Performance Measure and Efficiency of Chemical Skin Burn Classification Using KNN Method

Abstract Chemical burn injury is one of the major accidents in the world. The aim of this research is to develop an automated method of determining the severity of chemical skin burns. The severity of the chemical skin burn can be classified into three grades, namely Superficial, Partial thickness and Full thickness. Towards achieving this aim, a database of chemical skin burn images has been created by collecting images from various Hospitals. The initial pre-processing involves the contrast enhancement in L*a*b colour space. The pattern classifier technique namely K-Nearest Neighbour Classifier (KNN), has been applied on chemical skin burn images to classify them as Superficial, Partial and Full thickness burns. The help of dermatologists and plastic surgeons has been taken to label the images with chemical skin burn grades and the labelled images are used to train the classifiers. From the many features that were extracted from the images, two significant features such as the mean and then DCT were selected that best embody the differing characteristics of the three grades of chemical skin burns. The algorithms are optimized on features of pre-labelled images, by fine-tuning the classifier parameters.. The efficiency of the analysis and classification of the KNN method is about 67.5% for grade1, 82.5% for grade 2 and 75% for grade 3. The design and development of such a classifier is clinically very significant particularly, when it is used in emergency remote areas.