ABSTRACTS

摘要

We propose to apply a principal component analysis (PCA) to a scale space. The PCA is used for a face recognition as eigen face. Because the operation to scale space from an input image is continuous, it is necessary to extend the conventional finite PCA based on square matrix to the infinite dimension. In this research, we resolved the infinite eigen problem to integral equation using the spectral theory, and we proposed to approximate the solution of the equation by the polynomial equation. We apply the spectral decomposition to the gaussian scale space and scale normalized LoG space, which is used in SIFT, and we clarify the eigen solution. As the applications of the proposed method, we introduce the method of a generation of gaussian blur image with any scale and the method of high accuracy keypoints detection.