Automatic Classification of Dermoscopic Images and Diagnosing System

Abstract

Melanoma and non-melanoma skin cancers have a rapidly rising increased incidence, implying that cancer is a global epidemic. Our proposed model employs an encoder-decoder framework with CNN models to recognize and restore hair pixels from photos. We compare our method to six state-of-the-art techniques based on classic image processing techniques by utilizing resemblance metrics for evaluating the baseline hairless picture and one with generated hair. The Wilcoxon signed-rank method is used to test the strategies. The results, both qualitatively and statistically, demonstrate how our model works and how our loss function improves the restoration capabilities of the recommended model.