Improved Technique to Diagnose Skin Cancer using Advanced Image Processing and Machine Learning Techniques

Abstract

Skin cancer is being classified among the mortal and exaggerating forms of cancer since decade. Notwithstanding, the early diagnosis of skin cancer is very important and it is an extravagant procedure. The presence of human skin is tough to examine and to model. This is because of its complex surface. The difficulty of the irregular edge, tone, appearance of thick hair and other alleviating features generate the skin tough to be analyzed. Human skin has some non-identical sorts of textures that diseased skin can characterize between the textures of the healthy one. In consequence, considerable achievements have not been brought into the evolution of diagnosis approach for skin cancer. On the other hand, skin cancer diagnosis in dermoscopic portrayals is an exceptionally troublesome chore. Notwithstanding, the early diagnosis of skin cancer is very important and it is an extravagant procedure. In consequence, considerable achievements have not been brought into the evolution of diagnosis approach for skin cancer. For precise diagnosis and categorization of skin cancer, particular features are recommended that one may categorize benign and malignant illustration. A midst analysts, governing the efficacious mechanism of skin cancer diagnosis is an all-important matter of contention. Ascertaining the more proficient methods of diagnosis to minimize the amount of errors is a crucial subject among analysts. Image processing is used to recognize the affected area by disease, its form and color etc. The reason behind it is to minimize the percentage of miscalculations. Computer vision can bring about influential contribution in skin cancer detection. Computer Aided Diagnosis provides support in the premature diagnosis of skin cancer. Programmed skin abrasion segregation is a stimulating chore because of the low contrast between abrasion and encompassing skin and asymmetrical periphery. In this paper we present an amalgamation of statistical features, GLCM features and GLRLM features. The recommended methodology is enacted on 100 images from PH2 database. In this work 2 segmentation tactics namely Otsu’s thresholding and Region growing have been explored. The performance of two systems is measured in terms of sensitivity, specificity, accuracy, precision, recall and f-measure