彩色图像鲁棒聚类分割快速算法

Fast robust clustering algorithm for colour image segmentation

给出一种具有鲁棒性的彩色图像聚类分割快速算法,以改善基于马氏距离聚类分割算法(MFCM)的实时性和抗噪性。利用彩色图像红、绿、蓝3通道分量构造三维直方图,统计出现频次不为零的灰度级组数目,用于取代像素值进行聚类。将图像像素邻域均值嵌入MFCM算法的目标函数,采用拉格朗日乘子法获取其迭代求解表达式,可得相应图像分割聚类算法。随机选取伯克利标准图像库中3幅彩色图像,添加不同强度的高斯噪声,进行分割测试。实验结果表明,改进算法对噪声图像的分割具有一定鲁棒性,对无噪声彩色图像分割相比MFCM算法具有更高执行效率。

The fuzzy clustering image segmentation algorithm （MFCM） based on Mahalanobis distance has inadequacy in aspect of real time performance and noise immunity. A fast robust colour image segmentation algorithm based on improved clustering method is therefore presented. The image pixel neighbourhood average is embedded into its clustering objective function, and a 3D-histogram is constructed by means of three channel components of colour image in ROB space to fast segment colour image. New clustering iterated formulas of sample membership and clustering centres are obtained by using Lagrangian multiplier method to solve the optimal fuzzy clustering objective function, and the colour image segmentation algorithm is then constructed by combining 3D-histogram with the new clustering iterated formulas. Experimental results of segmentation for colour image corrupted by Gaussian noise show that the proposed segmentation algorithm is robust to segment image with noise, and its fast algorithm based on 3D-histogram has higher iterative efficiency for noiseless colour image segmentation compared with the MFCM algorithm.