4D Facial Expression Recognition

Facial expressions contain various information which represent human behaviors. It is an important way to express moods and an effective way to communicate with each other without language. People can judge others’psychological behaviors by recognizing their facial expressions. Peoples’expressions equal7%language performance,38%sound performance and55%facial expressions. Study on recognizing facial expressions becomes a hot spot in fields of pattern recognition and artificial Intelligence. Because of the development of3D scanning technique and multi attitudes on recognizing2D facial expressions, an increasing number of researchers pay attention to3D/4D facial expressions recognition, and researchers have made great advances on this in recent years. This paper also does some research in this field.Face expressions recognition system generally includes image data preprocessing, feature extracting, expression determining. This article does its exploratory research in the following four aspects:The goal of this paper is to design an automatic facial expressions recognizing system, so this paper collected original data to reccongnize facial expressions directly, not any hand-actuated mark and all of data preprocessing, feature extracting and expression identifying are dealt with by computer.During the process of data preprocessing, each original frame of3D point cloud data preprocessing are sheared, smoothed, filled holes, coordinated, and aligned grids to get normalized cloud data.After data preprocessing feature extracting is accomplished, this paper extracts dynamic feature information of4D facial expression. There are each facial action unit being extracted in each threshold value range of depth information and direction information when expressions change trend of the change.In the process of identifying facial expression, the paper establishes dynamic coding eigenvector of six basic expressions, and using nearest neighbor method to input expression for judgement.The test indicates that the algorithm recognition average recognition rate achieves83%. This results demonstrate that our algorithm is effective.