The median filtering heuristic is considered to be an indispensable tool for the currently popular variational optical flow computation. Its attractive advantages are that outliers reduction is attained while image edges and motion boundaries are preserved. However, it still may generate blurring at image edges and motion boundaries caused by large displacement, motion occlusion, complex texture, and illumination change. In this paper, we present a non-local propagation filtering scheme to deal with the above problem during the coarse-to-fine optical flow computation. First, we analyze the connection between the weighted median filtering and the blurring of image edge and motion boundary under the coarse-to-fine optical flow computing scheme. Second, to improve the quality of the initial flow field, we introduce a non-local propagation filter to reduce outliers while preserving context information of the flow field. Furthermore, we present an optimization combination of non-local propagation filtering and weighted median filtering for the flow field estimation under the coarse-to-fine scheme. Extensive experiments on public optical flow benchmarks demonstrate that the proposed scheme can effectively improve the accuracy and robustness of optical flow estimation.

中值滤波试探法被认为是当前流行的变分光流计算不可缺少的工具。其吸引人的优点是，在保留图像边缘和运动边界的同时，可以减少异常值。但是，由于大的位移，运动遮挡，复杂的纹理和照明变化，它仍然可能在图像边缘和运动边界处生成模糊。在本文中，我们提出了一种非本地传播滤波方案来解决从粗到精的光流计算过程中的上述问题。首先，我们分析了从粗到精的光流计算方案下加权中值滤波与图像边缘和运动边界模糊之间的联系。其次，为了提高初始流场的质量，我们引入了非局部传播滤波器，以减少离群值，同时保留流场的上下文信息。此外，我们提出了一种非局部传播滤波和加权中值滤波的优化组合，用于在粗到精方案下进行流场估计。大量的公共光流基准测试表明，该方案可以有效提高光流估计的准确性和鲁棒性。