一种优化的图像配准算法

An optimized image registration algorithm

为了降低传统尺度不变特征变换(SIFT)算法的特征点检测与匹配的时间复杂度,提出一种优化的图像配准算法,即采用Trajkovic算法检测特征点,并采用SIFT算法的分配描述符方法分配特征点描述符参数,再用稀疏降维原理对特征点描述符参数进行降维处理,最后,采用基于双向匹配的相似性度量算法进行特征点匹配。模拟实验选择检测图像的特征点数、匹配对数、正确匹配对数、匹配正确率、配准时间与配准时间下降率6个指标作为评估标准,结果表明,优化算法在特征点配准正确率方面与传统SIFT算法相当,但在特征点配准速度方面有明显提升。

To reduce the time complexity of the traditional SIFT algorithm related to the feature points detection and matching,an optimized image registration algorithm is proposed. Namely the feature points are extracted by Trajkovic algorithm,and the SIFT algorithm distribution descriptor method is adopted to allocate feature points descriptor parameters. Then sparse dimension reduction principle is used to reduce the feature points descriptor dimensions. Finally,the similarity measure algorithm based on bidirectional matching is used to match them. Six evaluation criteria including the feature points detection of image,matching pair,correct matching pair,matching accuracy,registration time and registration time drop rate are adopted in the simulation experiments. The simulation results show that the optimization algorithm in comparison with traditional SIFT algorithm is equivalent in the accuracy of feature point matching. However,it has obvious improvement in terms of feature points matching speed.