Two-Dimensional Technique for Image Presentation and its Application

In contrast with PCA, the two-dimension presentation technique (TDP) developed recently is very efficient. With TDP, we can easily extract feature vectors of an image matrix by projecting the image matrix rather than the corresponding vector onto projecting axes. In this paper, we present complete properties of TDP in detail and the property of decorrelation associated with TDP is originally revealed. The differences and similarities between TDP and PCA are also analyzed and presented. Furthermore, local-TDP approach is proposed to perform face recognition. Local-TDP aims to draw local characteristic of face images. Especially, local-TDP appears to be beneficial to weaken the side effect on face recognition of varying imaging conditions. The possible reason is that the varying imaging conditions mainly bring strong difference for parts of the image, while the influence on other parts is little. As a result, the similarity between the extracted local features of two face images of one individual may become larger in comparison with holistic features of face images. The conducted experiment also indicates that local-TDP is competent for extracting invariant features of face images with varying illumination