基于熵和独特性的角点提取算法

Corner detection algorithm based on entropy and uniqueness

针对角点提取在图像配准中的应用,利用图像窗口的互相关系数定义了邻域窗口的独特性,提出一种基于熵和独特性的角点提取算法。算法首先通过Canny算子提取图像边缘,然后通过计算边缘点所在圆形邻域的熵和独特性筛选出角点,并通过不断修正剩余候选角点的独特性达到输出角点分散分布的目的。通过与Harris算法及区域特征提取的Sift算法实验对比,表明该算法能够对角点准确提取、精确定位,具有较好的抗噪性和方向无关性,且提取的角点分散分布,尤其适用于图像配准,其局限性在于不具有尺度不变性。

Comer detection is a basic problem in image processing domain. Aiming at the application of comer detection to image registration, based on correlation coefficient, the uniqueness measure at a pixel was defined, and a comer detection algorithm based on entropy and uniqueness was presented. Firstly, Canny edge detector was used to detect the edge of the image, and then entropy and uniqueness of the circle windows centered at the edge pixels, were computed. The comers were detected by selecting edge pixels with high entropy and uniqueness. And the uniqueness of remaining edge pixels was modified repeatedly in order to acquire widely dispersed comers. Compared with Harris comer detection and Sift region detection, the algorithm was more efficient in detecting comers accurately, with precise location, good noise resistance and orientation independence, and was especially suitable for image registration due to the widely dispersed comers detected, except that the comers were not scale invariant.