Texture Aided Depth Frame Interpolation

Yongbing Zhang,Jian Zhang,Qionghai Dai
DOI: https://doi.org/10.1016/j.image.2014.05.005
IF: 3.453
2014-01-01
Signal Processing Image Communication
Abstract:Recent development of depth acquiring technique has accelerated the progress of 3D video in the market. Utilizing the acquired depth, arbitrary view frames can be generated based on depth image based rendering (DIBR) technique in free viewpoint video system. Different from texture video, depth sequence is mainly utilized for virtual view generation rather than viewing. Inspired by this, a depth frame interpolation scheme using texture information is proposed in this paper. The proposed scheme consists of a texture aided motion estimation (TAME) and texture aided motion compensation (TAMC) to fully explore the correlation between depth and the accompanying textures. The optimal motion vectors in TAME and the best interpolation weights in TAMC are respectively selected taking the geometric mapping relationship between depth and the accompanying texture frames into consideration. The proposed scheme is able to not only maintain the temporal consistency among interpolated depth sequence but also improve the quality of virtual frames generated by interpolated depth. Besides, it can be easily applied to arbitrary motion compensation based frame interpolation scheme. Experimental results demonstrate that the proposed depth frame interpolation scheme is able to improve the quality of virtual view texture frames in both subjective and objective criterions compared with existing schemes.
What problem does this paper attempt to address?