Abstract
A new concept is introduced to extract the essential and distinct characteristics of images. A key idea is to introduce the eigen patterns, which represents the characteristics of images independent of the resolutions and position of images on the screen. The eigen pattern of an image is obtained by projecting the pixel information in x-y screen coordinate to a red, green and blue coordinate systems. Thereby, this projection removes the location, angle and size information in a screen, and extracts the essential characteristics of the target image. We apply the notion of eigen patterns to human faces identification/synthesis with high resolution. As a result, we have succeeded in recovering and identifying the 10 human faces.
