Distance Dependent Depth Filtering in 3D Warping for 3DTV

Abstract

Depth image-based rendering (DIBR) is the process of synthesizing some new "virtual" views from one "real" view and the associated per-pixel depth information. The most important problem in this process is to deal with the newly exposed areas (holes) appearing in the virtual images. One common solution to decrease the number of holes is to pre-process the depth map, before the warping. In this paper, we present a new filtering technique for depth image-based rendering. In order to reduce or completely remove the newly exposed areas an efficient smoothing is necessary for the sharp depth changes near object boundaries. In the meantime it is useless to filter the smooth areas in the depth map. Our solution is based on a weighted Gaussian filter taking into account the distance to the contours. By this way, the geometric distortions and the computation time are reduced compared to a uniform filtering of the depth map. We present some results in the context of creation of stereoscopic views for 3D TV.

Depth-Based Multiview Distributed Video Coding

“…The optimal value of this parameter is empirically determined (Section V). The second processing step consists in slightly enlarging the mask, in order to take into account the fact that there may be other actually occluded pixels falling outside the estimated mask [34]. We show in Fig.…”

Plenoptic Layer-Based Modeling for Image Based Rendering

“…There are many ways to use the layer model to synthesise new image views, various image surface warping techniques applied to the entire image have been proposed in [47][48][49]; although the resultant output is a complete image, it may be significantly distorted and often fails to fully take into account the occlusions and disocclusions inherent in the set-up. An alternative approach is a rigid layer shift accounting for the occlusion ordering on the layers, this models a scene more accurately but dissoclusion may lead to gaps in the final output which need to be filled as discussed in [50].…”

Adaptive wavelet coding of the depth map for stereoscopic view synthesis

“…To deal with this issue, interpolation techniques [6] can be used to fill in these occluded areas. Another way consists in pre-processing the depth map [7] or using two reference images [8].…”