Multi-view Image Generation from Depth Maps and Texture Images Using Adaptive Smoothing Filter

Since depth-image-based rendering (DIBR) is one of the rendering processes to generate virtual views from the texture image and the depth map, we investigate DIBR as a core technique for threedimensional television. In this paper, we propose a new scheme to generate multi-view images using DIBR. In order to improve visual quality of multiview images at newly exposed areas, we preprocess the depth map using an adaptive smoothing filter. In conventional algorithms, the depth map is collapsed since they applied a filter to the whole depth map. After extracting object boundaries from the depth map, we apply smoothing filters with different window sizes according to the strength of depth discontinuities. With the proposed scheme, we can not only maintain reliable depth map qualities, but also generate high-quality multi-view color images. Experimental results show that our scheme outperforms previous works in multi-view image generation and supports various functionalities.