3DTV, as the next revolution in visual technology, promises to bring to the customers a new generation of services. Enjoy three-dimensional entertainments, navigate freely around a sportive show, to name but a few of the new promising 3DTV applications. Another target fields can be expected, like Digital Cinema, IMAX theaters, medicine, dentistry, air-traffic control, military technologies, computer games, and more. In the meantime, the development of digital TV and autostereoscopic displays allow to easily introduce 3D in broadcast applications like television. The creation and the transmission of autostereoscopic content have to be thought with the new broadcast constraints, and especially with two of them: the adaptivity with respect to the different receiver capabilities and the backward compatibility allowing to extract the 2D information for existing 2D displays.
Among the various studies, recent researches give much attention to 3DTV, more specifically to depth image-based rendering (DIBR) approaches. Indeed, DIBR technique has been recognized as a promising tool which can synthesize some new ”virtual” views of the scene from the so-called video-plus-depth data representation, instead of using the former 3DTV proposals, such as 3D models or stereoscopic images. The video-plus-depth data representation uses a regular color video enriched with the depth map providing the Z-distance for each pixel. This format is currently standardized by the MPEG within the MPEG-C part 3 framework of the compression of the per-pixel depth information within a conventional MPEG-2 transport stream.
Classically the correlation between the texture motion vectors and the depth sequence motion vectors is not exploited in the coding process. One of the aims of my past researches was to reduce the amount of information for describing the motion of the texture video and the depth map sequences by sharing one common motion vector field. Intuitively, the texture video and the depth map sequences have common characteristics, since they describe the same scene with the same point of view. For that reason, in both domains (color-surface structure and distance information) boundaries coincide and the direction of motion is the same. My approach exploits the physical relation between the motion in both videos, the texture and depth map videos.
Furthermore, considering a fixed global bitrate we proposed a new distribution of the available bandwidth by establish an objective ratio between the number of bits of the texture and depth map. We optimize the bandwidth repartition between the texture and the depth map data.
Once the 3D data have been encoded and transmitted over the TV broadcast, some “virtual” sequences are generated from the transmitted video at the TV set-top boxes for the 3DTV purposes. However, due to the 3D rendering, some areas that were occluded in the transmitted sequence become visible in some “virtual” sequences. To deal with this problem we propose a new filtering technique for DIBR allowing to reduce or completely remove the holes without degrading the entire depth map. The proposed filter account the distance to object boundaries. The advantage is not to introduce distortion in the depth map when it’s not necessary. Experiment results have illustrated the high efficiency of the proposed method.
Keywords : 3DTV, DIBR, video-plus-depth, depth map, MPEG-C part 3, motion vector, bit allocation.
