摘要： 传统图像拼接方法对尺度、视差和光照变化较大的图像配准效果不佳。为此，提出一种基于仿射-尺度不变特征转换(ASIFT)算法的无缝图像拼接方法。采用能够抵抗强仿射情况的ASIFT算法检测特征点，并对特征点进行提取与匹配。利用随机抽样一致算法反复迭代，找到精确的变换矩阵初始值，根据变换矩阵进行2幅图像之间的统一坐标变换，使用加权平滑算法完成图像的无缝拼接。实验结果表明，与基于SIFT特征的拼接方法相比，该方法的图像拼接效果较好。

关键词: 尺度不变特征转换, 仿射-尺度不变特征转换算法, 图像拼接, 随机抽样一致算法, 变换矩阵, 图像配准, 加权平滑算法

Abstract: A seamless image stitching method based on Affine-SIFT(ASIFT) algoithm is proposed in this paper. It implements automatically and makes up the disadvantages of traditional image registration methods under greatly different scale, parallax and illumination conditions. It adopts ASIFT algorithm for the detection of feature points, which is robust to the case of strong affine. Then it does the extraction and matching of feature points. With Random Sample Consensus(RANSAC) algorithm, it finds the accurate initial transform matrix through iteration, and accomplishes the unity of coordinate transformation between two images. The seamless image stitching is completed with smoothing algorithm. Experimental results demonstrate that this method has better performance than the method based on SIFT.

Key words: Scale-invariant Feature Transform(SIFT), Affine-SIFT(ASIFT) algorithm, image stitching, Random Sample Consensus(RANSAC) algorithm, transform matrix, image registration, weighted smoothing algorithm

中图分类号: 

• TN911.73