Study Of Face Recognition Algorithm Based On Log-gabor Wavelet Transform And Subspace Analysis Method

In face recognition, the subspace method has been widely used in featureextraction because it is simple and effective. The traditional face recognitionalgorithms which is based on the subspace theory assume that the space face islinearly separable, However, space face is always believed to be a low-dimensionalsub manifold embedded in high-dimensional ambient space, and many methods ofsubspace feature extraction adopt manifold learning to explore the intrinsic structureof face patterns. As nonlinear manifold learning algorithms are extremely complexand it is difficult to realize incremental learning, this has seriously hindered itsapplication in the field of artificial intelligence and pattern recognition. Localsensitive discriminant analysis (LSDA) is a linear dimensionality reduction method,and has both an excellent ability to maintain the local linear structure and trulyreflects the data manifold structure.Because of its good classification performance ithas become one of the mainstream methods in the manifold learning. However,manifold learning algorithm cannot effectively eliminate the redundant informationin the vector of image characteristics; under the influence of light, pose, expressionchanges and other factors, face image features exist in a complex non-linearstructure of space, and as a linear algorithm, LSDA has inevitable defects it cannotexplore these nonlinear factors very well.This dissertation mainly deals with methods of subspace feature extraction basedon manifold learning,the main work and contributions are presented in the followingaspects:1. The face feature extraction algorithm based on Log-Gabor Wavelet transforms.Through the detailed analysis of research and experimental comparison between theGabor transform and Log-Gabor transform it is proved that the Log-Gabor transformhas a lot of advantages in face feature extraction.2. Introduction of the principle and process of the local sensitive discriminantanalysis algorithm (LSDA) and its improved algorithm OLSDA.3. First use the Log-Gabor wavelet to obtain the high order correlation of theoriginal image data set. Second use the subspace method based on manifold learningin the low-dimensional vectors, which includes the core of the algorithm orthogonallocal sensitive discriminant analysis (OLSDA). Third select the appropriateclassification to classify the extracted facial features effectively.Finally, completethe whole face recognition process.