A two level k-means segmentation technique for eczema skin lesion segmentation using class specific criteria

In the paper we have proposed a two level k-means segmentation technique for eczema skin lesion segmentation. Two class criteria is used for classifying the normal skin and eczema skin lesions using Mahalanobis distance. In order to further improve the segmentation performance normalized color spaces are used. Our experiments include RGB and CIElab color models; and their color space normalized-I (CSN-I) versions. For pre-processing Frankle-McCann retinex and adaptive light compensation is used. The experimental results show that our proposed algorithm gives better results with normalized color spaces. We have achieved a segmentation accuracy of 86.07% for RGB normalized I color space for the G channel with adaptive light compensation.