基于改进ORB和匹配策略融合的图像配准方法

Image Registration Method Based on the Fusion of Improved ORB and Matching Strategy

针对Oriented Fast and Rotated Brief(简称ORB)算法在图像配准中不具有尺度信息、BRIEF算法不具有旋转不变性以及特征匹配精度低等问题,本文提出一种基于改进ORB和匹配策略融合的无人机航拍图像配准方法。首先,通过简化多尺度空间检测特征点,采用稳健的RootSIFT描述符进行特征描述,用来提升特征描述的稳定性。然后,结合距离比值、双向匹配和余弦相似性测度(Cosine Similarity)匹配策略获得初始匹配。最后,利用随机抽样一致性算法(RANSAC)确定最终的特征点匹配关系,计算空间变换参数,实现图像的精确配准。实验结果表明,该方法在图像尺度、旋转和光照变化方面均表现出良好的性能,并提高了图像匹配准确率和配准精度。

Aiming at the problems that ORB(Oriented Fast and Rotated Brief)algorithm does not have scale information in image registration,BRIEF algorithm does not have rotation invariance and low feature matching accuracy,this paper proposes an unmanned aircraft aerial image registration method based on the fusion of improved ORB and matching strategy.First,by simplifying the multiscale space detection of feature points,the robust RootSIFT descriptor is used for feature description,which is used to improve the stability of feature description.Then,the initial matching is obtained by combining the distance ratio,two-way matching and cosine similarity matching strategy.Finally,the Random Sampling Consistency Algorithm(RANSAC)is used to determine the final feature point matching relationship,calculate the spatial transformation parameters,and achieve accurate image registration.The experimental results show that the method has good performance in image scale,rotation and illumination change,and improves the accuracy of image matching and registration accuracy.