution; once the algorithm adapts to a local optimum, the step size may decay prematurely, causing the search to stagnate and making it difficult to escape local optima.

[0004] In summary, existing intelligent optimization algorithms often encounter search stagnation when processing high-dimensional data, easily getting stuck in local optima, and are unable to accurately obtain the optimal threshold combination for image pixel classification, resulting in poor image segmentation quality. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of the prior art by providing an optimized method, system, device, and medium for image segmentation, thereby solving the problems in the prior art.

[0006] This invention specifically provides the following technical solution: an optimization method for image segmentation, comprising: Acquire an original grayscale image to be segmented and generate a two-dimensional histogram of the original grayscale image; Two-dimensional Rényi entropy is used as the objective function to evaluate the segmentation quality of two-dimensional histograms, and multiple threshold points for segmenting the two-dimensional histograms are obtained. The coordinate values ​​of the threshold points are used as candidate solutions, and multiple candidate solutions are combined into a population; where the threshold point is a two-dimensional coordinate point in the two-dimensional histogram. Based on the objective function, an adaptive dimensionality selection differential optimization strategy is used to search the population. The number of dimensions to be updated and the number of offspring to be generated are dynamically determined according to the current search progress. A differential mutation strategy is then used to operate on the individuals in the population to generate the first set of candidate solutions. For each individual in the population, a second set of candidate solutions is generated through a collaborative update using a parallel Rime optimization mechanism and a randomized step-size covariance matrix adaptive evolution strategy RSCMA-ES mechanism. According to the preset greedy selection mechanism, the population and the global optimal solution are updated by combining the first set of candidate solutions and the second set of candidate solutions; The process of obtaining the global optimal solution is iterated, and all the global optimal solutions obtained in the iteration process are used as the optimal threshold point combination. The two-dimensional histogram is segmented by the optimal threshold point combination to obtain the segmentation result image.

[0007] Preferably, the differential mutation strategy uses the current-to-best / 1 mutation operator, combining the information of the current individual in the population, the globally optimal individual, and two randomly selected individuals to generate a new individual as the first set of candidate solutions.

[0008] Preferably, in the RSCMA-ES mechanism, the update of a step-size control evolution path is decoupled from the directional dependence of a covariance matrix.

[0009] Preferably, the decoupling of the update of a step-size control evolution path from the directional dependency of a covariance matrix is ​​specifically as follows: A policy is randomly selected from a pre-defined policy pool, a random direction vector is generated, and sampling is performed based on the current policy parameters and the random direction vector to generate... A new candidate solution; Utilize the optimal Each generation updates the mean vector, replaces the eigenvector matrix of the covariance matrix with a random direction vector, updates the step size to control the evolution path, and decouples the step size adjustment from the principal axis direction of the covariance matrix.

[0010] Preferably, the Rime optimization mechanism simulates physical phenomena to perturb the position of the individual for each dimension of the current individual, based on a dynamically adjusted attachment threshold.

[0011] Preferably, the generation of the two-dimensional histogram of the original grayscale image specifically involves: The original grayscale image to be segmented is subjected to nonlocal mean filtering to obtain the filtered image; A two-dimensional histogram is constructed by combining the pixel gray values ​​of the original grayscale image to be segmented with the corresponding pixel gray values ​​of the filtered image.

[0012] Preferably, the segmentation of the two-dimensional histogram using the optimal threshold point combination specifically involves: The two-dimensional histogram is segmented into multiple category regions by using the optimal combination of threshold points, and these multiple category regions are used as the basis for classifying the regions in the segmented image.

[0013] This invention provides an image segmentation optimization system, comprising: The data acquisition module is used to acquire an original grayscale image to be segmented and generate a two-dimensional histogram of the original grayscale image. The population partitioning module uses two-dimensional Rényi entropy as the objective function to evaluate the segmentation quality of the two-dimensional histogram, obtains multiple threshold points for segmenting the two-dimensional histogram, uses the coordinate values ​​of the threshold points as candidate solutions, and combines multiple candidate solutions into a population; where the threshold point is a two-dimensional coordinate point in the two-dimensional histogram. The first optimization module is used to search the population based on the objective function using an adaptive dimension selection differential optimization strategy, dynamically determine the number of dimensions to be updated and the number of offspring to be generated according to the current search progress, and use a differential mutation strategy to operate on the individuals in the population to generate a first set of candidate solutions. The second optimization module is used to generate a second set of candidate solutions for each individual in the population by co-updating the parallel Rime optimization mechanism and the randomized step-size covariance matrix adaptive evolution strategy RSCMA-ES mechanism. The update module is used to update the population and the global optimal solution according to a preset greedy selection mechanism, combining the first set of candidate solutions and the second set of candidate solutions; The segmentation module iterates through the process of obtaining the global optimal solution and uses all the global optimal solutions obtained during the iteration as the optimal threshold point combination. The optimal threshold point combination is then used to segment the two-dimensional histogram to obtain the segmentation result image.

[0014] The present invention provides a computer device, including a memory and a processor, wherein the memory stores a program, and when the program is executed by the processor, the processor performs the steps of the above-described image segmentation optimization method.

[0015] The present invention provides a storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the above-described image segmentation optimization method.

[0016] Compared with the prior art, the present invention has the following significant advantages: This invention generates a two-dimensional histogram from the acquired original grayscale image and uses two-dimensional Rényi entropy as the objective function. Based on the objective function, it searches a population with threshold point coordinates as candidate solutions and uses a differential mutation strategy to operate on individuals, generating a first set of solutions. During the search, the search behavior is adaptively adjusted according to the optimization progress, achieving an intelligent balance between exploration and utilization throughout the optimization process, improving solution efficiency and accuracy. Furthermore, through an innovative parallel execution Rime optimization mechanism and RSCMA-ES mechanism, each individual in the population is updated to obtain a second set of solutions, achieving the effect of injecting continuous random perturbation. This effectively prevents the step size from prematurely decaying due to over-adaptation to local optima, allowing the algorithm to maintain its exploration vitality in the later stages of the search, thus greatly enhancing its ability to escape local optima. The optimal solution is updated using the two sets of optimized solutions, and the two-dimensional histogram is segmented using the final optimal threshold point combination to obtain the segmentation result image. This provides efficient optimization motivation while reducing unnecessary overhead in high-dimensional space, resulting in segmentation results with good performance on multiple objective indicators. Attached Figure Description

[0017] Figure 1 This is an overall flowchart of a multi-level threshold image segmentation method based on the optimization method of the present invention, provided by an embodiment of the present invention; Figure 2 This is a schematic diagram of the framework of an optimization algorithm provided in an embodiment of the present invention; Figure 3 This is a conceptual comparison diagram of the search behavior between the RSCMA-ES mechanism and the standard CMA-ES mechanism in this embodiment of the invention; wherein... Figure 3 (a) is a diagram of the CMA-ES mechanism. Figure 3 (b) is a diagram of the RSCMA-ES mechanism; Figure 4 This is a schematic diagram illustrating the working principle of the ADSDO mechanism in an embodiment of the present invention; Figure 5 This is a flowchart of the optimization algorithm initialization phase in an embodiment of the present invention; Figure 6 This is a flowchart illustrating the specific execution of the ADSDO mechanism in this embodiment of the invention. Figure 7 This is an execution flowchart of the Rime optimization mechanism in an embodiment of the present invention; Figure 8 This is an internal workflow diagram of the RSCMA-ES mechanism in this embodiment of the invention. Detailed Implementation

[0018] 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, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0019] This invention aims to address the technical problems of existing intelligent optimization algorithms when dealing with high-dimensional complex problems, such as easy getting trapped in local optima, imbalance between exploration and utilization, and insufficient global search capability. To achieve the above objectives, this invention provides an optimization method applied to computer devices. The method first initializes a population containing multiple candidate solutions and then iteratively optimizes the population. In each iteration, the following steps are performed:

[0020] First, an Adaptive Dimension-Selective Differential Optimization (ADSDO) strategy is executed. This strategy dynamically determines the number of dimensions that each individual needs to update and the number of offspring that need to be generated in this iteration based on the current search progress, and uses an improved "current-to-best / 1" differential mutation strategy to operate on the individuals in the population to generate the first set of candidate solutions.

[0021] Secondly, for each individual in the population, the Rime optimization mechanism and a Randomized Step-size Covariance Matrix Adaptation Evolution Strategy (RSCMA-ES) mechanism are executed in parallel to generate a second set of candidate solutions. The RSCMA-ES mechanism introduces a randomly generated direction vector into a computation formula for updating the step-size-controlled evolution path, replacing the traditional eigenvector matrix that depends on the covariance matrix. This decouples the update of the step-size-controlled evolution path from the specific direction of the covariance matrix. This design effectively prevents premature step-size decay due to over-adaptation to the local environment in the later stages of optimization.

[0022] Finally, based on the preset active greedy selection mechanism, and combining the first and second sets of candidate solutions generated above, the population and the global optimal solution are updated, and better solutions are retained for the next generation.

[0023] Example 1: This embodiment describes the specific method of applying the optimization method to a multi-level threshold image segmentation task, and its overall process is as follows: Figure 1As shown, specifically, an image segmentation optimization method in this embodiment includes the following steps: Step S1: Acquire an original grayscale image to be segmented and generate a two-dimensional histogram of the original grayscale image.

[0024] Input image preprocessing: Given an original grayscale image to be segmented. Applying nonlocal means (NL-means) filtering to the image yields a filtered image containing rich nonlocal structural information. .

[0025] Constructing a 2D histogram: Based on the original grayscale image to be segmented Image after nonlocal mean filtering The corresponding pixel grayscale values ​​together construct an NL-means two-dimensional histogram. This histogram... The position value represents the grayscale value in the original image. And the corresponding pixel in the filtered image has a grayscale value of The frequency of occurrence of pixel pairs is normalized to obtain a two-dimensional probability distribution.

[0026] Step S2: Use the two-dimensional Rényi entropy as the objective function to evaluate the segmentation quality of the two-dimensional histogram, and obtain multiple threshold points for segmenting the two-dimensional histogram. Use the coordinate values ​​of the threshold points as candidate solutions, and combine multiple candidate solutions into a population; where the threshold point is a two-dimensional coordinate point in the two-dimensional histogram.

[0027] Define the optimization objective function: use the two-dimensional Rényi entropy as the objective function to evaluate the image segmentation quality. For Threshold points (each threshold) (These are two-dimensional coordinates), and these thresholds segment the two-dimensional histogram into... Each category region. Total two-dimensional Reynytic entropy. Defined as the sum of the Reˊnyi entropies of all category regions. The goal of image segmentation is to find an optimal set of threshold combinations that maximizes the total two-dimensional Reˊnyi entropy. maximize.

[0028] Step S3: Based on the objective function, the population is searched using an adaptive dimension selection differential optimization strategy. The number of dimensions to be updated and the number of offspring to be generated are dynamically determined according to the current search progress. A differential mutation strategy is then used to operate on the individuals in the population to generate the first set of candidate solutions.

[0029] The differential mutation strategy uses the current-to-best / 1 mutation operator to generate a new individual by combining information from the current individual in the population, the global best individual, and two randomly selected individuals, which serves as the first set of candidate solutions.

[0030] The number of dimensions to be updated and the number of offspring to be generated are both set to decrease monotonically as the current search progress increases.

[0031] Step S4: For each individual in the population, a second set of candidate solutions is generated through a collaborative update using the parallel Rime optimization mechanism and the randomized step-size covariance matrix adaptive evolution strategy RSCMA-ES mechanism.

[0032] The Rime optimization mechanism perturbs the position of the individual by simulating physical phenomena for each dimension of the current individual, based on a dynamically adjusted attachment threshold.

[0033] In the RSCMA-ES mechanism, the update of the evolution path controlled by a step size is decoupled from the directional dependency of a covariance matrix to prevent premature decay of the step size during the optimization process. That is:

[0034] A policy is randomly selected from a pre-defined policy pool, a random direction vector is generated, and sampling is performed based on the current policy parameters and the random direction vector to generate... A new candidate solution. The randomly generated direction vector is dynamically selected from a policy pool containing at least two different generation strategies.

[0035] Utilize the optimal The mean vector of each generation is updated by using a random direction vector instead of the eigenvector matrix of the covariance matrix in a calculation formula for updating the step size control evolution path, thereby decoupling the step size adjustment from the principal axis direction of the covariance matrix.

[0036] Step S5: Based on the preset greedy selection mechanism, update the population and the global optimal solution by combining the first group of candidate solutions and the second group of candidate solutions.

[0037] Step S6: Iterate through the process of obtaining the global optimal solution, and use all the global optimal solutions obtained in the iteration process as the optimal threshold point combination. Use the optimal threshold point combination to segment the two-dimensional histogram to obtain the segmentation result image.

[0038] The two-dimensional histogram is segmented into multiple category regions using the optimal combination of threshold points. These multiple category regions are then used as the criteria for classifying the regions in the segmentation result image, generating the final segmentation result image. The final segmentation result image is then subjected to pseudo-color processing to obtain a Jet color mapping image.

[0039] As shown in steps S3, S4, S5, and S6, the optimization problem is solved: the problem of finding the optimal threshold combination is transformed into an optimization problem. The variable to be optimized is... The coordinates of a two-dimensional threshold point, the dimension of the problem is... The objective function is the total two-dimensional Rényi entropy as defined above. The optimization problem is solved using the DRSCRIME optimization method detailed in Example 2.

[0040] Example 2: This embodiment provides an optimization method based on the DRSCRIME algorithm. (Refer to...) Figure 2 This method framework integrates a basic Rime optimization module, an Adaptive Dimension Selection Difference Optimization (ADSDO) module 210, and a Randomized Step Size Covariance Matrix Adaptive (RSCMA-ES) module 220. Its detailed execution process is as follows:

[0041] Step 1: Initialization Phase (refer to...) Figure 5 ): Set the global parameters of the algorithm, including population size. Problem Dimension Search boundaries and the maximum number of function evaluations As a termination condition, initialize the internal state variables of the RSCMA-ES module, including the mean vector. Global step size Unit covariance matrix and evolutionary path and Based on this, randomly generate containing The initial population of candidate solutions Then, the fitness of each candidate solution is evaluated, and the individual with the best fitness is recorded as the initial global optimal solution. and its fitness value .

[0042] Step 2: Iteratively optimize the main loop: Number of function evaluations Not achieved Previously, steps three and four were executed repeatedly.

[0043] Step 3: Global Exploration Enhancement Phase (ADSDO Module 210, see reference) Figure 6 ): This phase aims to enhance population diversity and global exploration capabilities through efficient differential mutation strategies.

[0044] Dynamic parameter determination: First, based on the current search progress Dynamically calculate the number of dimensions to be updated in the current iteration. and the number of offspring to be generated for each parent generation Both of these parameters change with... The algorithm decreases as the number of iterations increases, allowing for a smooth transition from extensive exploration in the early stages to refined utilization in the later stages. The working principle of ADSDO is as follows: Figure 4 As shown.

[0045] Differential mutation and offspring generation: Iterate through every parent individual in the current population. For each parent, loop... This process generates the corresponding number of new offspring. When generating each offspring, a random selection is made. Each dimension is used to generate new values ​​by applying the current-to-best / 1 difference mutation operator to these dimensions.

[0046] Evaluation and Greedy Selection: After boundary conditions are applied to newly generated offspring, their fitness is immediately evaluated and compared with their direct parents. Only when the offspring's fitness is better is the offspring used to replace the parent. During this process, if the fitness of any new offspring is better than the current global optimum, the current optimum is updated synchronously. and .

[0047] Step 4: Individual Collaborative Update Phase (Rime Module and RSCMA-ES Module 220): After the ADSDO phase, for each individual in the population ( from arrive Perform a fine-grained search in parallel on two tracks.

[0048] Rime mechanism update (see reference) Figure 7 For each dimension of the current individual, based on a dynamically adjusted attachment threshold, the "Soft-rime search strategy" or "Hard-rime puncture mechanism" is probabilistically activated to simulate physical phenomena that perturb the individual's position, balancing wide-area exploration and convergence toward the optimal region.

[0049] RSCMA-ES mechanism update (see reference) Figure 8 This mechanism provides each individual with powerful local search and escape capabilities. Its core lies in the fact that it performs a complete round of adaptive parameter updates for each population member within a single macro-iteration.

[0050] Parameters and offspring generation: First, a policy is randomly selected from a pre-defined policy pool to generate a direction vector. Then, based on the current strategy parameters... And the random direction vector B, sampled to generate A new candidate solution is identified and evaluated, i.e., boundary checks are performed, and fitness is calculated.

[0051] Parameter adaptation: Sort offspring and utilize the optimal... Update the mean vector for each child generation. Subsequently, the step size control evolution path is adaptively updated. Covariance matrix evolution path Global step size Covariance Matrix If it is not the initial stage, update the covariance matrix. The key step is... The update uses a random vector This decouples the step size adjustment from the principal axis direction of the covariance matrix, such as... Figure 3 (a) and Figure 3 As shown in (b), even if the sampling distribution (ellipse) has adapted to the local environment, the step size adjustment direction (arrow) is still random, which can generate exploratory "jumps" to effectively escape the local optimum (blue star) and find the global optimum (red star).

[0052] Matrix maintenance and global update: To ensure numerical stability, the eigensystem is periodically updated, and the covariance matrix is ​​maintained. Perform eigenvalue decomposition into B, and constrain C. Simultaneously, adjust the global step size, check for minimum and maximum values, and update immediately if any newly generated offspring is better than the current global optimum. .

[0053] Step 5: Final Selection and Iterative Check: After all individuals have undergone collaborative updates, an "active (forward) greedy selection mechanism" is applied to filter the final candidate solutions, ensuring that only individuals with improved performance can enter the next generation. Then, the process is checked. Has it been achieved? If the desired result is not achieved, return to step three to begin a new iteration; otherwise, the optimization process terminates.

[0054] Step Six: Output: After the loop terminates, output the final globally optimal solution. and its fitness value.

[0055] Effect verification: To verify the beneficial effects of the present invention, a segmentation experiment was conducted on a BSD dataset containing 51 complex natural scene images.

[0056] To objectively benchmark the performance of the method of this invention, nine advanced or classic intelligent optimization algorithms, including RIME, DE, SSA, WOA, CLPSO, IWOA, IGWO, DECLS, and CLSGMFO, were selected for comparison. To ensure the fairness and reproducibility of the results, all algorithms were executed under uniform experimental parameters. Specifically, for each test image in the dataset, the number of segmentation thresholds (TH) was set to 5, 10, 15, 20, 25, and 30, respectively. The population size of all comparison algorithms was uniformly set to 30, and the maximum number of iterations was 100. To reduce the impact of randomness, each algorithm was run independently 30 times at each threshold level, and the statistical results were recorded for subsequent performance evaluation.

[0057] Comparative results based on three core metrics—Feature Similarity Index (FSIM), Peak Signal-to-Noise Ratio (PSNR), and Structural Similarity Index (SSIM)—show that the performance advantage of the method in this invention becomes increasingly significant with the increase of the number of thresholds (i.e., the increase of problem dimensionality), demonstrating excellent scalability. In terms of FSIM metrics: In the low-dimensional scenario with TH=5, the method of this invention is already competitive (ranked second); when the number of thresholds increases to 10 or more, the method of this invention ranks first on average in all higher-dimensional scenarios.

[0058] In terms of PSNR metric: the method of this invention shows an absolute advantage, consistently ranking first in average across all six threshold levels from low to high dimensions.

[0059] Regarding the SSIM metric: the advantage of the method of this invention is also established as the dimensionality increases. Its average ranking improves from fourth place at TH=5 to first place at all higher threshold levels of TH=10 and above.

[0060] In summary, these comprehensive data demonstrate that, compared with the aforementioned comparative methods, the present invention exhibits superior accuracy, robustness, and scalability in solving complex image segmentation problems of different dimensions, as shown in Tables 1, 2, and 3.

[0061] Table 1. Comparison results of FSIM for all algorithms on the BSD dataset Table 2. PSNR comparison results for all algorithms on the BSD dataset Table 3. SSIM comparison results for all algorithms on the BSD dataset Compared with the prior art, the present invention has at least the following beneficial effects: The algorithm significantly enhances its ability to avoid premature convergence: through the innovative RSCMA-ES mechanism, the step size adaptation process is decoupled from the directionality of the covariance matrix, injecting continuous random perturbations. This effectively prevents the step size from decaying prematurely due to over-adaptation to local optima, allowing the algorithm to maintain its vitality for large-scale exploration in the later stages of the search, thereby greatly improving its ability to escape local optima.

[0062] A dynamic balance between exploration and exploitation is achieved: the introduced ADSDO mechanism can adaptively adjust the search behavior according to the optimization process. In the early stage, global exploration is enhanced by updating more dimensions and generating more offspring; in the later stage, the amount of updates is reduced to focus on the fine-grained exploitation of advantageous regions, thus achieving an intelligent balance between exploration and exploitation throughout the optimization process, improving solution efficiency and accuracy.

[0063] The method exhibits superior performance and scalability in high-dimensional complex problems: ADSDO's dimension selection strategy effectively reduces unnecessary computational overhead in high-dimensional spaces, while RSCMA-ES's anti-stagnation mechanism ensures continuous optimization motivation. The synergistic effect of these two mechanisms makes the performance advantage of this invention increasingly apparent when dealing with high-dimensional problems. Experimental results show that in handling high-dimensional (e.g., 100-dimensional) benchmark functions and image segmentation tasks with up to 30 thresholds, the accuracy, robustness, and stability of this invention comprehensively surpass many classic and advanced comparative algorithms.

[0064] Significant results in practical applications: When this invention is applied to multi-level threshold segmentation tasks of natural, medical (breast cancer, kidney pathology) and remote sensing images, by optimizing the two-dimensional Rényi entropy, the segmentation results obtained are superior to the comparison methods in multiple objective indicators such as feature similarity (FSIM), peak signal-to-noise ratio (PSNR) and structural similarity (SSIM), and the visual effect is better, which proves the effectiveness and superiority of this invention in solving practical complex optimization problems.

[0065] This invention proposes an image segmentation optimization system, comprising: a data acquisition module, a population partitioning module, a first optimization module, a second optimization module, an update module, and a segmentation module.

[0066] The system comprises several modules: a data acquisition module for acquiring an original grayscale image to be segmented and generating a two-dimensional histogram; a population partitioning module using two-dimensional Rényi entropy as the objective function to evaluate the segmentation quality of the two-dimensional histogram, obtaining multiple threshold points for the segmentation histogram, using the coordinates of these threshold points as candidate solutions, and combining multiple candidate solutions into a population; where each threshold point is a two-dimensional coordinate point in the two-dimensional histogram; and a first optimization module using an adaptive dimensionality selection differential optimization strategy to search the population based on the objective function, dynamically determining the number of dimensions to be updated and the number of offspring to be generated according to the current search progress, and employing differential mutation. The strategy operates on individuals in the population to generate a first set of candidate solutions. The second optimization module is used to generate a second set of candidate solutions for each individual in the population through a parallel Rime optimization mechanism and a randomized step-size covariance matrix adaptive evolution strategy RSCMA-ES. The update module is used to update the population and the global optimum by combining the first and second sets of candidate solutions according to a preset greedy selection mechanism. The segmentation module is used to iterate the process of obtaining the global optimum and use all the global optimum solutions obtained in the iteration process as the optimal threshold point combination to segment the two-dimensional histogram and obtain the segmentation result image.

[0067] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a program, and when the program is executed by the processor, the processor performs the steps of an image segmentation optimization method.

[0068] According to the disclosed embodiments, the computer device can communicate with one or more external devices (e.g., keyboard, pointing device, Bluetooth communication, etc.) or with any device that enables the computing device to communicate with one or more other computing devices (e.g., router, demodulator, etc.).

[0069] The present invention also provides a storage medium storing a computer program thereon, characterized in that the computer program, when executed by a processor, implements the steps of an optimized image segmentation method.

[0070] According to the disclosed embodiments, the storage medium can be a non-volatile computer-readable storage medium, such as, but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this invention, the storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0071] The above description, in conjunction with specific preferred embodiments, provides a more detailed explanation of the present invention. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such deductions or substitutions should be considered to fall within the scope of protection of the present invention.

Claims

1. An optimization method for image segmentation, characterized in that, include: Acquire an original grayscale image to be segmented and generate a two-dimensional histogram of the original grayscale image; Two-dimensional Rényi entropy is used as the objective function to evaluate the segmentation quality of two-dimensional histograms, and multiple threshold points for segmenting two-dimensional histograms are obtained. The coordinate values ​​of the threshold points are used as candidate solutions, and multiple candidate solutions are combined into a population. The threshold point is a two-dimensional coordinate point in the two-dimensional histogram; Based on the objective function, an adaptive dimensionality selection differential optimization strategy is used to search the population. The number of dimensions to be updated and the number of offspring to be generated are dynamically determined according to the current search progress. A differential mutation strategy is then used to operate on the individuals in the population to generate the first set of candidate solutions. For each individual in the population, a second set of candidate solutions is generated through a collaborative update using a parallel Rime optimization mechanism and a randomized step-size covariance matrix adaptive evolution strategy RSCMA-ES mechanism. According to the preset greedy selection mechanism, the population and the global optimal solution are updated by combining the first set of candidate solutions and the second set of candidate solutions; The process of obtaining the global optimal solution is iterated, and all the global optimal solutions obtained in the iteration process are used as the optimal threshold point combination. The two-dimensional histogram is segmented by the optimal threshold point combination to obtain the segmentation result image.

2. The image segmentation optimization method as described in claim 1, characterized in that, The differential mutation strategy uses the current-to-best / 1 mutation operator to generate a new individual by combining information from the current individual in the population, the globally optimal individual, and two randomly selected individuals, which serves as the first set of candidate solutions.

3. The image segmentation optimization method as described in claim 1, characterized in that, In the RSCMA-ES mechanism, the update of a step-size control evolution path is decoupled from the directional dependence of a covariance matrix.

4. The image segmentation optimization method as described in claim 3, characterized in that, The method of decoupling the update of a step-size-controlled evolution path from the directional dependency of a covariance matrix is ​​as follows: A policy is randomly selected from a pre-defined policy pool, a random direction vector is generated, and sampling is performed based on the current policy parameters and the random direction vector to generate... A new candidate solution; Utilize the optimal Each generation updates the mean vector, replaces the eigenvector matrix of the covariance matrix with a random direction vector, updates the step size to control the evolution path, and decouples the step size adjustment from the principal axis direction of the covariance matrix.

5. The image segmentation optimization method as described in claim 1, wherein the Rime optimization mechanism, for each dimension of the current individual, simulates physical phenomena to perturb the individual's position based on a dynamically adjusted attachment threshold.

6. The image segmentation optimization method as described in claim 1, characterized in that, The specific steps for generating the two-dimensional histogram of the original grayscale image are as follows: The original grayscale image to be segmented is subjected to nonlocal mean filtering to obtain the filtered image; A two-dimensional histogram is constructed by combining the pixel gray values ​​of the original grayscale image to be segmented with the corresponding pixel gray values ​​of the filtered image.

7. The image segmentation optimization method as described in claim 1, characterized in that, The segmentation of the two-dimensional histogram using the optimal threshold point combination is specifically as follows: The two-dimensional histogram is segmented into multiple category regions by using the optimal combination of threshold points, and these multiple category regions are used as the basis for classifying the regions in the segmented image.

8. An optimization system for image segmentation, characterized in that, include: The data acquisition module is used to acquire an original grayscale image to be segmented and generate a two-dimensional histogram of the original grayscale image. The population partitioning module uses two-dimensional Rényi entropy as the objective function to evaluate the segmentation quality of the two-dimensional histogram, obtains multiple threshold points for segmenting the two-dimensional histogram, uses the coordinate values ​​of the threshold points as candidate solutions, and combines multiple candidate solutions into a population. The threshold point is a two-dimensional coordinate point in the two-dimensional histogram; The first optimization module is used to search the population based on the objective function using an adaptive dimension selection differential optimization strategy, dynamically determine the number of dimensions to be updated and the number of offspring to be generated according to the current search progress, and use a differential mutation strategy to operate on the individuals in the population to generate a first set of candidate solutions. The second optimization module is used to generate a second set of candidate solutions for each individual in the population by co-updating the parallel Rime optimization mechanism and the randomized step-size covariance matrix adaptive evolution strategy RSCMA-ES mechanism. The update module is used to update the population and the global optimal solution according to a preset greedy selection mechanism, combining the first set of candidate solutions and the second set of candidate solutions; The segmentation module iterates through the process of obtaining the global optimal solution and uses all the global optimal solutions obtained during the iteration as the optimal threshold point combination. The optimal threshold point combination is then used to segment the two-dimensional histogram to obtain the segmentation result image.

9. A computer device, characterized in that, The device includes a memory and a processor, wherein the memory stores a program that, when executed by the processor, causes the processor to perform the steps of an image segmentation optimization method as described in any one of claims 1 to 7.

10. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the image segmentation optimization method according to any one of claims 1 to 7.

Images