Comparing Image Feature Extraction Methods Using Dermoscopy Noisy Images

Abstract

Melanoma is a rare type of skin cancer that is regarded as the most threatening. Many studies offered a variety of effective melanoma detection strategies. The primary goal of this study is to examine existing research efforts in response to the essential and difficult problem of feature selection and extraction for skin cancer melanoma detection from dermoscopy pictures. Speeded-Up-Robust-Features (SURF), Scale-Invariant-Feature-Transformation (SIFT) and Histogram-of Oriented-Gradients (HOG) are three popular feature extraction methods that we compare. For Dermoscopic noisy pictures, we evaluate the performance of feature detection techniques. The number of valid matches found by the algorithm between the original and noisy dermoscopy images determines the efficiency of three strategies. We use dermoscopy images corrupted by three different types of noise: Gaussian noise, Salt & Pepper noise, and Speckle noise. SIFT is stable for the majority of noisy dermoscopic images, according to the findings of the trials, however it is slow. In this paper, we discovered that SURF is the best, with performance comparable to SIFT. HOG, on the other hand, proves its worth when it comes to retrieving image edge and texture information.