EDGE BASED IMAGE SEGMENTATION

Abstract

Image segmentation has emerged as a leading field of research due to its wide span of usage and applicability. The segmentation process involves partitioning of an image into various groups of disconnected regions with attributes such as intensity, colour, tone or texture which are both uniform and homogeneous in nature. Various techniques have been developed over time for improving segmentation of an image and the participation of a pixel within a region or cluster as a crisp or a fuzzy entity has been debated widely by various researches. K-Means, Fuzzy C-Means (FCM), Fuzzy Co-Clustering for Images (FCCI) have been widely regarded as some of the most efficient algorithms for the image segmentation process by various researchers and scholars. All these are iterative process in which results are improved over the course of various iterations based on an over heading optimization function. FCM has been widely acknowledged as a very efficient clustering method but it loses its efficiency over many situations like involvement of noise or distorted images. Many variations of FCM have been proposed to overcome these drawbacks. Edge based weight distribution is one of the most efficient variation of FCM given which ensures overcoming of FCM inefficiency with noisy images. FCCI is an improvement over FCM in terms of its optimization to ensure better and well defined segmentation of an image. In the proposed technique, a combination of FCCI and edge based weight distribution method has been explained which ensures efficient results as it incorporates the advantages of both the techniques. The proposed technique has been applied over 100 natural images taken from Berkeley’s image segmentation database and the results have been compared with the above mentioned algorithms on the basis of mean square error as a quantitative measure.