基于一致性随机采样的图像特征匹配鲁棒确认

误匹配点的存在影响了计算图像问变换关系的准确性,从而导致较差的图像匹配效果.通过随机采样一致性算法,提出了一种剔除错误匹配,精确确认图像匹配特征,从而计算图像间几何变换矩阵的鲁棒方法.该方法首先基于特征向量相似性准则,得到初始匹配点对,再利用特征点周围的灰度信息进行权值计算,在用随机采样一致性算法拟合几何变换矩阵的迭代过程中,得到使目标函数最小的匹配关系以筛选由噪声等引起的误匹配点对,从而精确计算图像间的几何变换关系矩阵,实现图像的精确配准.实验结果表明,该算法具有良好的噪声鲁棒性,得到了理想的图像配准效果.

Robust verification for image feature matching based on RANSAC

Error matching points usually lead to an inaccurate transformation from images into poor image registrations. In this paper, a method for verifying accurate matching features was proposed based on random sample consensus (RANSAC), and a robust algorithm for estimating geometric transform matrix between images was presented. Firstly, initial matching features were obtained from the similarity of feature vectors, and then weights could be gained from the gray information around each feature point. In the process of iterative fitting geometric transform matrix using RANSAC, error matches were eliminated by minimization cost function. Finally, precise image registration was achieved. Experimental results show that the algorithm has good noise robustness and ideal image registration.