An Efficient Depth Image-Based Rendering with Depth Reliability Maps for View Synthesis

Yi Lai,Xuguang Lan,Yuehu Liu,Nanning Zheng
DOI: https://doi.org/10.1016/j.jvcir.2016.09.015
IF: 2.887
2016-01-01
Journal of Visual Communication and Image Representation
Abstract:In this paper, an efficient depth image-based rending (DIBR) with depth reliability maps (DRM) is proposed to improve the quality of synthesized images. First, a DRM-based occlusion-aware approach is developed to obtain a segmentation mask, which can explicitly indicate where the information in an intermediate image should be blended preferably. Next, an improved weight model for view creation is introduced to enhance the quality of synthesized images. Finally, a distance and depth-based sub pixel weighted (DDSPW) algorithm is presented to solve the visibility and resampling problems. Experimental results demonstrate that the treated DIBR schemes have better performance for view synthesis than the other three methods through the subjective visual perception and objective assessments in terms of peak signal to noise ratio and structural similarity index. (C) 2016 Elsevier Inc. All rights reserved.
What problem does this paper attempt to address?