基于HUBER函数的序列图像位移场估计算法

摘要

An algorithm based on Huber function to estimate displ acement field is presented. Different from the optical flow, the method directl y detects disp lacement field between two frames in a sequence. Using Huber function, the metho d can alleviate the oversmooth problem at the “motion edge”. To get the displacement field, t he conjugated gradient method is used to minimize the energy function. In order to save the computational consumption, the hierarchical strategy based on wavelet transforms is taken. The experiment shows that th e average residual error using the method is smaller than usin g block-matching method for the large-scale elastic deformed image.%提出了一种基于Huber函数的序列图像位移场估计算法。与光流法不同的是，它 能直接计算位移场，还能在一定程度上减轻对于 变化剧烈的“运动边界”过渡平滑的问题。该算法采用共轭梯度法优化能量方程，并利用小 波分解进行分级计算，能快速可靠地得到稳 定的位移场。实验表明，对有较大弹性形变的图像，与块匹配法相比，该算法得到的平均 匹配残差明显减小。