Research On Face Recognition Based On Pseudo - Eigenvalue And Error Detection

The face recognition technology has been more mature under the ideal conditions (the user and the environment with controllable). However, the performance of face recognition technology still need to be improved under non-ideal conditions (with cover, attitude transformation and lighting transform). Based on the analysis of the existing algorithms, this paper makes analysis and research for the existence of a series of problems in blocking recognition of people face. Face image division, operator error computing performance, and improve the image feature extraction method robustness of several aspects are. The main innovation of this paper is showed as follows:(1) In order to overcome problem that high time complexity and threshold selection difficult caused by reconstruction algorithm in facial reconstruction, the paper proposed recognition algorithm based on linear block error detection and nonlinear face recognition algorithm. In both algorithms, images were preprocessed firstly, then use linear operator error (traditional operator error, the error log transformation of the operator) and nonlinear operator errors (Gaussian error operator, based on a logarithmic transformation of Gaussian error count son), calculate the block error face images of each region, and in the feature fusion segment occlusion region, it was given less weight to reduce the impact on the recognition results of the blocked area. Through experimental comparison analysis found that occlusion area detection accuracy rate and the error count there is a direct relationship between the child’s pros and cons, operator error and feature extraction algorithms together determine the recognition result is good or bad. The results show that using of non-linear transformation of Gaussian error number recognition algorithm operator has more advantages other algorithms.(2) KPCA algorithm based on pseudo-eigenvalues to select sample point is proposed. For classical kernel principal component analysis (KPCA) and the number of training samples to calculate the cost of the algorithm into question the existence of a positive correlation, in this paper, thought of choice of site based on pseudo-eigenvalues will be applied to KPCA, and whether keeps the point still in the training set based on the sample points on the representative of the overall size, which will contain a large number of samples from the training set of redundant information removed, not only reduce the time complexity of the algorithm but also improve the recognition rate. The results showed that: among the selected sample parameters ξ and there is a close relationship, and determine the validity of the identification system ultimately.(3) KPCA algorithm based on pseudo-eigenvalues to select sample point will be applied to block error detection in face recognition algorithm, this paper proposes a shade the face recognition algorithm based on pseudo feature and improved error detection. The improved algorithm combines the advantages of both effectively, and after AR face database testing, the experimental results show that the algorithm is indeed feasible and effective.