基于DIBR的立体图像合成方法的研究与实现

发布时间：2018-05-11 22:05

本文选题：Mean + Shift　；参考：《东北大学》2011年硕士论文

【摘要】：随着计算机立体图像技术的迅速发展,3D信息技术正在革命性地影响和改变着人们的沟通与工作方式,并广泛应用于教育、医疗、交通、展览、展示、电影等诸多行业领域,其产品也越来越受到人们的青睐。因此立体图像的合成技术对提高3D信息技术具有重要的研究意义和应用价值。传统DIBR算法虽然可以快速生成虚拟视角图像,但是由于自然场景中不同物体的景深通常是不一样的,所以虚拟视角图像中会出现大范围的空洞区域和不连续区域,导致虚拟视角图像质量下降,最终影响立体图像的效果。因此,本文针对如何有效地解决虚拟图像中存在的空洞问题和不连续问题量,以及立体图像合成方法进行了研究。本文在阅读大量参考文献的同时,对DIBR(Depth Image Based Rendering)算法进行了研究,并结合多种图像处理算法来进行立体图像的合成,解决了基于传统方法建立的立体图像效果不明显的问题。本文利用Mean Shift图像分割算法将图像中的各个区域分割出来,为下一步获取深度图奠定了必要的基础。本文在研究传统的光流计算方法的基础上,提出了基于块光流计算的方法提高速度和获得不同区域的光流的新方法,通过比较各区域光流的大小来判断其远近关系,然后分别为各个区域赋予深度值来得到深度图。由于传统的DIBR算法在虚拟视角图像中存在空洞和遮挡的问题,导致信息的丢失和虚拟视角图像的不连续而影响立体图像的效果,因此本文先对深度图像进行高斯滤波预处理,消除深度图像中的噪声；然后将虚拟图像扩大到八个视角,减少了虚拟视角图像中的空洞和不连续区域；进一步利用图像插值技术填补和消除虚拟视角图像中的不连续区域。实验结果表明,利用本文方法得到的立体图像的视觉效果比较理想。
[Abstract]:With the rapid development of computer stereo image technology, 3D information technology is revolutionizing and changing the way people communicate and work, and is widely used in education, medical treatment, transportation, exhibition, film and many other industries. Its product also receives people's favor more and more. Therefore, stereo image synthesis technology has important research significance and application value to improve 3D information technology. Although the traditional DIBR algorithm can quickly generate the virtual view image, because the depth of field of different objects in the natural scene is usually different, there will be a wide range of void and discontinuous areas in the virtual view image. The result is that the quality of the virtual view image is decreased, and the effect of the stereo image is affected. Therefore, how to effectively solve the void problem and discontinuity problem in virtual image and the method of stereo image synthesis are studied in this paper. In this paper, while reading a lot of references, we have studied the DIBR(Depth Image Based Rendering) algorithm, and combined with various image processing algorithms to synthesize stereo images, which solves the problem that the effect of stereo images based on traditional methods is not obvious. In this paper, the Mean Shift image segmentation algorithm is used to segment each region of the image, which lays a necessary foundation for the next step to obtain the depth map. Based on the study of the traditional optical flow calculation method, a new method based on block optical flow calculation is proposed to improve the speed and obtain the optical flow in different regions. Then each region is given the depth value to get the depth map. Because the traditional DIBR algorithm has the problem of hollowness and occlusion in the virtual view image, which leads to the loss of information and the discontinuity of the virtual view image, so the depth image is preprocessed by Gao Si filter. The noise in the depth image is eliminated; then the virtual image is expanded to eight angles to reduce the void and discontinuous region in the virtual view image; the image interpolation technique is further used to fill and eliminate the discontinuous region in the virtual view image. The experimental results show that the visual effect of the stereo image obtained by this method is ideal.
【学位授予单位】：东北大学
【学位级别】：硕士
【学位授予年份】：2011
【分类号】：TP391.41

【相似文献】