Optimal selection of features using wavelet fractal descriptors and automatic correlation bias reduction for classifying skin lesions

The non-invasive computerized image analysis techniques have a great impact on accurate and uniform evaluation of skin abnormalities. The paper reports a method for the texture and morphological feature extraction from skin lesion images to differentiate common melanoma from benign nevi. In this work, a 2D wavelet packet decomposition (WPD) based fractal texture analysis has been proposed to extract the irregular texture pattern of the skin lesion area. On the whole 6214 features have been extracted from each of the 4094 skin lesion images, by analyzing the textural pattern and morphological structure of the lesion area. For the identification of the most efficient feature set, an improved correlation bias reduction method has been introduced in combination with support vector machine recursive feature elimination (SVM-RFE). An automatic selection of correlation threshold value has been introduced in this proposed work to eliminate the correlation bias problem associated with SVM-RFE algorithm. With these selected features, the support vector machine (SVM) classifier with radial basis function is found to achieve the classification performance of 97.63% sensitivity, 100% specificity and 98.28% identification accuracy. The results show that the scheme presented in this paper surpasses the performance of the other state-of-the art techniques for the differentiation of melanoma from other skin abnormalities.