Abstract

This paper introduces a novel approach for segmenting skin images based on the use of Wavelet Networks (WN). The proposed WN model is a fixed-structure network that does not require traditional training. After constructing the wavelet framework, translation and scaling parameters are determined through a two-stage screening process, which helps in selecting the most effective wavelets. To refine the network architecture and estimate optimal weights, the orthogonal least squares (OLS) algorithm is employed. The dual screening stages enhance the global efficiency of the wavelet matrix and improve function estimation, particularly for larger scales. In this work, the R, G, and B components of the skin image serve as the input features for the network. The image is then segmented to accurately identify the boundaries of skin lesions. The proposed segmentation algorithm was tested on 30 skin images and evaluated across 11 different metrics, using the segmentation results provided by an expert pathologist as the ground truth. Experimental results demonstrate that the proposed method outperforms several existing state-of-the-art techniques commonly applied in medical image analysis.