摘要
针对传统Canny算子在滤波时会模糊边缘且需要人工设置高低阈值的缺点,提出了一种基于三维块匹配的改进自适应阈值Canny边缘检测算法,并用于太赫兹三维层析成像。该算法一方面对滤波方法进行了改进,用三维块匹配(BM3D)滤波算法结合引导滤波算法代替高斯滤波算法以减少图像边缘信息的丢失;另一方面,针对传统人工设定阈值的不确定性,将梯度图进行块匹配后对三维图像块组使用最大类间方差法(OTSU)以自适应确定高低阈值。最后利用该算法对含有噪声的图像进行边缘检测处理,发现在高斯噪声方差为20时滤波后的峰值信噪比(PSNR)从22.202提升至27.151,验证了该算法去除噪声的有效性。三维块匹配改进自适应阈值Canny边缘检测算法(BM-OTSU-Canny)减少了错误边缘的数量,同时保留了连接性较好的边缘点,改善了边缘细节信息的提取效果。
Aiming at the shortcomings of traditional Canny operator in filtering,which can blur the edge and need to set high and low thresholds manually,an improved adaptive threshold Canny edge detection algorithm based on 3D block matching is proposed for terahertz 3D tomography.On one hand,the algorithm improves the filtering method by replacing the Gaussian filtering algorithm with the 3D block matching(BM3D)filtering algorithm and the guided filtering algorithm to reduce the loss of image edge information.On the other hand,in view of the uncertainty of the traditional manual threshold,the maximum inter-class variance method(OTSU)is used to adaptively determine the high and low thresholds of 3D image blocks after block matching of gradient images.Finally,the edge detection of images containing noise is carried out using the algorithm.It is found that when the Gaussian noise variance is 20,the filtered peak signal-to-signal ratio(PSNR)increases from 22.202 to 27.151,which verifies the effectiveness of the algorithm in removing noise.By using BM-OTSU-Canny algorithm,the number of false edges is reduced,and at the same time,the edge points with better connectivity can be retained,and the extraction effect of edge details is improved.
作者
於康杰
方波
李剑敏
王震
蔡晋辉
邬佳璐
何正龙
YU Kangjie;FANG Bo;LI Jianmin;WANG Zhen;CAI Jinhui;WU Jialu;HE Zhenglong(College of Metrology and Measurement Engineering,China Jiliang University,Hangzhou 330018,China;North Electro-Optics Group Co.Ltd,Xi'an 710000,China;Hangzhou Dahua Instrument Manufacturing Co.Ltd,Hangzhou 311400,China)
出处
《量子电子学报》
CAS
CSCD
北大核心
2023年第4期458-468,共11页
Chinese Journal of Quantum Electronics
基金
国家重点研发计划重大科学仪器设备开发专项课题(2018YFF01013005)。
