This document proposes a noninvasive method to detect diabetes mellitus using facial images. It extracts texture features using a Gabor filter bank and color features in the Lab color space from divided facial blocks. These features are classified using k-NN and SVM algorithms. Experimental results show classification accuracies up to 94.28% for k-NN and 97.14% for SVM, providing an alternative to invasive blood glucose tests. The method aims to detect differences in facial skin associated with diabetes status based on traditional Chinese medicine hypotheses.