一种面向图像拼接的快速配准算法

A fast registration algorithm for image stitching

针对传统图像匹配算法在高分辨率的无人机图像匹配过程中时间开销较大这一问题,提出了一种SIFT—PCA相结合的算法,该算法从两个方面进行改进。极值点搜索策略方面:增加相邻像素点的比较域,由原来的26个极值判定像素点扩充到42个,从而减小特征点的数量,为以后的配准奠定基础。特征向量维数方面:利用PCA主成分分析算法降低矩阵维数。最后采用最近邻法和改进的RANSAC进行特征点匹配。实验结果表明:改进后的算法时间开销较少且后续拼接效果较好。

To solve the problem that traditional images matching algorithm has high time cost in highresolution UAV image registration process,a new algorithm which is based on SIFT was presented. It improved SIFT algorithm from two aspects. One is extreme point search strategy,the other is a combination of PCA technology. On the aspect of search strategy,it increased the adjacent pixel of the comparison field,from the original 26 to 42. There by reducing the number of feature points and laid a foundation for future registration. On the aspect of PCA technology,the dimensionality of eigenvectors is reduced.Finally,the nearest neighbor method and the improved RANSAC are used to match the feature points. The experimental results show that the improved algorithm has less time overhead and better splicing effect.