Applying L-SRC for Non-invasive Disease Detection Using Facial Chromaticity and Texture Features

Abstract

Diseases like hyperuricemia and hysteromyoma along with prediabetes (a serious health condition) are causing more suffering and hardship than ever before. Recently, computerized non-invasive diagnostic methods inspired by Traditional Chinese Medicine (TCM) have proved to be reasonable and effective using the face and/or tongue to perform disease detection. These methods no longer require bodily fluids to be extracted (e.g., a blood test), which further relieves the pain of patients and allows doctors to focus on more labor intensive activities. In this paper, we propose a novel classifier based on the fusion of the linear discriminant analysis (LDA) and the sparse representation based classifier (SRC) named L-SRC, to perform disease detection. Specifically, we collect facial images using a non-invasive capture device from those suffering from hyperuricemia, hysteromyoma and prediabetes, and feed it to the L-SRC classifier to perform classification. The experimental results show that L-SRC can discriminate samples belonging to the three classes with healthy control more effectively, achieving accuracies of 72%, 70.95% and 76.60% respectively. This indicates a promising application prospect in the future.