3-D Face Recognition Based on Support Vector Machine

• Main Authors: Zhi-Wei Freng, 馮芝瑋
• Other Authors: Jiann-Der Lee
• Format: Others
• Language: zh-TW
• Published: 2003
• Online Access: http://ndltd.ncl.edu.tw/handle/27544241304049806231

碩士 === 長庚大學 === 資訊工程研究所 === 91 === Abstract This thesis presents a novel scheme for 3D face recognition. The whole system consists of four stages, named preprocessing, feature extraction, feature selection and recognition. First, the body data are obtained by a 3D body scanner located in Chang Gung Memorial Hospital. These data are then cut to extract the facial region (region of interest) for normalization. In the region, fifteen feature points on the profile and frontal face are determined by using a corner detection algorithm based on the concept of enhanced chain-code. Using these feature points, a feature vector with 21 coefficients is formed for face recognition stage. To reduce the dimension of the feature vector, an efficient feature selection strategy is also proposed. Finally, the refined feature vector is used for face recognition with the aid of support vector machine (SVM). The training of support vector machine is based on the statistical learning theory. Compared with traditional neural network, SVM have shown its superior performance in computation time. With the feature selection strategy, the size of the feature vector is not only reduced from 21 to 9, but also increased the feature stability. For SVM, the training time just needs 4.6 seconds. From the experimental results, it is observed that the recognition rate of the proposed scheme is 100 % for non-rotated data set. Even the data set are rotated 40o about the base line through the nose tip; the accuracy rate can also achieve 92% above. These characteristics prove the validness and effectiveness of the proposed approach for 3D face recognition task.