Noninvasive Diabetes Mellitus detection based on texture and color features of facial block

Abstract

Diabetes Mellitus (DM) is becoming one of the fastest growing and most common non-infectious diseases in the world today. This has set a huge measure of burden on governments and medical service authorities. As of now, researchers have identified that DM can be recognized in a non-invasive approach through the analysis of human facial blocks. In this paper, a novel way has been proposed to detect DM using texture and color features of facial blocks. For DM detection, four facial blocks have been selected which characterize facial image, color and texture features have been extracted from these blocks using color space transformation and 2-D Gabor Filter (GF) respectively. Input RGB image is transformed into Lab color space to extract color features. Extracted texture and color features represent the samples, where each facial block is defined by a single texture and color value. Healthy verses DM samples have been classified with accuracy 94.28% using k-Nearest Neighbors (k-NN) and 97.14% using Support Vector Machine (SVM) classifiers.