Feature Extraction And Its Application On Face Recognition

Feature extraction is an important research topic in computer vision and pattern recognition fields. The curse of high dimensionality is usually a major cause of limitations of many practical technologies, while the large quantities of features may even degrade the performances of the classifiers. Currently, many feature extraction methods have been proposed, and the most well-known ones are principle component analysis (PCA) and linear discriminate analysis (LDA). But traditional feature extraction is performed as a"batch model", all the training samples should be prepared before the training process. When the training sample is added like a queue, it's time-wasting for traditional feature extraction algorithm, so it's necessary to do some incremental research after the feature extraction. Major content of this paper is as follows:(1) Sub-pattern CAPCAThe paper presents a novel supervised face recognition method called sub-pattern based class-augmented principal component analysis (SpCAPCA). Firstly, the original images were divided into several sub-patterns by presented approach. Secondly, the SpCAPCA method was applied to the sub-patterns obtained from the previous step. Then the obtained features were combined according to a certain order. Therefore, the dimension of the original images could be reduced. This approach could not only extract the local features of the images effectively, but also adapt the complicated illumination conditions.(2) The incremental laplacian eigenmaps based on cam weighted methodThe paper presents a novel Incremental Laplacian eigenmaps which using cam weighted distance as a neighborhood selected method. Firstly, the dimension of the original data should be reduced with Laplacian eigenmaps. Then we obtain the neighbors of each added sample by using the cam weighted distance and construct the feature of the current data using such neighbors. Finally, the existing datas'embedding results need to be updated. Simulation results testify the accuracy of the proposed algorithms, other experiments show that it also has a promising performance in sample classification, face recognition, stability and time-consuming.(3) Incremental 2DMMCThe paper presents a novel incremental 2DMMC method. Firstly, extract the feature of the initial face train-set using 2DMMC, and then update the current projection space by I2DMMC. The method can not only avoid SSS problem, it can reduce the system computation and update the projection space dynamically.