图像匹配中去除误配的对抗性优化方法的改进

摘要

Applying the adversarial optimization approach to remove the error matches in image matching usually causes the removal of correct matches, especially when multiple iterations are run. Concerning this drawback, a limit was put on the number of iterations, and a subsequent processing was added, which re-estimated the homography matrix using a Random Sample Consensus (RANSAC) -like method. The experiments show that the improved method can preserve almost all the correct matches with smaller root mean square error. And in the aspect of computing speed, the time the improved method needs is less than half the time the original method needs.%将对抗性优化方法用于图像匹配中的去除误配,容易将正确匹配对也移除,尤其是在迭代次数较多的时候.针对该缺陷,限制对抗性优化方法的迭代次数,并增加后续处理过程,以类似随机抽样一致性(RANSAC)的方法来重新估计单应性矩阵.实验结果表明,改进的方法能保留几乎所有正确匹配,并且具有更小的均方根误差.尤其是在计算速度方面,改进的方法比原方法的计算时间缩短了一半以上.