一种改进的全景图自动拼接算法

An Improved Fully-automatic Image Mosaic Algorithm

提出了一种带有图像自动排序功能的全景图全自动拼接算法。首先利用相位相关法对输入图像进行排序,同时估算角点匹配区域以加快角点匹配速度,提高匹配稳健性;然后使用RANSAC算法去除外点,采用奇异值分解法配合LM非线性优化方法求解变换参数,提高配准精度;最后使用多频带融合算法进行图像混合以改善线性加权融合算法带来的高频细节模糊;算法中还设有检测机制以确保LM优化顺利进行。与现有算法相比,对于图像重叠区域较小、图像中含有运动物体、重复性纹理等较难自动拼接的情况,使用该算法可以有效提高拼接成功率,缩短总体拼接时间,而且合成的图像清晰度较高。实验结果表明该算法具有较好的实用价值。

In this paper, an improved fully-automatic image mosaic algorithm is presented. Firstly, phase correlation is used to sort the unordered image sequence and roughly compute the translation offset between adjacent images, which speeds up comer match procedure and improves matching stability as well. Secondly, RANSAC algorithm is used to eliminate outliers to ensure effectiveness of the matched comer pairs. Then, Singular Value Decomposition-Least Square （SVDLS） method and Levenberg-Marquardt optimization are used to robustly determine the 8 parameters transform model, which aims to reduce the registration error and also improve the stability when handling very small number of matched comers. At the last of the algorithm, a multi-band blending technique is used to generate the final panorama. Invalid parameters are verified by the translation offset to make Levenberg-Marquardt optimization more successful and robust. The experimental results show that the proposed image mosaic algorithm exceeds existing ones at matching speed, stability, and it has less blur or ghost effect after blending, especially when there are noise, moving objects, repeated texture and small overlaps presented in the images.