Analysis of Dermoscopic Images for Classification of Skin Lesions Using Advanced Image Processing Techniques

Skin Cancer is the most common disease worldwide, which has high prevalence even in the developed countries. Most skin lesions begin as abnormal growths that can convert into cancer in certain types of skin tissues. Traditional methods of clinical diagnosis often require trained specialists, MRI devices, X-ray machines and laboratory testing which aren't always accessible, especially in rural or under-resourced areas. Recent advancements in science and technology have enabled the use of computer-based models to diagnose skin diseases by analyzing symptoms. Although, the issue that requires attention is the lightening condition of skin diseases images as a result, images exhibit weak contrast. This processing part of the dermatology images are still less explored. In this research, we have applied various advanced image processing techniques like Histogram Equalization, Adaptive Histogram Equalization, Contrast-Limited-AdaptiveHistogram Equalization, and Gamma correction, afterwards their effects are studied using Mean-Squared-Error, and Peak-Signal-to-Noise-Ratio. Furthermore, the study also analyzed the effect of the processing techniques on classification performance using various deep convolutional neural networks.