The Key Techniques For Converting 2D To 3D Video

With the improvement of people's life quality, the requirement of the advanced visualization techniques of image and video becomes urgent. Meanwhile, the hardware technologies have gained great development in recent year, thereby more and more researchers start to concern about the evolution of video visualization ----from two dimension (2D) to three dimension (3D). The paper begins with the introduction of the background and development of 3D visualization technology, and then provides a review of the latest development of 3D video. It shows that the autostereoscopic display system has prospective future, where 2D video + Depth map format provided by Philips 3D display device is expected as one of the popular format for 3D video visualization. Thus, how to get the depth information only from monocular 2D video sequence becomes the main issue of this thesis. That is, we try to find a proper solution without a need to perform a complete innovation from the very beginning of the video industry.Through the analysis of important cues in Human Visual System (HVS) about how people distinguish three dimensional information, the thesis proposes three methods to obtain relative depth information between different objects. Firstly, as human eyes are more sensible to the edge of objects and recognize it as the depth discontinuities, we propose a novel color edge detector based on quaternion filters. Under the processing mechanism of human brains, people can bring the objects to the foreground with the edge of objects. Based on quaternion algebras, this thesis introduces a color edge sobel detector and tries to expand the traditional edge detectors into quaternion field. Secondly, as human eyes are more sensible to recognize distinctive motion of objects as the front scene, we present a method to acquire the depth information from the motion cue in the video. A set of quaternion Gabor filters are established in this method to reliably obtain the motion field. Basing on the gravity cue proposed by Philips Research, there is another feature of HVS that a viewer is generally provided a depth profile where the bottom part of the image has a higher depth than the top part. This thesis proposed an algorithm to further optimize the depth map based on gravity cue. Finally, we assign confidence value to the abovementioned three depth maps according to their importance in HVS. As a result, the weighted depth map is expected to provide the viewers a pleasing 3D visual experience.The conversion from 2D video to 3D video is a new research topic. It is closely related to computer vision, HVS, image and video processing. The content of this thesis contributes to the realization of conversion from 2D to 3D video.