Texture and color feature based WLS framework aided skin cancer classification using MSVM and ELM

Abstract

This paper proposes a multilayer decomposition aided method based on textural and color feature for detection and classification of skin cancer images. Firstly, images are decomposed into a piecewise base layer and detail layer by weighted least squares (WLS) framework based edge-preserving decomposition. From detail or enhanced layer of original image, normalized symmetrical Grey Level Co-occurrence Matrix (GLCM) and Histogram of Oriented Gradients (HOG) are taken as textural feature descriptor and color histogram obtained from base or smoothened layer of image is considered as color feature vector. These feature values extracted from smoothened and enhanced images are fed to Multiclass Support Vector Machine (MSVM) and Extreme Learning Machine (ELM) for classification. An average accuracy of 94.18% and 90.5% with MSVM and ELM, respectively are obtained while classifying four types of skin cancer cells (Squamous cell carcinoma, Basal cell carcinoma, Melanoma, Actinic keratosis) for DermNet NZ database.