Free Viewpoint Video Synthesis Based on DIBR.
Xuesong Gao,Keqiu Li,Weiqiang Chen,Zhongyao Yang,Wenguo Wei,Yangang Cai
DOI: https://doi.org/10.1109/mipr49039.2020.00064
2020-01-01
Abstract:The Free Viewpoint Video (FVV) system is an advanced visual media mode, which could satisfy the audience's demand of watching from any viewpoints. It has aroused great interests in the field of multimedia and vision. This paper proposes a set of solutions, including depth estimation, virtual view interpolation and post processing, to synthesize and optimize the virtual view images. Specifically, deep learning based method is proposed to compute depth images, which is more efficient than traditional stereo matching algorithm. The virtual view can be synthesized with the help of DIBR algorithm in an arbitrary position between two reference cameras. To remove the noise caused by rendering, some post processing techniques, such as median filtering and hole filling, are employed. The extensive experimental results demonstrate that the quality of the synthesized free viewpoint video has been significantly improved using our method.