The Evaluation And Improvement Of Post-processing Of 3D Video Based On Disparity Infomation

With the increasing demand of people's entertainment and the increasing 3D movies on the market, 3D video has gradually become a burgeoning research hotspot in the field of computer vision and multimedia. Because of the added property of'depth'in 3D video compared with traditional 2D video, the evaluation and processing of 3D video should has some difference with 2D video's. I am focusing on the post-processing and evaluation of 3D video and have proposed a method based on the disparity infomation to evaluate the quality of 3D video.In the end,I also proposed a new post-processing algorithm based on the traditional 2D post-processing and the disparity got from the 3D video.Because the stereo cameras are impossiblely placed absolutely parallelly, the image pairs are not rectified, it will make us uncomfortable when looking at them in 3D effect and it's hard to get the disparity map.We have to rectify the images as pre-processing to simplify the calculation of the disparity map. Although we can rectify the images based on camera calibration, it will be cumbersome processes and the results will be affected by the intrinsic and external parameters of the camera's change.In addition, we can get few parameters from the video in reality.Thus it is important to do researching on the rectification with no camera parameters.We have proposed a method using fundamental matrix to rectify the stereo images, and it proved to be a good way to make the scan line rectified to meet the requirements of the subsequent matching steps.For the aim of getting the disparity information of the 3D video, when doing the stereo matching, we used the method based on'Areas'to get dense disparity map. In order to get better depth data of the interested region when shooting the scene, we adjusted the angle of stereo camera with a small bias, keeping it in convergence status. I have well studied on the image correction and stereo matching based on the model,and proposed an stereo matching method based on the results of the image segmentation using Meanshift. First, we calculate the initial disparity map and choose GCP to do the seed expanding. Then we can fit the disparity plane and get the high quality result using BP algorithm to do the global optimization finally.To make the whole algorithm high-efficiency,I have studied on how to speedup the matching processing useing GPU and the result is satisfactory. In chapter 4, we have some discussion on the evaluation of 3D video. Firstly, I processed the 3D video with the traditional post-processing method and evaluating the effect objectively and subjectively. Then we proposed a new post-processing method based on the disparity information we have got.It can balance the blockyness and the depth loss well.It proved to be a good method that can get better result around the objects'edges.In the end, chapter 5, we have some simple review and outlook to all the work. It's worthy pointing out that, the improved post-processing method proposed in chapter 4 needs accurate disparity around the edges.Therefore, it's an important research topic on how to get right disparity around the edges in the future.