基于RANSAC的图像拼接方法

为解决常用SUSAN角点检测的阈值都是固定的,检测出的角点经归一化互相关后直接使用RANSAC(即随机抽样一致性算法)得到的单应性矩阵准确度不高等问题,给出了一种改进的RANSAC图像拼接方法.利用具有自适应阈值的角点提取算法抽取图像的角点特征,采用引导匹配的方法重新进行归一化互相关,增加了准确匹配点的个数,提高了单应性矩阵的估算效果,最后采用拉普拉斯金字塔对配准后图像进行分层融合、拼接.实验结果表明,该方法比常用方法具有更好的矩阵估算效果,拼接效果良好.

Image mosaic method based on RANSAC

To solve the fixed thresholds of common SUSAN corner detection and the homography accuracy problems caused by the corner directly preliminary match using NCC, an improved RANSAC method is put forward here. The image feature points is extracted by corner extraction algorithm with adaptive threshold and NCC is carried out again by the guide matching method. The number of matching points is increased and estimated effects of the homography is improved. Finally, hierarchical fusion for image registrated is achieved using laplace pyramid method. The results show that this method is better than the common method for estimating the effect of the matrix and has a good mosaic effect.