像素自相关矩阵的阈值自适应角点检测算法

发布时间：2018-06-17 01:33

本文选题：图像处理 + 算法　；参考：《农业工程学报》2017年18期

【摘要】：针对Harris角点检测算法中角点响应函数(corner response function,CRF)系数阈值与非极大值抑制系数阈值需要人为设定所造成的可变性和随机性等问题,该文提出一种通过计算图像每个像素的自相关矩阵行列式值,构造特征角点图像进行自适应阈值分割的改进Harris角点检测算法。该算法首先通过计算原图像经过方向滤波和低通滤波后各像素的自相关矩阵行列式值,以此构造特征角点图像;然后采用OTSU算法计算特征角点图像分割阈值,从而筛选出预选区域;最后结合改进的非极大值抑制方法提取有效角点。通过5组角点检测对比试验结果数据分析,不同类型图像的角点检测准确率均有提高,高分二号遥感影像的角点检测准确率提高27.06个百分点,可以初步得出,该算法相比传统Harris角点检测算法不但能够自动计算角点检测的最佳阈值,而且能够更准确地定位角点和去除边缘伪角点,从而提高了角点检测的精确度,该研究可为农业遥感影像数据检测提供参考。
[Abstract]:In order to solve the problem of variability and randomness caused by artificial setting of corner response function CRF coefficient threshold and non-maximum suppression coefficient threshold in Harris corner detection algorithm, In this paper, an improved Harris corner detection algorithm is proposed by calculating the determinant value of autocorrelation matrix for each pixel of the image and constructing the feature corner image for adaptive threshold segmentation. Firstly, the algorithm constructs the feature corner image by calculating the determinant value of autocorrelation matrix of each pixel after the original image is filtered by direction and low pass filter, and then calculates the segmentation threshold of the feature corner image by using Otsu algorithm. Finally, the effective corner is extracted with the improved non-maximum suppression method. Through the analysis of the data of five groups of corner detection and contrast test results, the accuracy of corner detection of different types of images has been improved, and the accuracy of corner detection of high score 2 remote sensing image has increased by 27.06 percentage points, which can be preliminarily obtained. Compared with the traditional Harris corner detection algorithm, this algorithm can not only automatically calculate the optimal threshold of corner detection, but also more accurately locate the corner and remove the pseudo-corner of the edge, thus improving the accuracy of corner detection. This study can provide reference for agricultural remote sensing image data detection.
【作者单位】：三峡大学理学院;中国科学院遥感与数字地球研究所;
【基金】：遥感科学国家重点实验室课题(Y6Y00200KZ)
【分类号】：TP391.41

【相似文献】