A Comparative Analysis of Skin Cancer Image Denoising and Segmentation using Deep Learning Techniques

Recently the skin cancer patient count is increasing every year, if it continues in future there will be mass strength of people could affect with this disease. It may threaten human life. Early stage detection and diagnosis of skin cancer through automated identification mechanism plays a crucial role in today’s world. The dermoscopic images are varied by hair artifacts, marks, and ill-defined boundaries. Such factors complicate the detection process. The accurate detection and segmentation of skin lesion from rest of the skin requires an efficient automated technique. Hence, segmenting affected portion from noisy image may not be accurate and it decreases the segmentation rate. By resolving this problem, this study proposed Dynamic Environmental Reflection Removal denoising technique to remover or reduce noisy content from the image. Also, this study compares the performance of this proposed denoising technique with several other approaches. Furthermore, segmentation is carried out on a benchmark dataset namely HAM10000 with deep learning techniques such as ANN, RNN, and LSTM has gained segmentation accuracy of 89%, 90%, and 94%. The result shows that the segmentation accuracy is significantly increased while processing the denoised images.