Depth-Aware Preprocessing for Reducing Viewer Discomfort in DIBR Systems

Abstract: The advent of 3D-TV in homes has led the 3D-tv technology to develop rapidly even though voices have been raised that the technology is not mature enough to fully replace the current technology with 2D-displays. Viewers have experienced that viewing 3D-video for an extended period of time by using most of the com- mercially available techniques will cause the viewer to feel discomfort. Increasing comfort for the viewer is at the core of solving the problems with market penetration of 3D-TV. One reason for discomfort is inaccurate depth in a 3D video; another may be the technology in the display used to render the video. In this thesis the problems related to inaccurate depth information will be assessed. If there is a good way to measure whether depth is poorly recorded or estimated various methods can be applied to increase comfort at the cost of decreasing 3D- experience for the viewer. This can be done by smoothing poor depth such that it improves the comfort at the cost of reducing the 3D effects. In this report it is shown how the depth in a 3D-video for DIBR systems can be evaluated and in the case where the depth information is a poor estimation of the actual depth it is shown how discomfort can be decreased with simple methods. Suggestions for how to extend it to more fine-grained methods are also given. Objective results are used to estimate what the subjective experience would be like. Different aspects of the quality of the 3D movie are determined by assigning a score to each of them. These scores are multiplied with a set of weights which are attained from a subjective evaluation phase. This subjective evaluation phase gives reference values to the algorithm. The aspects examined are: Variance in color frames Movement in color frames Contour matching between color frames and depth maps.