A Hybrid Feature Extraction Method for Skin Lesion Detection

Early detection is important for the effective cure of skin cancer because it is a disease with potentially fatal consequences. The formal method used to diagnose skin cancer is the biopsy method, which is a laborious and extremely unpleasant process. This research study suggests a thorough method for the early detection of skin cancer. To improve the quality of Digital Dermoscopic (DD) images by eliminating noise and emphasizing important characteristics, a multi-filter Fusion and Equalization (MFE) is suggested as part of the preprocessing stage. To ensure accurate delineation of lesion boundaries, a boundary contrast-based Otsu threshold approach (BCOT) is presented for segmentation. To enhance the robustness of feature extraction, incorporating local variations in texture along with spatial relationships is crucial. resulting in a more powerful and discriminative feature set that captures both structural and textural information for improved recognition tasks. This is achieved through the combined utilization of the grey level co-occurrence matrix (GLCM) and the Local Optimal Oriented Pattern (LOOP) method. GLCM captures intensity-based information, while LOOP further enhances the analysis of pixel patterns, collectively enriching the extracted features. In the classification step a machine learning algorithm, Random Forest (RF) is utilized to classify the skin lesions. The Kaggle database contains a dataset of 800 images of benign and malignant types of oncological diseases that were sourced from the ISIC and are used in this study. This method provided an accuracy level of 98.75%. MATLAB R2023b is utilized in this study to build the skin cancer detection model.