半导体光电, 2016, 37 (1): 136, 网络出版: 2016-03-22
基于SIFT特征和误匹配逐次去除的图像拼接
Image Mosaic Based on SIFT Feature and Mismatch Successive Elimination
随机抽样一致性算法 SIFT特征 匹配点按块随机选取 误匹配逐次去除 RANSAC algorithm SIFT feature random block selecting mismatch successive elimination
摘要
针对随机抽样一致性算法(RANSAC)计算量大、耗时长、匹配点选取不当会影响变换矩阵精度、阈值的鲁棒性较差, 以及不能完全去除误匹配等不足, 提出了一种基于SIFT特征和误匹配逐次去除的图像拼接算法。该算法首先提取图像的SIFT特征, 并利用近似的最近邻搜索算法(BBF)进行特征初始匹配, 然后利用一种误匹配逐次去除的迭代算法正确地估计图像间的变换矩阵。在这种误匹配逐次去除的迭代算法中, 采用预检测模型的方法, 减少了迭代运算的数据量, 提高了拼接速度; 采用匹配点按块随机选取的方法保证了变换矩阵的稳定性和精确度; 通过逐次筛选去除误匹配, 且在筛选过程中采用自适应阈值, 完全去除了误匹配。实验结果表明, 该算法在保证较高精度和鲁棒性的情况下, 缩短了拼接时间, 提高了拼接效率。
Abstract
The conventional RANSAC algorithm has many shortcomings, such as large calculating amount, long calculating time, the bad influence of unsuitable matching point selection on transformation matrix precision, poor threshold robustness, and incomplete mismatch elimination. Therefore, an image mosaic algorithm based on SIFT feature and mismatch successive elimination was proposed in this paper. Firstly, the SIFT features of images were extracted and initially matched using BBF algorithm. Then an integration algorithm of mismatch successive elimination was used to evaluate the transformation matrix between images correctly. In this algorithm, the pre-detection was applied to reduce the calculating amount and increase the mosaic speed, a random block selecting method which improved the precision was adopted to select samples, and the mismatch was completely eliminated for the use of successive elimination and adaptive threshold. Experimental results show that this algorithm reduces the mosaic time and increases the mosaic efficiency with its high precision and good robustness.
张静, 袁振文, 张晓春, 李颖. 基于SIFT特征和误匹配逐次去除的图像拼接[J]. 半导体光电, 2016, 37(1): 136. ZHANG Jing, YUAN Zhenwen, ZHANG Xiaochun, LI Ying. Image Mosaic Based on SIFT Feature and Mismatch Successive Elimination[J]. Semiconductor Optoelectronics, 2016, 37(1): 136.