基于尺度不变特征的影像自动拼接

Automatic Image Stitching Based on Scale Invariant Feature

影像拼接主要存在拼接错位和拼接缝的问题。本文首先采用RANSAC算法剔除SIFT特征匹配的误匹配点并进行初始单应矩阵估计;其次引入光束法平差消除误差的累积,实现精确的全局对准从而避免拼接错位;最后采用多波段融合消除拼接缝效应。试验证明,本文实现的影像自动拼接算法很好地克服了以上问题,得到视觉效果较好的拼接影像。

There are two problems in image stitching： stitching dislocation and seam line. In this paper, firstly, we used RANSAC algorithm to eliminate error matching point of SIFT feature matching and estimate the initial homography matrix Secondly, introduced bundle adjustment to eliminate error accumulation, and realized accurate global alignment to avoid stitching dislocation Finally, by multi-band fusion to remove the seam line. Experimental results proved the effectiveness of the proposed method.