Abstract:

Spectral clustering minimizing normalized cut criterion has a high computational complexity and an inaccurate result in color image segmentation. In order to overcome these disadvantages, the paper firstly used Fuzzy C-Means (FCM) to deal with three channels of color image, obtained pre-segmentation image from these channels cluster results to construct undirected weighted graph; and then minimized normalized cut criterion using discrete particle swarm optimization algorithm instead of spectral clustering; finally, pre-segmentation result was obtained by the optimal particle. The experimental results show that the method is less time-consuming, and obtains a precise segmentation result in color image segmentation.

Key words: normalized cut, particle swarm optimization algorithm, Fuzzy C-Means (FCM), color image segmentation

摘要：

为克服谱聚类算法求解归一化彩色图像划分时计算复杂度高、寻优能力差的不足，先对彩色图像各通道进行模糊C均值聚类，综合各通道聚类结果获得待分割图像，构造无向带权图；再使用二进制离散化粒子群算法替代谱聚类算法求解归一化划分准则的最小值，最后通过最优粒子获得分割结果。实验表明该方法耗时少，能完整准确地提取彩色图像中的目标。

关键词: 归一化划分, 粒子群优化算法, 模糊C均值聚类, 彩色图像分割