一种改进的模糊C-均值(FCM)彩色图像分割算法

An improved FCM clustering algorithm for color image segmentation

针对传统模糊C-均值(FCM)聚类算法计算量大,聚类中心对初始值敏感和聚类数目不能自适应确定的缺点,提出了一种改进的FCM算法。首先对图像进行采样量化,并在满足视觉一致性的L*a*b*颜色空间计算并统计图像的色差信息,然后依据全局色差阈值选取初始聚类中心,对图像进行聚类分析,同时根据准则函数确定最佳聚类数,实现了聚类中心的优化选取和最优聚类数目的确定,有效减少了计算量。研究结果表明,改进后的FCM算法不仅较好地克服了传统FCM算法的缺点,而且聚类效果好,处理速度快,聚类效果与人的视觉感应保持了良好的一致性。

Aiming at overcoming the shortages of fuzzy C-means（FCM） algorithm including its large amount of calculation,sensitivity for initial clustering center and the number of clusters can＇t be identified adaptively,an improved method was proposed.Firstly,the original image was sampled and quantified and transformed to Lab color space where the chromatic aberration（CA） was calculated and censused.Then,the initial clustering centers were selected based on global CA threshold and the optimize numbers of clusters were determined by criterion function.Finally,the optimize clustering centers and numbers were realized with less amount of calculation.The results indicate that the improved FCM algorithm not only overcomes the shortages of traditional FCM algorithm effectively,but also gets a good clustering result and maintains a consistency with people＇s vision costing less time.