Improved Skin Lesion Detection and Segmentation by Fusing Texture and Geometric Features

Abstract

Melanoma is the greatest carcinogenic skin cancer. In the last years, the prevalence degree of melanoma has risen by 50 percent. There is a necessity to provide an onscreen system for the diagnosis of skin lesions. The system will reduce the unnecessary biopsy and the cancer can be diagnosed at an early stage. In this paper a framework is proposed for the automated skin lesion detection in an input image. A segmentation algorithm based on texture is used to classify normal skin class or lesion class. Also, fusion of texture and geometric features is presented in this work. SVM classifier is trained to identify lesions as malignant melanoma or benign lesion. The system yielded an efficiency of 84.7%, 89.4% and 83.5% for Haralick features, features given by Soh and Clausi and Histogram based features respectively. The fusion based on texture and geometric features enhanced the performance of the system. The evaluated performance metrics are better than the previous methods. The improved system helps diagnose at an early stage reducing the mortality rate.