Abstract: Light-field cameras have now become available in both consumer and industrial applications. It is an important research subject to reconstruct the object by using the light field camera. In the course of practical research, the spatial information of the Lytro camera and the angle information are reused in the same sensor, which leads to the low resolution of the image, and the reconstruction effect is not ideal. In order to solve this problem, this paper presents a method of sub-pixel image depth estimation, performing the sub-pixel shifts based on multi-label of sub-aperture images in the frequency domain, building the matching cost volume reference to center view image. Then the use of the guide filter suppresses noise while keeping the edge of the image well, and matching cost behavior of multi label is optimized, the accurate depth estimation results are obtained. Finally, the surface rendering and texture mapping of the target depth map are processed, and finer results are obtained. The experimental results show that the proposed algorithm can solve the problem of fuzzy reconstruction in the reconstruction of complex objects with good performance.

Key words: sub-pixel precision, multi-label, image matching, graph cut, depth map estimation, 3D reconstruction

摘要： 光场相机目前已广泛应用于消费领域和工业应用领域，利用光场相机对目标物进行深度重建成为了一项重要的研究课题。在实际研究过程中，Lytro相机空间信息与角度信息复用于同一传感器，导致图像分辨率较低，从而使得重建效果不甚理想。为解决这一问题，提出了一种亚像素精度的光场图像深度估计方法，在频率域对子孔径图像进行多标签下的亚像素偏移，以中心视角图像为参照，建立像素匹配代价行为；使用引导滤波抑制噪声的同时保持了图像边缘；对多标签下的匹配代价行为进行优化，得到精确的深度估计结果。对目标深度图进行表面渲染、纹理映射等重建处理，得到较为精细的重建结果。实验结果表明，该算法在对复杂度较高的物体进行重建时，解决了重建模糊等问题，有较好的表现。

关键词: 亚像素精度, 多标签, 图像匹配, 图像分割, 深度估计, 三维重建