SIFT特征匹配的显微全景图拼接

Microscopic panorama splicing based on sift features matching

针对细胞图像序列模糊、传统的特征提取方法鲁棒性不强、伪匹配点对较多、图像匹配耗时过长、融合效果不佳等问题,提出了一种强鲁棒性、快速和精确的图像拼接算法.该算法首先用基于尺度不变(SIFT)算法提取细胞图像特征点,接着采用改进的BBF(Best-Bin-First)算法对特征集进行初始的双向匹配,然后采用随机抽样一致性(RANSAC)算法对匹配点对进行进一步提纯并估算出单应性矩阵,最后根据细胞图像序列之间的单应性矩阵关系将其投影到统一标准的平面坐标系下,用具有塔型结构的多分辨率融合算法对图像进行无缝融合完成全景图拼接.实验结果证明:该算法提取到的特征点分布均匀且数量适中,误配情况明显减少,能够有效地实现显微全景图的无缝拼接.

To overcome the problem of cell image sequence blur, bad robust of the traditional extraction method, too many pseudo matching points, long time of the image matching, and bad effect of integration, etc., this paper presented a good robust, quick and accurate image stitching algorithm. The presented algorithm is based on, firstly, scale-invariant（SIFT） algorithm to extract featured points of the cell image and improved BBF（Best-Bin-First） algorithm to initially set a two-way matching of the featured set. Then, the matching point was further purified and homography estimated by the use of random sample consensus（RANSAC） algorithm. Finally, the homography relationship of the cell image sequences was projected to standardized plane coordinate system for seamless fusion of the image through a tower- typed structure fusion algorithm of multi-resolution to complete panorama stitching. The results show that moderate amount extracted feature points are evenly distributed, mismatches are reduced, and the algorithm can realize seamless fusion of microscopic panorama effectively.