MULTI-VIEW VIDEO GENERATION FROM 2-DIMENSIONAL VIDEO

This paper proposes a region-based method to generate a multi-view video from a conventional 2-dimensional video. It first segments an image according to its color information and our scheme is based on the resulting segmented regions. The color information is used to extract the boundary of an object, which is very hard if the motion information is used. To classify the homogeneous regions, both the luminance and chrominance components are used. A pixel-based motion estimation is also performed to obtain accurate motion information. Then, we convert the motion values to corresponding depth values for each segmented region. Finally, we generate multi-view video by applying a rotational transformation method to 2-dimensional input images using the obtained depth information of each region. The ability and the resulting image quality. are tested by a subjective assessment, called Double Stimulus Continuous Scale, and compared with an existing method. The experimental results showed that the image quality of the proposed algorithm was much better than the precious one.