Tongue verification with manifold learning

As an important human organ, tongue carries abundant individual features and has been considered as a novel biometrics. However, hitherto little work has been done it. The main reason is that it is difficult to capture the features of human tongue. This paper presents a novel framework for modeling and recognizing human tongue based on image sequences. This method exploits manifold learning dimensionality reduction, leading to a low-dimensional embedding of tongue image sequence for verification. To match the embedded manifolds, the Hausdorff distance is used for similarity measures. Classification is then achieved in a nearest neighbor framework. Based on the tongue images database, the promising experimental results demonstrate that not only the validity of tongue biometric but also the effectiveness and robustness of the proposed method.