基于分布反馈螺旋扰动连续蚁群算法的多阈值图像分割研究

Research on Multilevel Threshold Image Segmentation Based on Distribution-Feedback Spiral Perturbation Continuous Ant Colony Optimization

摘要: 多阈值图像分割是图像处理领域中的重要研究内容，其核心在于根据图像灰度分布自动确定一组最优阈值，从而实现图像区域的有效划分。随着阈值数量的增加，候选阈值组合的搜索空间迅速扩大，传统穷举方法计算代价较高，而普通群智能优化算法又容易出现收敛速度慢、寻优精度不足以及陷入局部最优等问题。针对上述问题，本文提出一种基于分布反馈螺旋扰动的改进连续蚁群优化算法(DFSACO)，并将其应用于多阈值图像分割任务。该方法首先利用连续蚁群优化算法构建阈值搜索模型，通过解档案保存优质阈值组合，并采用高斯核采样方式生成候选解；随后设计分布反馈螺旋扰动算子，根据当前解档案的分布中心、最优解位置以及档案离散程度，对候选解进行自适应扰动更新，从而在保持算法局部开发能力的同时增强全局搜索能力。最后，采用Kapur熵作为图像分割目标函数，对多阈值组合进行优化求解。本文方法能够有效提升连续蚁群优化算法在复杂阈值空间中的搜索能力，为群智能优化算法在图像分割中的应用提供了一种新的思路。

Abstract: Multilevel threshold image segmentation is an important topic in image processing, and its key objective is to automatically determine a set of optimal thresholds according to the gray-level distribution of an image. As the number of thresholds increases, the search space of candidate threshold combinations expands rapidly. Traditional exhaustive methods usually suffer from high computational cost, while common swarm intelligence algorithms may face slow convergence, insufficient optimization accuracy, and premature convergence. To address these issues, this paper proposes an improved continuous ant colony optimization algorithm based on a distribution-feedback spiral perturbation strategy and applies it to multilevel threshold image segmentation (DFSACO). The proposed method first constructs a threshold search model using continuous ant colony optimization, where an archive is employed to store high-quality threshold combinations and Gaussian kernel sampling is used to generate candidate solutions. Then, a distribution-feedback spiral perturbation operator is designed to adaptively update candidate solutions according to the distribution center, the current best solution, and the dispersion degree of the archive. In this way, the algorithm can enhance global exploration while maintaining local exploitation. Finally, Kapur’s entropy is adopted as the objective function to optimize the multilevel threshold combination. The proposed method improves the search capability of continuous ant colony optimization in complex threshold spaces and provides a feasible approach for applying swarm intelligence algorithms to image segmentation.