Free Viewpoint Generation Research Based On DIBR

With the development of television technology, the traditional 2D television can not satisfy people's demand for an immiscible experience watching TV. So Three-Dimensional Television (3DTV) becomes a hot research topic in resent years. 3DTV is considered to be the next generation after high definition television (HDTV) in the television history. Therefore, it is significant to research the technologies of 3DTV. Most of the state-to-art 3DTV display technologies rely on the display tools such as stereoscopic glasses and stereoscopic helmet, which restricts the popularity of 3DTV. Another method to accomplish 3DTV system is to transmit several video streams for one scene for the sending end to the receiving end. However, the transmission band requires is so broad that it will be very difficult to build a real-time television system using this method. Furthermore, the equipments used in this system will be very expensive.In this paper, an improved virtual view generation method based on depth maps was proposed to solve this problem. The traditional image rendering technologies can be divided into two kinds:the model based rendering (MBR) and the image based rendering (IBR). The quality of the rendered images is quite good using the MBR method, but computation complexity will be very high if the scene to be rendered is complex. So this kind of method is not fit for real-time rendering. The computation complexity for IBR is much lower compared with MBR and it is much easier for IBR to achieve realism since realism is in the images themselves. However, this benefit has to be paid by dense sampling of the real world with a sufficiently large number of natural camera view images. In other words, the storage and transmission of massive data is the main problem for IBR. In this condition, we propose a depth maps based image rendering method to generate any viewpoint of a scene with the color image and the depth image known. DIBR technology is the combination of MBR and IBR since geometry information is used while no model need to be constructed.Many holes appear in the newly generated images. The interpolation method for filling these holes will degrade the quality of the generated images. In this paper, we smoothed the depth map before image rendering and results showed that the quality of the images get much better. In the meantime, geometry distortion was brought in. So a new smoothing method called asymmetric was adopted to solve the distortion problem. We show in the paper this method handled the distortion problem well.Using our method in the 3DTV system, can help reduce the data amount that need to be transmitted, relieve the pressure of the bandwidth and achieve real-time display. The proposed smoothing method can enhance image quality without reducing the sense of depth. A subjective access show that the image quality generated using our method is better than that using the previous method.