Background Estimation Based on Device Pixel Structures for Silhouette Extraction

Abstract

We propose a novel technique for background estimation on the tabletop systems that have an information display on the tabletop and observe tabletop objects with a camera. In such systems, the background of an image observed with the camera includes the image shown on the tabletop display. The background varies according to the information presented on the tabletop display. Although we can estimate the background image from a displayed image based on traditional geometric registration and photometric correction, the background image is not sufficiently accurate in terms of pixel value blending for background subtraction. We estimate the background image in high accuracy based on the pixel structures of the display and the camera. We experimentally evaluate the methods by comparing the estimated image with the observed image.