图像边缘轮廓自适应阈值的角点检测算法

发布时间：2017-12-28 11:19

本文关键词：图像边缘轮廓自适应阈值的角点检测算法　出处：《中国图象图形学报》2016年11期 　论文类型：期刊论文

【摘要】：目的基于边缘轮廓的角点检测算法的检测性能虽然相对比较稳定,但是它对边缘轮廓的局部变化敏感,并且只是给予一个经验门限去提取角点,为此提出一种对局部变化和噪声稳健的基于图像边缘轮廓自适应阈值的角点检测算法。方法该算法利用各向异性高斯方向导数滤波器对不同边缘和角点模型进行表征,提取表征边缘和角点的灰度及几何变化的不变属性,并通过正则化计算得到区别边缘和角点的自适应阈值。该算法首先利用Canny边缘检测器检测输入图像的边缘映射并从边缘映射中提取出边缘轮廓;然后利用各向异性高斯方向导数滤波器对所提取出的边缘曲线进行滤波平滑,计算出每一像素点的响应并与自适应阈值作比较,把响应大于阈值的点作为候选角点;最后,对候选角点进行非极大值抑制得到最终角点集。结果提出的算法分别与Harris算法,HeYung算法,以及ANDDs算法在仿射变换和高斯噪声的实验环境下进行比较,其性能指标为平均重复率与定位误差;并且对每个角点检测算法在无噪声和有噪声的情况下进行了角点匹配比较。4种算法的两个指标的平均排名为Harris 3.375,HeYung 2.625,ANDDs 2.625,本文算法1.375。本文算法在仿射变换以及高斯噪声的情况下有着良好的平均重复率和定位误差,优于其他3种算法。匹配实验中的错误点以及丢失点也少于其他3种算法。结论图像的特征检测在计算机视觉领域是一个重要的课题,在许多视觉系统中,检测特征往往作为复杂计算的第1步。因此,这一步的可靠性会极大地影响着视觉系统整体的结果。而角点作为图像的重要特征,对其研究具有重大意义。本文算法不同于传统的基于边缘的角点检测器仅利用边缘轮廓的信息,还利用到图像边缘像素的灰度信息。而且,本文算法还采用一个自适应全局阈值,避免了角点的误判。正则化的灰度变化有效减少了噪声或者光照对检测性能的影响。通过角点匹配实验、仿射变换实验以及高斯噪声实验,可以看出,本文的角点检测器拥有良好的检测性能,并且对噪声具有稳健性。
[Abstract]:Based on the detection performance of corner edge detection algorithm is relatively stable, but its local change on the edge of sensitive, and just give an empirical threshold to extract the corner point, so this paper put forward a kind of local variation and noise robust image corner edge contour detection algorithm based on adaptive threshold. The algorithm uses anisotropic Gauss direction derivative filter to characterize different edges and corner points, extracts invariant attributes that represent the grayscale and geometric changes of edges and corners, and obtains adaptive thresholds of differentiated edges and corners by regularization calculation. The algorithm uses edge mapping Canny edge detector of the input image and extract the edge contour from the edge map; then using anisotropic Gauss directional derivative filter to smooth the edges of the extracted curve, calculate the response of each pixel and compared with adaptive threshold, the response is greater than the threshold point as the candidate corner; finally, the candidate corner points of non maximum suppression final corner sets. The results of the proposed algorithm with the Harris algorithm, HeYung algorithm and ANDDs algorithm in the experimental environment of affine transformation and Gauss noise were compared, the performance index is the average repetition rate and position error; and each corner detection algorithm in the absence of noise and noise conditions compared with corner matching. The average ranking of the two indexes of the 4 algorithms is Harris 3.375, HeYung 2.625, ANDDs 2.625, and the algorithm 1.375. In the case of affine transformation and Gauss noise, this algorithm has a good average repetition rate and location error, which is better than the other 3 algorithms. The error points and loss points in the matching experiment are also less than those of the other 3 algorithms. Conclusion image feature detection is an important topic in the field of computer vision. In many visual systems, detection features are often the first step of complex computation. Therefore, the reliability of this step will greatly affect the overall results of the visual system. As an important feature of the image, the corner point is of great significance to its research. The algorithm is different from the traditional edge based corner detector using only the information of the edge contour, and also the gray information of the edge pixels of the image. Moreover, this algorithm also uses an adaptive global threshold to avoid the misjudgment of the corner points. The regularized gray change effectively reduces the effect of noise or illumination on detection performance. Corner matching experiment, affine transformation experiment and Gauss noise experiment show that the corner detector has good detection performance and robustness to noise.
【作者单位】：西安工程大学电子信息学院;
【基金】：国家自然科学基金项目(61401347) 陕西省自然科学基础研究计划基金项目(2016JM6013) 西安工程大学控制科学与工程学科群建设基金项目(107090811)~~
【分类号】：TP391.41
【正文快照】： 0引言图像的边缘轮廓上有着许多图像特征信息,例如角点、拐点。而角点是图像非常重要的特征,它往往被定义为图像边界曲线的曲率极大值点或者多个区域边界交汇点。图像的角点检测在机器视觉以及图像处理领域中有着极其重要的作用,它的应用已经遍及到许多领域。例如目标识别[1

【相似文献】