Tree structured wavelet transform signature for classification of melanoma

The purpose of this work is to evaluate the use of a wavelet transform based tree structure in classifying skin lesion images into melanoma and dysplastic nevus classes based on the spatial/frequency information. The classification is done using the wavelet transform tree structure analysis. Development of the tree structure in the proposed method uses energy ratio thresholds obtained from a statistical analysis of the coefficients in the wavelet domain. The method is used to obtain a tree structure signature of melanoma and dysplastic nevus, which is then used to classify the data set into the two classes. Images are classified by using a semantic comparison of the wavelet transform tree structure signatures. Results show that the proposed method is effective and simple for classification based on spatial/frequency information, which also includes the textural information.