基于多尺度Gabor滤波器的兴趣点检测

Interest Point Detection Based on Multi-scale Gabor Filter

为了提高兴趣点的定位准确性,提出一种基于边缘轮廓的兴趣点检测算法。首先利用Canny边缘检测器提取边缘,将提取的边缘进行间隙填补。然后利用多尺度Gabor滤波器的虚部与输入图像卷积运算较好的提取灰度变化信息,并获得不同尺度下边缘轮廓线上像素点的归一化信息熵值。最后利用不同尺度下边缘轮廓线上像素点归一化信息熵的乘积作为新的兴趣点测度提取兴趣点。与Harris,CPDA,He&Yung三种经典算法相比较,实验结果表明论文提出的算法是最优的,图像经过加噪后仍能很好的提取到兴趣点,具有较好定位准确性。

A new corner detection algorithm based upon multi-scale information entropy is proposed to enhance locationaccuracy. First,the Canny edge detector is used to extract the edge contour,which enhances the edge location accuracy. Sec-ond ,a group of imaginary parts of multi-scale Gabor filters are adopted to acquire the intensity variation of the gray level im-age. Third,under each scale,the information entropy along the gradient direction of edge pixels are computed,then the multiplication of different normalized information entropy is used as a new corner measure. The ex-perimental results show that the proposed method has better noise robustness and location accuracy than Harris,CPDA and He ＆ Yung algorithms.