一种新颖的基于像素跳跃加速的匹配算法

Novel template matching method based on the acceleration using the pixel-jump

提出了一种新颖的基于像素跳跃加速的匹配算法,并给出了算法中所涉及到的跳跃像素数N(动态N值和静态N值)选取的一般原则和方法。算法针对相关峰阳面的情况,为防止跳过相关峰峰值点,模板根据相关峰面所呈现出的特性,实时动态调整N值大小进行精细跳跃搜索,而针对相关峰阴面此时模板已经划过相关峰峰值点的情况,模板根据所给出的固定N值进行跳跃搜索。从而使模板搜索具有粗中有细的人眼搜索策略,集中搜索相关峰区域,极大地减少了不必要的冗余运算。通过大量的仿真实验结果表明,在保证匹配概率相近的前提条件下,本算法较未加速的互相关算法,可以将匹配时间减少至10倍以上;而且在基准图大小固定时,随着实时图尺寸的增大,算法消耗匹配时间少的优势会越明显,甚至优于二层金字塔分解加速方法。

A new matching algorithm based on the acceleration using the pixel-jump is proposed, meanwhile, the general methods and principles determining the amount of the pixel-jump N （including are introduced in this paper. According to the condition of the adret of the both static value N and dynamitic value N） correlation peak, the template timely and flexibly adjusts the value of Nto carry out a fine search in order not to omit the climax of the correlation peak. As for the nightside of the correlation peak, the template performs a bounded search according to the given value N because it has passed the climax of the correlation peak. Thereby the searching is analogous to visual perception mechanism of human being, which focuses on the area of the correlation peak, searching in a highly discriminating way. Therefore, it reduces unnecessary redundant operation. The simulation experimental results show that the improved algorithm maintains a high probability of correct matching, and the running time is less than 1/10. Moreover, when the size of reference image is fixed, the advantage of this time-saving algorithm is more obvious as the increase of the size of the real time image. The algorithm is even better than the two-level pyramid decomposing accelerating method.