Multiple-view Video Coding Using Depth Map in Projective Space

In this paper a new video coding by using multiple uncalibrated cameras is proposed. We consider the redundancy between the cameras view points and efficiently compress based on a depth map. Since our target videos are taken with uncalibrated cameras, our depth map is computed not in the real world but in the Projective Space. This is a virtual space defined by projective reconstruction of two still images. This means that the position in the space is correspondence to the depth value. Therefore we do not require full-calibration of the cameras. Generating the depth map requires finding the correspondence between the cameras. We use a “plane sweep” algorithm for it. Our method needs only a depth map except for the original base image and the camera parameters, and it contributes to the effectiveness of compression.