针对光场数据中的深度估计问题, 提出一种基于光场极平面图(epipolar plane image, EPI)的高精度深度估计方法。选用双边模型对EPI进行降噪处理, 以滤除噪声并更好地保留边缘信息; 通过采用类间方差来确定Canny边缘检测的自适应阈值, 结合最小二乘法拟合直线求取斜率, 获得边缘的深度信息; 然后运用线性插值的方法进行深度扩散, 求出非边缘区域的深度值。实验结果表明, 所提出的方法能够较好地进行深度估计, 并且能够较为有效地消除遮挡对于深度估计的影响。

In order to solve the problem of depth estimation in light field, a high-precision depth estimation method based on epipolar plane image (EPI) was proposed. Firstly, in order to filter the noise and preserve the edge information much better, the bilateral model was adopted. Then, the algorithm, which was the adaptive threshold Canny edge detection with the largest variance between the classes, was used. To obtain the slope, the least square method was used to fit the line, and the depth information of the edge could be obtained. The method of bilateral linear interpolation was to obtain the deep diffusion. The experimental results show that the proposed method could perform deep reconstruction much better and could effectively eliminate the influence of occlusion on depth estimation.