EXTRACTION OF ATTRACTIVENESS MAP OF IMAGES BASED ON PRINCIPAL COMPONENT ANALYSIS FOR IMAGES TAKEN BY PLURAL CAMERAS

This paper describes a method that extracts attractiveness map of input images. It estimates the degree of attractiveness of each part of the images using plural images taken by plural cameras. The attractiveness is expected to indicate the high possibility of existing of interest objects. We assume that the novel objects are contained in only a few images among the plural images. The degree of attractiveness of each part of the particular input images containing objects is obtained by subtracting the reconstructed background images from the input images. The adequate background images for the particular input images are inferred from other input images which does not contain novel objects. Since illumination condition is inferred from the input images which does not contain objects, appropriate back-ground images can be reconstructed without suffering the effect of contained objects. The effectiveness of the method is confirmed by computer simulations.