Research And Implementation On Stereo Images Synthesis Based On DIBR

With the rapid development of three-dimensional graphics, three-dimensional information technology is revolutionarily influencing on the communication and working way of people. Three-dimensional information technology has been widely used in education, health care, transportation, exhibitions, films and many other industries, and its sproducts are accepted by more and more people. Therefore, stereo image synthesis technology has important research significance and application value in the 3D technology. Traditional DIBR method could quickly generate visual images, but due to the pepth of different object is not the same in a image, there will be some empty areas and discontinuous areas in the visual images, these areas make the quality of visual images poor, and finally affects the quality of stereo image. So, this paper dose research on how to elimate the empty and discontinuous areas in the visual images and the synthesis method of stereo image.This paper mainly dose research on the DIBR (Depth Image Based Rendering) algorithm after lots of references being read. The stereo image is synthesized with various image processing algorithms, and the three-dimensional effect of the stereo image is more obvious than traditional image. This paper uses the Mean Shift segmentation algorithm, by clustering and several times iterations, all the regions of the image are segmented, and this is the necessary foundation for the depth image in the next step. Compared with the traditional optical flow calculation method, the method block-based optical flow computation used in this paper is not only faster, but also access to different areas of optical flow, by comparing the optical flow of different regions, we could determine their distance relationship, and the different value could be given for different values, and finally the depth image could be obtained, the final depth image can be obtained. In the traditional DIBR algorithm, the empty and occlusion hole of the virtual view image leads to the loss of information, and makes the virtual view image not continuous, this condition finally makes the effect of the stereo image bad. This paper firstly uses the Gaussian Filter as a pretreatment for the depth image, to effectively eliminate the noise in the depth image, and then uses image interpolation techniques to fill and eliminate the discontinuities areas in the eight virtual view image. The results show that the stereo image that is obtained by the methods of this paper, their visual effect is superior than the traditional stereo image.