A novel improvement to PCA for image classification
Weihao Zheng, Yan Zhang · 2011
Block PCA was proposed as an improvement to principal component analysis (PCA) several years ago. Block PCA is noticeable owing to its promising performance in image classification such as face recognition. It achieves this by dividing an image into a number of blocks and applies the PCA method to the obtained blocks rather than the whole images. With this paper, we propose a novel improvement to PCA for image classification. This improvement is very different from PCA in the following way: besides it requires that the transform results of all of the training samples have the maximum variance, it also requires that the transform results of all of the training samples from the same subject have the maximum variance. The experiments indicate the effectiveness and feasibility of the proposed improvement.