Region partition and feature matching based color recognition of tongue image

Color recognition of tongue image is one of the key tasks in a computerized tongue inspection system in traditional Chinese medicine (TCM). Most existing methods carry out pixel-wise classification in a tongue image without considering regional information which is essential for TCM doctors to retrieve the color distribution of the substances and coatings of the tongue in its diagnosis. This paper presents a novel color recognition scheme. A color–texture segmentation algorithm has been firstly developed to obtain a number of homogenous regions; then these regions were classified into different categories of colors of substances and coatings based on Earth Mover’s distance (EMD). We also have employed the locally linear embedding (LLE) technique to eliminate the discrepancies among samples. Finally, the experimental results have shown that the proposed approach was superior to the pixel-wise classification algorithm especially in achieving meaningful areas of substances and coatings of tongue and its performance was well accepted by TCM doctors.