Research On Some Problems Of Facial Expression Recognition

Facial expression recogniton is a technology which compares facial expression features with the ones stored in database in advance to identify facial expression. The technology has a high social value and development prospects in pattern recognition, intelligent control and other respects.The research of facial expression recognition mainly includes facial expression extraction, feature reduction and classification, among which feature extraction and feature reduction are the key technologies which can ensure the recognition and on-line performance of algorithms. This thesis devotes main efforts to facial expression recognition research in the stationary images and the comparision researches of DCT (Discrete Cosine Transform) and 2DPCA (Two-Dimensional Principal Component Analysis) feature dimension reduction algorithms, mainly including:(1) Some problems of facial expression features automatic extraction and higher feature dimension are studied. Firstly, facial expression feature points are extracted by use of ASM (Active Shape Model). Secondly, creating the feature vector related with the feature points is explained based on Gabor wavelet. Finally, PCA (Principal Component Analysis) along with LDA (Linear Discriminant Analysis) is used to finish the feature dimension reduction and the facial expression recognition, and simulation experiment is given. This method improves the degree of automation of facial expression feature extraction, reduces the feature dimension effectively, and have a better recognition of facial expression.(2) The effective dimension reduction of facial features is studied. With respect to the similarities between the principal of image compression and image feature reduction, this thesis brings the classic algorithm(DCT) in image compression to the domain of feature dimension reduction of facial image. Firstly, the feasibility of DCT to be used in image feature dimension reduction is analyzed, and the basic principle of the image feature dimension reduction based on DCT is given. Then 2DPCA algorithm is introduced, and the differences between the algorithms of DCT and 2DPCA are compared. At last, the experiments by use of DCT and 2DPCA algorithms are operated on the facial expression images, and the results prove the effectiveness of the presented method in this paper.