一种改进的KAZE特征检测算法

An improved algorithm for feature detection based on KAZE

特征点检测算法是图像匹配的基础,在机器视觉、飞机导航、图像拼接、三维重建等领域中得到了广泛的应用.其中基于非线性扩散的KAZE特征检测算法鲁棒性强、匹配率高,但实时性明显低于其他算法.针对以上问题,提出一种简单有效的改进KAZE算法.该算法通过改进特征点搜索策略、利用圆改进M-SURF特征向量描述方法并降维、引入余弦相似性度量等步骤来提高算法实时性.实验结果表明:改进后的KAZE算法在保证原算法鲁棒性、匹配率的基础上,减少了运行时间,增强了算法的实时性.

Feature detection is the basis of image matching, it has been widely used in machine vision, air- craft navigation,image stitching,3D reconstruction and so on. Among them, KAZE algorithm based on nonlinear diffusion performs better on robustness and matching rate than others, but it is slower. To solve the above problem, a simple and effective algorithm is proposed to improve KAZE. The algorithm reduces running time by improving search strategy of feature point, using round to improve M-SURF and reducing feature vector dimension, making the cosine as similarity measure. The experiments results show that the algorithm can enhance real-time under the premise of ensuring original robustness and matching rate.