Analysis of Range Images Used in 3D Facial Expression Recognition Systems

Abstract

With the creation of BU-3DFE database the research on 3D facial expression recognition has been fostered; however, it is limited by the development of 3D algorithms. Range image is the strategy for solving the problems of 3D recognition based on 2D algorithms. Recently, there are some methods to capture range images, but they are always combined with the preprocess, registration, etc. stages, so it is hard to tell which of these generated range images is of higher quality. This paper introduces two kinds of range images and selects different kinds of features based on different levels of expressions to validate the performances of proposed range images; two other kinds of range images based on previously used nose tip detection methods are applied to compare the quality of generated range images; and finally some recently published works on 3D facial expression recognition are listed for comparison. With the experimental results, we can see that the performances of two proposed range images with different kinds of features are all higher than 88 % which is remarkable compared with the most recently published methods for 3D facial expression recognition; the analysis of the different kinds of facial expressions shows that the proposed range images do not lose primary discriminative information for recognition; the performances of range images using different kinds of nose tip detection methods are almost the same what means that the nose tip detection is not decisive to the quality of range images; moreover, the proposed range images can be captured without any manual intervention what is eagerly required in safety systems.