An Intelligent Saliency Segmentation Technique and Classification of Low Contrast Skin Lesion Dermoscopic Images Based on Histogram Decision

Abstract

Skin cancers primarily malignant melanoma is mortal and tough to recognize in the final stages. To minimize the increasing death rate it is a most essential goal to recognize the skin cancer at its first stage. Skin lesion classification is becoming challenging more and more due to low contrast images. In this research, we propose an intelligent method by implementing the histogram decision to separate the low contrast images into a large amount of dataset. This decision is helpful in the pre-processing stage for the enhancements just in low contrast image either applied into all dataset by avoiding the time complexity. The saliency-based method is applied for lesion segmentation and achieved 95.8 % accuracy. Feature selection is performed by the entropy method after the extraction of deep color and PHOG features. In this research, the SVM classifier is applied on three benchmark datasets ISIB 2016, ISIB 2017 and PH2. Through our proposed fusion feature vector, the best classification results in achieved are 99.5% accuracy on the dataset ISIB2017.