结合相位和SIFT特征相关的图像自动配准算法

摘要

In this paper, a robust image alignment algorithm is presented for solving disorder sequential images' mosaic problem.Firstly, input images' multi-resolution pyramids are constructed, and phase-correlation sub area pairs are searched in the low resolution layer.Then,these area pairs' SIFT feathers are cross-correlated to verify the matching situation, input images' matching relation and each projective transformation matrix are initially estimated.Finally, the matched original images' SIFT feathers are cross-correlated again for exactly solving the transformation parameters.Phase-correlation method and SIFT feather correlation method are combined in this algorithm,and the multi-resolution method is also applied.So,the velocity and stability of image alignment operation are optimized by this algorithm.%提出一种鲁棒性较好的图像配准算法解决乱序输入图像序列自动拼接问题.首先为输入图像建立多分辨率金字塔,在较低分辨率图像层上进行相位相关区域搜索;接着对得到的相关区域进行SIFT特征相关验证,得到图像序列的配准关系和透视变换矩阵的初步估计;然后结合估计参数对原始图像的SIFT?特征再次进行相关匹配,最终实现图像序列变换参数的精确求解.该算法结合相位相关和SIFT特征相关两种配准算法的优点并应用由粗到精的匹配策略提高了图像配准的速度和稳定性.