Abstract: Motion estimation is an essential process in the super-resolution reconstruction which directly effects the quality of resultant image. To reduce the searching points of motion estimation and improve the search speed, this paper proposes an improved small cross-diamond search algorithm based on Gaussian Pyramid. A Gaussian Pyramid with two layers is constructed. The first layer is used for estimating the initial motion vector by using the small cross-diamond search algorithm, and the early termination strategy is adopted to terminate the algorithm in advance, while the second layer evaluates the final offset value based on the initial motion vector produced in the first layer. Experiment is conducted with standard image sequences. The experimental results show that the proposed algorithm can greatly improve search speed within the desirable search accuracy, and especially for the sequences with large motion deviation.

Key words: Gaussian Pyramid, small cross-diamond search algorithm, early termination strategy, motion vector

摘要： 运动估计是图像超分辨率重建中的关键环节，直接影响超分辨重建的结果质量。为减少运动估计搜索点数，提高搜索速度，提出一种基于高斯金字塔分层思想的小十字形搜索算法。算法把图像构建成一个两层高斯金字塔，在上层使用小十字形搜索算法估计初始运动矢量，并通过提前终止策略来提前结束搜索；在下层以上层估计结果作为初始值，估计最终图像运动矢量。在标准图像序列上进行的实验结果表明，该算法在保持搜索精度的前提下能明显提高搜索速度；特别对于在运动偏差较大的情况下，提高效果更显著。

关键词: 高斯金字塔, 小十字形搜索算法, 提前终止策略, 图像运动矢量