Research On Facial Expression Recognition Based On Manifold Learning

One of the difference between human and animal is rich and complex expressions. The most direct expression of inner thoughts through joy, anger, sadness and joy. Without learning and rigorous training generally difficult to achieve undemonstrativeness, So if we can capture the subtle facial expression status were analyzed, then it can get important information from the unexpected. Facial expression recognition is a very important topic in the field of image processing and pattern recognition, but also is an important part of the realization of humancomputer interaction intelligent. In recent years, due to awareness of its importance,many researchers begin to pay attention to the related research.This paper based on the non dynamic expression image as the research object,study the correlation algorithms and algorithms based on Manifold Learning. The topological structure of manifold, face recognition and facial expression recognition are compared with experimental analysis.The main research work of this thesis is:1. The traditional subspace algorithm in face recognition:To describe Principal component analysis(PCA) and Linear descriptive analysis(LDA). A face recognition algorithm based on Manifold Learning : Isometric mapping(ISOMAP), locally linear embedding(LLE) and local tangent space alignment(LTSA) are proposed and analyzed, and experimental comparison. Experimental results show that the LTSA algorithm is compared with other algorithm has certain advantages.2. Improved LLE algorithm, the idea of LTSA applied in LLE, the linear reconstruction of LLE and ILLE is proposed. This algorithm combines the advantages of LTSA and LLE. Can have very good results on the manifold structure to maintain and face recognition, but there is still a lot of defects.3. Study on the application of improved LTSA algorithm in manifold learning generalization on the basis of in face recognition, more of the LTSA algorithm indealing with the problem of not very effective to improve the increasing data set. For this kind of deficiency, LTSA algorithm of generalized improvement have been put forward. Finally, after a series of experiments to verify the algorithm compared with other algorithms to obtain the advantages of this algorithm.