Abstract: To overcome the defects of image segmentation by classic fuzzy C-means (FCM) clustering that considers nothing about image continuity, this paper introduces a new spatially constrained FCM image segmentation algorithm. The pseudo-likelihood of labeling is adopted in this algorithm to combine spectral similarity and spatial neighboring of image pixels. A new objective function is proposed and minimized. The experiments are conducted on simulated gray images and real color images. Experimental results show that the proposed approach is more effective and has better performance.

Key words: fuzzy C-means(FCM), Markov random fields(MRF), pseudo-likelihood, image segmentation

摘要： 针对传统模糊C-均值(FCM)图像分割算法没有考虑图像空间连续性的缺点，提出一种改进的空间约束FCM分割算法。该算法引入了Markov随机场理论中类别标记的伪似然度近似策略，将像素特征域相似性同空间域相邻性有机地结合起来，给出了新的像素样本聚类目标函数。实验证明，该算法能大大提高分割性能并改善分割的视觉效果。

关键词: 模糊C均值, Markov随机场, 伪似然度, 图像分割

CLC Number: 

• TP391