A Comparative Study of Features Extracted in the Classification of Human Skin Burn Depth

Abstract

The first burn treatment provided to patient is usually based on the first evaluation of the skin burn injury by determining the burn depths. In this paper, the objective is to conduct a comparative study of the different set of features extracted and used in the classification of different burn depths by using an image mining approach. Seven sets of global features and 5 local feature descriptors were studied on a skin burn dataset comprising skin burn images categorized into three burn classes by medical experts. The performance of the studied global and local features were evaluated using SMO, JRIP, and J48 on 10-fold cross validation method. The empirical results showed that the best set of features that was able to classify most of the burn depths consisted of mean of lightness, mean of hue, standard deviation of hue, standard deviation of A* component, standard deviation of B* component, and skewness of lightness with an average accuracy of 77.0% whereas the best descriptor in terms of local features for skin burn images was SIFT, with an average accuracy of 74.7%. It can be concluded that a combination of global and local features is able to provide sufficient information for the classification of the skin burn depths.