The Research On Virtual View Generation

Virtual view generation is one of the key technologies in Multi-view 3DTV system. In this thesis, virual view generation technology was on the focus. Disparity image generation technology and virtual viewpoint generation technology, as well as image scaling technology in autostereoscopic display were under the research.To deal with the disparity estimation problem in virtual view generation, segmented based stereo matching scheme was utilized. An improved voting based disparity plane estimation algorithm was proposed, while disparity plane refinement was implemented to revise the inaccurate disparity.Virtual viewpoint generation technology was researched under the parallel camera configuration. Two Methods of rendering were implemented:forward warping and backward warping. Holes in forward warping were first filled with the two reference images, while a priority-order median filtering algorithm was proposed to deal with the remaining holes. Disparity image of the reference views were first mapped into the virtual viewpoint in backward warping, which generate the virtual view disparity image, then holes in the virtual view disparity image were filled. Finally, virtual view image were rendered using the reference image and the virtual view disparity image.Image scaling technology was also researched in this thesis. Based on the sparse and direction-sensitive features of the coefficients of the Nonsubsampled Contourlet Transform, an edge direction adaptive interpolation algorithm was proposed, which was implemented in the image enlargement and image reducement. Experimental results showed that the proposed image scaling algorithm dealed with the blurring problem effectively, and improved the subjective quality of the interpolated images.