Extraction of features from cross correlation in space and frequency domains for classification of skin lesions

Abstract

This paper proposes a systematic approach for the feature extraction and subsequent classification of benign and malignant skin diseases of both the melanocytic and epidermal lesion categories from dermoscopic images. For this purpose, melanoma and nevus are considered as representatives of the melanocytic skin lesion category, whereas basal cell carcinoma (BCC) and seborrheic keratoses (SKs) are included under the epidermal lesion category. The present work explicates the extraction of spatial and the spectral features from conspicuous regions of skin lesions on the basis of similar visual impacts with the appropriate kernel patches, using the Cross-correlation technique. Depending on the dermoscopic features, the kernel patches have been chosen from a set of dermoscopic images comprising all the skin disease categories selected for this work. A multi-label ensemble multiclass skin lesion classification strategy has been introduced for the segregation of malignant and benign melanocytic and epidermal skin lesions, along with their subclass classification. It has been possible to identify both malignant and benign lesions of melanoma, nevus, BCC and SK disease classes, with sensitivities of 98.76%, 99.01%, 98.87% and 99.41% respectively.