随机抽样一致消除特征错配的一种加速算法

提出一种利用随机抽样一致方法（RANdom SAmple Consensus,RANSAC）消除图像匹配时错配的加速算法,具有平移,旋转和缩放不变性,同时允许其他方面（视角等）的较小变化,可以满足当前主流局部特征匹配算法的要求。RANSAC精度高,但时间复杂度高,时间复杂度与有效数据百分比有关,基于此提出了一类新的查找无效数据的方法,可以快速提高有效数据的比例,当它和RANSAC算法结合使用时,可以在保证高精度的情况下,有效地降低查找时间,提高图像匹配的实时性。实验表明该算法效果显著。

A Speed-up Calculation Method for Images Matching by Using RANSAC Process

A speed-up calculation method for images matching by using RANSAC Process is proposed. The proposed method is designed for the images with 3 levels of invariance： translation,rotation and scale,and at the same time it allows small changes in other respects（ view angle,etc.,which is most common for local invariant detector. Our method exploits three types of consistency to detect outliers to enhanced inlier ratio,which can lead to the time consumption of RANSAC decreasing significantly. The experimental results show that the classical RANSAC method can save significant time if our method is performed in advance.