基于SURF特征匹配的图像拼接算法

Image Stitching Based on SURF Feature Matching

提出了一种了基于SURF(speed up robust features)特征匹配的图像拼接算法。SURF方法是一种快速且鲁棒性较好的特征提取算法,用该算法提取图像特征后,使用改进BBF(best bin first)的快速匹配算法来寻找图像间的匹配点;用L-M算法对单应性矩阵进行优化时,本文提出使用梯度误差函数增强对光照变化的鲁棒性;最后采用多分辨率融合方法进行图像融合,有效地消除了拼接痕迹,并保持较高的分辨率。实验结果验证了该算法的高效性,对存在旋转、尺度缩放、视角以及光照变化的图像都具有良好的效果。

An image stitching algorithm based on SURF feature matching is proposed.As a state-of-art algorithm,SURF can provide fast and robust feature extraction.First,it is used to find interest points and compute stable descriptors.Then,a fast index and matching method based on BBF is proposed to get point correspondences.The homography array is optimized by the L-M algorithm in which a gradient error function is proposed to enhance the robustness of illumination change.Finally,the multi-resolutions fusion method is used to get a realistic seamless image that retains high image resolution.The experimental results show that it greatly improves the efficiency of stitching and can achieve good performance for images with rotation,scale,viewpoint and illumination change.