一种基于改进SURF特征描述子的快速算法  

An improved SURF Algorithm by Reconstructing Feature Descriptor

作　　者：时培弘 董淑福[1] 单勇[1] 

机构地区：[1]空军工程大学信息与导航学院,陕西西安710077

出　　处：《电子科技》2015年第10期8-10,14,共4页Electronic Science and Technology

基　　金：陕西省自然科学基金资助项目(2013JM8025)

摘　　要：在SURF算法中为保证旋转不变性需要计算出关键点主方向,并将关键点周围的梯度信息旋转到主方向后构造描述子,造成了时间上的大量开销。文中重新构造了一组各向同性特征描述子,在保证描述子旋转不变性的同时,避免了计算主方向及依据主方向旋转梯度的相关操作,进而节省了计算时间。实验结果表明,新算法在一定误差范围内计算速度比改进前的算法提高了近30%。SURF algorithm has been extensively applied in object recognition and computer vision for its speed and well-performed robustness. To keep rotation invariance in original algorithm, Bay got dominant orientation of keypoint to which gradients were computed relatively, and this part takes considerable time. In this essay, a couple of isotropic vectors are selected to reconstruct feature descriptor to keep rotation invariance as well as computing speed by using Integral Image thereby exempting algorithm from consuming time for above operations. The test has proved that the performance of proposed method is improved almost 30% allowing for error to a certain extent.

关 键 词：SURF算法 各向同性 特征描述子 旋转不变性 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]