Image segmentation based on Fuzzy entropy and Gestalt principles

The main purpose of image segmentation is achieve more information from specific regions of the image which help the analysis and interpretation of objects in scene. In this paper, an unsupervised method is provided for gray images segmentation. This method is a combination of two processing steps which are performed hierarchically and in parallel form. In the process of image segmentation with the proposed method, before the beginning of each stage of the hierarchical merging in image segmentation, degree of fuzzy homogeneity of the pixels in each subset of the image would be measured which including a region and all its adjacent regions, and regions that present higher mutual information in these subsets would be selected as a place of growth and then, in the next step, a gestalt principles based grouping method would be used in order to obtain meaningful information from these local optimal. The proposed method was compared to two other unsupervised algorithms. The segmentation results using proposed method on Berkeley dataset indicate good performance for PRI, VoI and Covering indices compared to other algorithms.

Keywords:

Fuzzy entropy , thresholds , Mutual information , Gestalt principles