图像序列中的特征点匹配是计算机视觉中的一个基本问题,也是目标识别、图像检索以及3维重建等问题的基础。为了提高图像匹配的精度,提出了一种针对两幅图像的高精度特征点自动匹配算法。该算法首先分析并提出两幅图像中相应特征点的邻域窗口之间的单应映射可以用仿射变换模型来近似;然后通过快速的基于仿射变换模型的迭代优化方法,不仅估计并矫正了相应邻域窗口之间的透视畸变,同时还补偿了在特征点检测阶段对相应特征点的定位误差,从而使匹配结果达到子像素级精度;最后通过真实图像的实验以及与现有算法的比较结果表明,该算法不仅得到了更多的匹配关系,还提高了特征点匹配的精度。

Feature point matching is a key problem of computer vision and is frequently used in object recognition,image retrieval and 3D reconstruction and so on.In this paper,an accurate feature points matching method for two-frame images was proposed.Since it was proved that the homography between two windows of corresponding feature can be geometrically approximated by an affine transformation model.The projective distortion of windows of corresponding feature was estimated and rectified by a fast iterative scheme based on the affine transformation model.At the same time,the location error of corresponding feature points produced at the feature detection stage was compensated using the estimated affine parameters.The matching results of corresponding feature points can achieve sub-pixel precision,which effectively improve the precisions of the final epipolar geometry.Experimental results of real images and the comparisons with other methods strongly demonstrate the validity and accuracy of the algorithm.