从同心拼图中恢复深度信息

同心拼图（Concentric Mosaics,CM)是一种重要的基于图像的绘制方法。如果利用深度信息则可以进一步提高绘制质量并减少数据量。作者发现在CM序列中也存在着近拟的极线平面图像（Epipolar Plane Image，EPI），而且像点在EPI图像上的轨迹斜率和物点的深度呈近似线性关系。基于此发现，该文提出了一种从CM序列中自动动恢复深度信息的方法。这个方法首先利用EPI图像的频谱分析对场景的深度分布范围做出估计，然后根据全光采样的原理在EPI的斜率空间均匀的分割投票箱，并对给定窗口内频谱能量进行投票，以求得能量的最大方向，从而获得这个窗口所对应的深度，最后组合成CM的场景深度图，实验结果证实了上述方法的有效性。

Depth Recovery from Concentric Mosaics

Concentric Mosaics is an important image based rendering method. However, the depth information is always necessary when rendering with less sampling rate or navigating smoothly between two CMs. This paper introduces a method to reconstruct the depth map of the scene from the CM. When capturing using CM setup, we can also extract approximate EPI (Epipolar Plane Image), as well as lateral motion capturing. The slope of the object's trajectory on the EPI has linear relationship with the depth of the object. First, we analyze the spectral energy of EPI to estimate the scope of the scene's depth. Then we divide the voting box based on plenoptic sampling rules. For each selected window on EPI, we estimate the slop with the maximum spectral energy. Finally, we get the depth map of the scene. Both synthetic and real CM sequence data are used to evaluate our algorithm.