Voting-based depth map refinement and propagation for 2D to 3D conversion

In this paper, a voting-based filter which is capable of enhancing the quality of depth map sequence and interpolates missing frames is proposed. The main concept is that if the information in the filter window is not consistent, then only the multitude decides the output. The minority will be considered as outliers. Compare to other depth map refinement method using joint bilateral filter, the outlier detection of the proposed scheme ensures that only appropriate information is involved so that the halo effects can be avoided. Moreover, in order to refine a sequence of depth maps, a space-time filtering extension is proposed. This extension refines each depth map according to the information of several adjacent frames rather than only one frame. As a result, the proposed method is capable of interpolating missing depth maps of a sequence and the errors on the depth maps are successfully reduced. The experimental results demonstrate that the voting-based filter not only provides a depth map sequence with good quality and temporal consistency but also provides several post-processing for depth maps in 2D to 3D conversion due to its flexibility.