The Role of Keypoint Sampling on the Classification of Melanomas in Dermoscopy Images Using Bag-of-Features

Integrating medical knowledge on a Computer Aided-Diagnosis systems for the detection of melanomas is an essential factor for the acceptance of the system by the medical community. Bag-of-Features, a popular classification method based on a local description of an image, can be used as a means to integrate medical knowledge while developing an automatic melanoma classification system. An important step of this algorithm is the correct identification of discriminative regions, due to the great impact that it has on the algorithm’s performance. This paper aims at comparing different strategies for the extraction of interest regions. The achieved results show that texture-based detectors perform better than a dense sampling strategy, achieving Sensitivity= 98% and Specificity= 86%.