Depth Map Generation Based On2D-To-3D Conversion

With the development of3D stereoscopic display technology,3D movies have get rapid development because it can bring immersive stereo effect to the audiences. There is also a huge market demand for3D movies now. However, making3D movies is much more complex than making2D contents. It costs more money and the production period is longer. So that there is a serious shortage of3D video programs and the number of3D contents cannot meet the market demand. In addition, many of the existing abundant2D video resources are irreproducible, but people also want to watch the corresponding3D version. Therefore, it is of great importance for3D video applications to investigate an efficient3D video production method and how to convert2D video into3D video effectively.This paper analyzed the key problems in2D to3D conversion technology, and expanded researches for the core techniques in it. Specifically achieved the following results:1. We proposed a depth map generation algorithm based on static images. Algorithms like image segmentation, background depth generation based on matching depth template, foreground object depth generation based on depth changing model have been used here. Our method not only improves the existing problems such as background depth distortion, foreground depth simplification and layer effect but also achieves good subjective quality.2. This paper put forward an optimization algorithm based on RGBD image for the noises and layer effect issues in the existing depth maps. The algorithm takes "surface normal" as depth cue, smoothing filters the depth information image and solves the problem that the depth value in a plane is not continuous. The experiment results show that we can generate better viewpoint images with better subjective quality.3. A2D to3D image conversion verification system is proposed. The system has the functions of image segmentation, depth map generation for foreground and background and3D effect verification.