| The light field records the irradiance of all rays in space. The 4D light field data, which is captured by the light field cameras, contains the spatial and angular information of rays that can be used to obtain depth information of the scene. Affected by the complexity of the scene, such as the occlusions and smooth texture regions, the precision of the estimated depth are not satisfactory. In this paper, we proposed a new depth estimation algorithm with pixel level accuracy, which uses the 4D structural information of light field data, the refocusing and 3D surface reconstruction in a narrow perspective based on the depth information was presented.In this paper, the depth computation was converted into calculating the disparity between adjacent views by using the correspondence between depth and disparity. The accuracy of the disparity was improved by using secondary matching and classified optimization methods, and obtained a high-precision depth map. Firstly, the disparity of each pixel in the image was calculated by using the area-based matching algorithm with weighting. Based on the analysis of error in matching process, a new confidence function was proposed. Combining with the characteristic that the smooth regions and occlusions that have different range of values when calculating the distance measurement function, the error regions was classified and identified. Thus, the secondary matching methods and optimization methods was designed to reduce the probability of mismatching, while taking into account the boundary reserved and area smooth. Thus, the disparity of each pixel was estimated accurately.The depth information was used to accomplish the digital refocusing, the 3D reconstruction and viewpoint roaming. According to the relationship between the depth of refocusing and the disparity, the refocusing algorithm based on the disparity was presented and we obtained the images refocusing at different depths. In addition, based on the geometrical theory of the imaging process, the 3D coordinates of the scene were obtained by using the depth information, and the 3D reconstruction of the scene was accomplished. Based on the 3D coordinates of the scene and the imaging process, the image of the scene can be achieved while the viewpoint is roaming in the depth direction.In this paper, using publicly benchmark light field dataset and measured light field dataset to obtain the algorithm verification, testing and evaluation of imaging accuracy. The results had shown that the proposed algorithm reduced the error in the smooth regions and occlusions. The new algorithm has better accuracy than the existing algorithms.In this paper, we proposed a high-precision depth information acquisition method based on a four-dimensional light field data. It has a good sense for the theory and application of the matching optimization in the process of disparity calculation, and the depth information acquisition of the smooth surface. While for the study of this capture mode, it can provide useful reference for the design of light field camera in some certain application. |
