Abstract: In order to improve the segmentation effect and speed up the segmentation procedure of noise images, this paper proposes an efficient image segmentation method, i.e. IPSOC, which combines Nonsubsampled Contourlet Transform（NSCT） with Particle Swarm Optimization（PSO） algorithm. In this method, an original image firstly is decomposed with multi-scale NSCT transform. Then low frequency coefficients at the highest level are used to reconstruct an approximate image, and after the two-dimensional histogram of the reconstructed image and its mean-filtered image are produced, its trace of the between-class scatter matrix is taken as the object function for searching the best threshold. Simultaneously, an improved PSO algorithm is selected as the parallel scheme, which makes some progress compared to the standard PSO, such as real-time updating the individual and the global optimal information, preventing the stagnation of particles, and regarding threshold-kept times as one of the termination conditions. Experimental results show that IPSOC obviously improves both segmentation speed and segmentation quality, compared with some methods based on genetic algorithm and artificial fish algorithm.

Key words: image segmentation, noise image, Nonsubsampled Contourlet Transform（NSTC）, improved PSO algorithm

摘要： 为提高含噪图像的分割效果和分割速度，将非下采样Contourlet变换和粒子群优化算法相结合，提出了一种有效的图像分割方法——IPSOC。该方法首先对待分割图像进行多尺度非下采样Contourlet变换，然后利用其最高级低频系数重构图像，计算重构图像与其均值图像的二维直方图中类间离散度矩阵的迹，并以之作为分割图像的目标函数来搜索最佳分割阈值。为加快阈值搜索速度，以改进的粒子群优化算法作为阈值分割的并行搜索策略，通过对基本粒子群优化算法进行个体及全局最优信息的实时更新，防止粒子停滞操作和阈值保持次数限定搜索终止条件等几个方面的改进，快速有效地获得分割图像。实验结果表明，该方法与基于遗传算法和人工鱼群算法的分割方法相比，明显提高了图像分割速度和分割质量。

关键词: 图像分割, 含噪图像, 非下采样Contourlet变换, 改进PSO算法