基于图像轮廓的角点检测算法

Corner Detection Algorithm Based on Image Contour

利用Gabor滤波器的生物视觉特性,设计了一个基于图像轮廓的角点检测器,提出利用轮廓像素的梯度方向变化信息作为角点测度。首先,用Canny边缘检测器对图像边缘进行提取;然后,利用多边形近似来改善边缘,在轮廓之间填补空隙得到封闭的轮廓;最后,将Gabor滤波器平滑后的轮廓像素的主方向角度差作为角点测度进行检测角点。实验选用15张测试图片分别在旋转和噪声下评估检测重复率和定位误差,通过与CSS、He&Yung、CPDA角点检测器进行对比实验,所提算法提高了平均角点重复率,使平均定位误差有所降低。

According to the biological visual characteristics of the Gabor filter, a corner detection based on image contour is designed, using gradient direction change information of contour pixels is proposed as the corner measure. First, the image edges are extracted by Canny edge detector. Then, polygonal approximation is used to improve the edge, to fill gaps between the contour. Finally, the main direction angle difference of the contour pixels which is smoothed by the Gabor filter as the corner measure to detect corners. Fifteen different test images are used to test repetition rate and location error under the rotation and noise, with CSS, He＆Yung, CPDA corner detection compared, the average angular repetition rate is improved, and the average positioning error is reduced.