一种用于动态视频超分辨率的多尺度最小二乘仿射块匹配图像配准方法

动态视频的超分辨率复原中，连续各帧图像间的精确匹配具有非常重要的意义。该文提出一种基于多尺度最小二乘仿射块匹配的图像配准方法。首先定义了一个指标Dmv来衡量图像的整体和局部匹配效果，并以此为基础设计了一种多尺度块选择机制，根据图像的运动情况选择匹配块大小，以兼顾图像中运动平坦和非平坦区域的匹配效果。与传统的块匹配方法不同，该文采用基于仿射模型的最小二乘配准方法实现各图像块的匹配，并通过修正步长的归一化处理解决了不同大小图像块在匹配时的收敛问题，从而在提高参数估计精度的同时降低了算法的运算量。最后，通过实验对算法的匹配性能及其对超分辨率复原算法整体性能的影响进行了测试。实验结果表明，该方法不仅可以实现更为准确的运动估计，当用于最大后验概率MAP超分辨率复原算法时，能够进一步有效提高算法的复原性能和实现速度。

A Least Square Affine-Based Block-Match Image Registration Algorithm for Dynamic Video Super Resolution

Registration of the consecutive frames is quite essential in dynamic video super resolution. In this paper, a multi-scale least square affine-based block-match method is proposed. An index is defined to evaluate global and local matching performances of the images. Then a multi-scale scheme is designed to adjust the block size automatically according to the motions between frames, which guaranteed well performances in both planar and un-planar regions. Different from the traditional block-match method, the affine-based least square estimation algorithm is introduced for registration of each pair of blocks. Convergence of the estimating process for different size of blocks is resolved by unification of the update step, which results in improvements in both estimating precision and speed. Finally, the proposed algorithm is evaluated in both the registration performance and its affects to the performance of the super resolution algorithm. Experimental results show that the proposed algorithm not only can provide more accurate motion estimation, when be applied to the Maximum A Posterior (MAP) based super resolution method, it shows obvious enhancement in both reconstruction performances and efficiencies.