Semi-supervised classification on “warm or cool” color in tongue images

Examination of the tongue condition is a standard diagnostic method in Traditional Chinese Medicine (TCM) and takes account of a wide variety of features including shape, texture, and color. The terms “warm”, “neutral”, and “cool” are used to refer to a kind of chromatics characteristic of the tongue color and are associated with various health states. In this paper, we propose a semi-supervised (cluster and label) scheme for tongue color analysis on “warm or cool”. In the training part, the proposed scheme makes use of a classical clustering algorithm, Expectation Maximization, to divide all pixels in tongue gamut into 150 clusters. Then we construct two auxiliary images for each cluster and manual labeling endows these clusters with category labels of “warm or cool”. Finally, each trained category on “warm or cool” is set up by sum some clusters approximately. In the testing part, we use a lookup table to divide all pixels in an input image into three distinct categories of “warm or cool”. In experiments conducted on a total of 392 tongue samples, our system achieved an accuracy of 91.1%.