提出了一种基于投影最大特征峰匹配的稳像算法来进一步提高灰度投影稳像算法的实时性。首先, 将参考帧与当前帧图像等分为若干区域子块, 根据灰度投影计算公式分别计算每个子块的水平投影和垂直投影; 依次计算两帧图像相应子块中对应的垂直和水平投影最大特征峰的位置差值, 作为相应子块的水平和垂直运动矢量。然后, 根据帧间运动矢量变化程度进行运动矢量修正。最后, 根据各子块的运动矢量统计结果进行全局运动估计。将该最大特征峰匹配算法运用到多光谱成像系统进行实验, 并与较经典的灰度投影算法做了比较。结果表明: 在保证相同的稳像效果前提下, 对于分辨率为1 024 pixel×1 024 pixel的图像序列在(-60, 60)搜索范围内进行运动估计时, 前者比后者的运算时间节省了近20%。提出的算法突破了现有投影稳像算法需要对可能的区域逐一搜索的繁琐匹配过程, 具有较好的实时性和良好的稳像效果。

A new image stabilization algorithm based on the characteristic peak of projection matching was proposed to improve the real-time characteristics of gray projection image stabilization method. Firstly, the reference frame and the current frame were divided into several sub image blocks, and the horizontal projection and vertical projection of each sub image block were calculated according to the projection calculation formula. Then, the horizontal and vertical position differences of the corresponding sub blocks maximal projection characteristic peaks were calculated as motion vectors, and the motion vectors were corrected according to the motion vector variance ratio between frame and frame. Finally, the global motion vectors were derived based on the sub motion vectors. The proposed algorithm based on the characteristic peak of projection matching was used in a multi-spectral imager and obtained results were compared with that of the traditional gray projection image stabilization method.The experimental results show that new algorithm has saved nearly 20% time as compared with that of the traditional one for 1 024 pixel×1 024 pixel resolution image sequence and the (-60, 60) regions of search. It does not have complex match processing that traditional method has to research the image projection area one by one, and can offer good real-time performance and stabilization effect.