R.V.V. Krishna¹ and S. SrinivasKumar^2
1Aditya College of Engineering &Technology and ²Jawaharlal Nehru Technological University, Vijayanagaram, India

Color image segmentation algorithms in the literature segment an image on the basis of color, texture, and also as a fusion of both color and texture. In this paper, a color image segmentation algorithm is proposed by extracting both texture and color features and applying them to the One-Against-All Multi Class Support Vector Machine classifier for segmentation. A novel Power Law Descriptor (PLD) is used for extracting the textural features and homogeneity model is used for obtaining the color features. The Multi Class SVM is trained using the samples obtained from Soft Rough Fuzzy-C-Means (SRFCM) clustering. Fuzzy set based membership functions capably handle the problem of overlapping clusters. The lower and upper approximation concepts of rough sets deal well with uncertainty, vagueness, and incompleteness in data. Parameterization tools are not a prerequisite in defining Soft set theory. The goodness aspects of soft sets, rough sets and fuzzy sets are incorporated in the proposed algorithm to achieve improved segmentation performance. The Power Law Descriptor used for texture feature extraction has the advantage of being dealt in the spatial domain thereby reducing computational complexity. The proposed algorithm is comparable and achieved better performance compared with the state of the art algorithms found in the literature.

Segmentation , Classification, Clustering, Fuzzy Sets, Homogeneity, Rough Sets, , Soft Sets, Multi Class SVM, Texture, Power Law Descriptor