Image segmentation is an important research topic of image processing and computer vision community since long. The variety and complexity of images and applications of segmentation make this problem a challenging and critical one. Algorithms based on fuzzy theory have found to provide improved accuracy of segmentation. Fuzzy schemes for segmentation employ fuzzy connectedness to capture the fuzziness associated with the measure of 'hanging togetherness'. The proposed work is a fully automatic segmentation method based on fuzzy affinity and fuzzy connectedness. This uses histogram thresholding to automatically select the salient seed points of segments before applying the fuzzy connectedness algorithm. Performance of the proposed approach is compared with Fuzzy C Means (FCM) algorithm employing sensitivity, specificity and dice similarity coefficient for the cases of natural image data set and medical image data set. The proposed algorithm exhibits superior performance figures for both the data sets when compared to Fuzzy C means algorithm. It is also observed that, unlike FCM, the algorithm treats small size spots as separate regions or segments which is very important in brain image segmentation to locate lesions from the rest of the tissues. This property of the algorithm provides increased accuracy for brain image segmentation. © 2012 Praise Worthy Prize S.r.l.
Jyothisha J. Nair and Govindan, V. K., “Automatic Segmentation Employing Fuzzy Connectedness”, International Review on Computers and Software, vol. 7, no. 2, pp. 561-567, 2012.