Research And Implementation On Virtual Viewpoint Generation Algorithm Of Autostereoscopic Display

Reproducing stereo perception and sense of reality of the objective world has become a dream since human invents display devices. In recent years, many researchers are interested in the technology of autostereoscopic display. It can reproduce the depth information of scenes and make viewers get stereo perception. Virtual viewpoint generation is one of the most important technologies of autostereoscopic display. It renders images at different viewpoints of a scene through the images photoed by two or more cameras of the same scene. The rendered image can meet the requirement of autostereoscopic display.The principle of stereo vision, stereoscopic display and the technology of virtual viewpoint generation are introduced in this paper, and it involves extracting the feature points from one of the stereo image pairs, matching them in the other one, processing images with Delaunay triangulation and render new images through morphing. Moreover, it draws disparity maps of the stereo image pairs using Mean Shift segmentation based support weight correlation window matching, renders new images at arbitrary viewpoint between the two viewpoints through interpolation, occlusion filling and noise eliminating after smoothes the disparity. Finally, it renders and displays eight images through virtual viewpoint generation algorithm based on extracting feature points and image segmentation respectively. It also analyses the effects of disparity estimation and image rendering influenced by removing the wrong matching, filling occlusion and setting bandwidth parameter of segmentation.A lot of experimental data and subjective evaluation show that the performance of the algorithm in this paper is good. The objects of the rendered image have clear outlines and most texture details are preserved, and viewers can gain stereo perception when the rendered image combined with the initial stereo image pairs.