Skin lesion segmentation and recognition using multichannel saliency estimation and M-SVM on selected serially fused features

The number of deaths caused by melanoma has increased remarkably in the last few years which are the carcinogenic type of skin cancer. Lately, computer based methods are introduced which are intelligent enough to support dermatologist in initial judgment of skin lesion. However, there still exists a gap for an optimal solution; therefore, machine learning community is still considering it a great challenge. The primary objective of this article is to efficiently detect and classify skin lesion with the utilization of an improved segmentation and feature selection criteria. Presented contribution is threefold; First, ternary color spaces are exploited to separate foreground from the background—utilizing multilevel approach of contrast stretching. Second, a weighting criterion is designed which is able to select the best solution based on extended texture feature analysis, related labels, boundary connections and central distance. Third, an improved feature extraction and dimensionality reduction criteria is proposed which combines conventional as well as recent feature extraction techniques. The proposed method is tested on PH2, ISBI 2016 and ISIC benchmark data sets and evaluated on the basis of multiple parameters including FPR, sensitivity, specificity, FNR and accuracy. From the statistics, it is quite clear that the proposed method outperforms numerous existing techniques with considerable margin. © Springer-Verlag GmbH Germany, part of Springer Nature 2018.