A Robust Feature Extraction for Melanoma Pattern Classification Using Normalization Technique

Abstract

ABSTRACT newline newline newline newline newlineMelanoma the deadliest form of skin cancer must be diagnosed early for effective treatment. Skin cancer is the most common type of cancer that light complexion and high exposure to sunlight occurs often in people with Malignant Melanoma is the rarest, but the most dangerous form of skin cancers that is the cause of 75% of death in white skins. Dermoscopy is one of the techniques used by dermatologists to diagnose melanoma. Therefore, it is necessary to develop a system that will help the dermatologists to produce reliable results. The goal of the synopsis is to develop a Normalized Melanoma Pattern Recognition System (NMPRS) that seeks to identify the global pattern associated with the lesion. The five different stages of NMPRS consist of lesion acquisition, pre-processing, lesion segmentation, feature extraction and lesion classification. At present this technique is the commonly used technique for providing diagnosis accuracy for cutaneous melanoma. The major global patterns are Reticular pattern, Homogenous pattern, Globular pattern and multi-component pattern. Numerous works have focused on extraction of skin lesion and find their malignancy but this research work focuses on melanoma types. The development of independent feature extraction modules, each of which performs well over state-of-art methods has been focused in this work. The major contributions are: i) A study of global patterns and finds the descriptors that identifies the most. ii) Use the well-known speedy and robust SURF based technique and Principal Component Analysis (PCA) to extract the features to form a feature space vector. iii) To develop a system for feature extraction that analyses color, border and geometrical features for pattern recognition which finds the invariance on color and border irregularity. iv) To develop a cluster based feature extraction technique that follows spatial encoding on melanoma pattern recognition. v) Fusion of well said feature extraction techniques that n