基于成像一致性检测的光场图像快速深度估计方法
2019-09-19徐翎丰
徐翎丰



摘要:光场相机可以通过单次成像同时记录场景中光线的强度和角度信息,这使其在深度估计中具有独特优势。但目前的深度估计方法普遍存在计算量大、耗时长的问题,使得深度估计方法的实时性较差,限制了深度估计的使用场景。针对这一问题,本文从成本量函数和优化方法两方面进行了改进。实验结果表明,本文方法在提高了深度估计精度的同时减少了计算耗时,提高了算法的实时性。
关键词:深度估计;成像一致性检测;数字重聚焦;光场
中图分类号:TP18 文献标识码:A
文章编号:1009-3044(2019)20-0233-02
开放科学(资源服务)标识码(OSID):
Abstract:The light field camera can simultaneously record the intensity and angle information of the light in the scene by one exposure, which makes it have a unique advantage in depth estimation. However, the current depth estimation method generally has a large computational complexity and a long time-consuming problem, which makes the depth estimation method have poor real-time performance and limits the usage scenarios of the depth estimation. In response to this problem, this paper improves by constructing a new cost function. Firstly, digital refocusing is performed by using the original image of the light field to obtain the light field images of different depth planes. Then, the image photo-consistency detection is performed on the light field image of each depth layer, and the cost function is constructed, and adaptively acquired by the principle of minimum cost. The optimal cost at each point is obtained from the initial depth map.Finally, the fast weighted median filter is used to optimize the final depth map. The experimental results show that the proposed method reduces the computation time while ensuring the accuracy of depth estimation and improves the real-time performance of the algorithm.
Key words: depth estimation; photo-consistency detection; digital refocus; light field
當前,快速、高精度三维重建和三维成像已经成为市场的一大需求。三维技术的核心——深度估计也逐渐成为研究热点。传统的深度估计设备主要是传统相机或是相机阵列,前者单次拍摄所获取的图像信息量低,后者便携性较差。光场相机的出现解决了这一问题。光场相机是当今一种新兴的相机种类,在传统的相机透镜后加入微透镜阵列使得它可以通过单次成像同时记录场景中光线的强度和角度信息。
目前,利用光场图像进行深度估计的方法主要基于立体匹配的方法、基于极平面图像(EPI)的方法以及基于重聚焦的方法。基于立体匹配的方法中比较有代表性的是Jeon等人[1]的方法。他们较好地解决了光场相机的窄基线问题,但算法依然会受场景中的阴影、纹理和弱纹理区域影响。基于重聚焦的深度估计方法中较有代表性的是Tao等人[2]的方法,他们用马尔可夫随机场将匹配和散焦两种线索融合以进行深度估计,但其全局优化方法不仅增加了计算时间,而且使得深度估计结果过于平滑。……
