Current Issue Cover
Log-Gabor梯度方向下的角点检测

高华(西安邮电大学经济与管理学院)

摘 要
目的 角点是图像的基本特征,在图像处理与计算机视觉系统中,经常作为复杂计算的第1步,例如,目标识别、目标跟踪等。因此,角点检测器的检测性能显得尤为重要。基于此,本文提出了一个既利用到图像边缘轮廓信息又利用到图像灰度信息的基于log-Gabor梯度方向一致性的角点检测算法,以提高角点检测器的检测性能。方法 根据角点的定义可知,角点在各个方向的灰度变化都很大,并且每个角点的梯度方向与相邻像素的梯度方向都具有很大差别。然而,相邻边缘像素点的梯度方向是一致的,都是垂直于边缘脊的方向。因此,本文利用角点与边缘像素的这一特性,构建了一个新的角点测度。该算法首先利用边缘检测器检测并提取图像的边缘映射;然后利用log-Gabor虚部滤波器提取边缘像素周围的灰度变化信息,找到边缘像素点的梯度方向,利用梯度方向计算新的角点测度;最后对角点测度进行阈值化处理,得到最终的角点检测结果。结果 提出的算法分别与CPDA算法,He Yung算法,以及Harris算法在标准轮廓图像和仿射变换下进行性能比较。平均重复率与定位误差分别作为评价角点检测器检测稳定性以及定位性能的指标。结果表明,从仿真实验结果可以看到,本文提出的角点检测算法能够较好的检测到真实角点,避免对角点的漏检与误检。从旋转变换、非统一尺度变换以及高斯噪声下的平均重复率和定位误差结果中可以看出,本文算法优于其他3中角点检测算法,包括检测稳定性能和定位性能。结论:基于边缘的角点检测算法大多只依赖于图像的边缘轮廓信息,没有考虑到图像的灰度变化,而基于灰度的角点检测算法大多只考虑到图像的灰度信息。本文提出的算法既考虑到图像的边缘形状也考虑到图像的灰度变化,并且利用log-Gabor虚部滤波器充分的提取图像的局部信息。在此基础上,利用图像边缘像素的梯度方向一致性构建了新的角点测度,以提高角点检测器的检测性能。实验结果表明,本文算法拥有者较好的角点检测稳定性与定位性能。
关键词
Corner detection using log-Gabor gradient direction

Gao Hua()

Abstract
Abstract: Objective: Corner is the basic feature of image, and corner detection is an extremely task in the image processing and computer vision system. Therefore, the performance of corner detector is very important. In order to improve the detection performance of corner detector, this paper presents a new corner detector which not only combines the edge contour information with the gray information of the image, but also utilizing the consistency of edge pixels with log-Gabor gradient direction. Method: According to the definition of the corner, we know that intensities around a corner are changed extremely in every direction. And the gradient direction of a corner with the adjacent pixels is greatly different. However, the gradient direction of adjacent edge pixels is the same, are perpendicular to the ridge of edge. Therefore, this paper use this characteristic to construct a new corner measure. The proposed algorithm first uses the Canny edge detector to detect and extract the edge map of an input image. Then, the imaginary parts of log-Gabor filters are used to smooth the edge pixels along multi-directions, and the corresponding gradient directions of pixels are found. We use the gradient direction to calculate the corner measure. Finally, both the corner measurement and the angle threshold are used to remove the false and weak corners. Result: The proposed detector is compared with three various corner detectors on some published test image shapes of different sizes and under affine transforms. We also evaluate the performance under the circumstance of Gaussian noise degradation. Average repeatability and localization error are the two evaluation criteria. Experimental results show that the proposed method attains excellent performance on average repeatability and localization error under affine transforms and Gaussian noise degradation. The number of false and missed corners on published test images is less than that of the three other corner detectors in the experiments. Conclusion: The edge-based corner detection algorithm mostly depends only on the edge shape of the image, without considering the change of image gray, and the gray-based corner detection algorithm only considering the gray information of the image. The proposed method not only takes into account the image edge shape, and also considers the gray changes. The imaginary parts of log-Gabor filters are used to smooth the edge pixels along multi-directions. Meanwhile, the consistency of gradient directions of edge pixels is used to construct the corner measurement. The experimental results show that our proposed algorithm has better stability and corner detection performance.
Keywords
QQ在线


关注微信