Research On Distortion Model And Interactive Content Generation In3DTV

In recent years,3D industry are flourishing with increasing stereoscopic movies,displays, mobiles, projectors and cameras, most of which are based on disparitytechnology. People are perusing comfortable viewing experience and perfect depthregeneration for stereoscopic videos. But these two requirements are often againsteach other in disparity-based systems. To resolve the conflict, this thesis focuses onthe stereoscopic distortion model and interactive content regeneration.Stereoscopic distortion model is the basis of3D information processing. It firstdiscusses the mapping relationship between the camera space and the viewing space.Then it employs the shape distortion factor and the depth distortion factor to explainthe Puppet-theater effect and the Cardboard effect. The comfortable viewing zone isintroduced and its three different definitions are compared. At last, the distortionmodel and the comfortable viewing zone are combined to draw a conclusion of the3Dshooting rules.In stereoscopic content regeneration, stereo matching algorithms are veryimportant. But depth maps estimated by automatic stereo matching methods presentartifacts in flat and texture-less regions. This paper presents a friendly interactiveframework to refine depth maps for stereo matching. The left view image is firstinteractively segmented with a localized random walk algorithm using user’s strokesaround depth discontinuities as prior information. Then, the depth of each segmentedregion is recovered by surface fitting with initial depth generated by some stereomatching algorithm. Both objective and subjective results show that the proposedmethod significantly improves the accuracy of recovered depth information. Theinteraction is simple, and the refinement processing is independent of stereo matchingmethods. The proposed method can be readily extended for refinement of depthsequences, and is versatile for various stereo vision systems.