一种基于特征点的稳健无缝图像拼接算法

A Robust Seamless Image Stitching Algorithm Based on Feature Points

针对传统的基于区域图像拼接方法中,计算量大、鲁棒性差以及不能很好地解决拼接后的接缝等问题,提出了一种稳健的基于特征点的无缝图像拼接算法。在SIFT(scale invariant feature transform)提取图像特征点并匹配的基础上,通过优化的随机采样一致性(random sample consensus)算法过滤匹配点,去除误匹配点,并用过滤后的匹配点求解两对应图像间单应性矩阵初值;然后利用L-M算法优化单应性矩阵对图像序列进行拼接;最后通过改进线性加权函数法进行图像融合,很好地解决了接缝问题,实现了图像拼接处的平滑过渡。实验表明,该方法对存在旋转、尺度缩放、视角以及光照变化的图像都具有良好的拼接效果,拼接精度可以达到亚像素级。

Considering the high computational complexity, low robustness, and inefficient in solving stitching seam of traditional region-based method, based on feature points, a steady seamless image stitching method is proposed. After utilizing SIFT（ scale invariant feature transform） to acquire and match interest points, optimized random sample consensus（RANSAC） is used to filter the matching points in order to eliminate mismatched ones, then eligible points are used to calculate initial value of homography matrix between two corresponding images; image sequences is connected by homogrphy matrix which is optimized by the L-M algorithm; at last, improved linear weighted function is used to better solve stitching seam, thus a seamless transition is accomplished. Simulation results show that proposed image mosaic method can achieve good performance for images with small overlap region, large deformation, and its accuracy can reach sub-pixel level.