基于双稀疏分解的复杂图像Canny边缘检测  

Canny Edge Detection of Complex Image Based on Double Sparse Decomposition

作　　者：孟青青 李登峰[2] 肖文韬 MENG Qingqing;LI Dengfeng;XIAO Wentao(School of Computer Science and Artificial Intelligence,Wuhan Textile University,Wuhan 430200;College of Mathematical Sciences,Wuhan Textile University,Wuhan 430200)

机构地区：[1]武汉纺织大学计算机与人工智能学院,武汉430200 [2]武汉纺织大学数理科学学院,武汉430200

出　　处：《计算机与数字工程》2024年第4期1164-1168,1234,共6页Computer & Digital Engineering

基　　金：国家自然科学基金项目“稀疏框架与相关问题研究”(编号:61471410)资助。

摘　　要：在复杂图像边缘检测中,如何剔除非相干因素的影响一直是研究的重点和难点。针对以上问题,提出了双稀疏分解方法,对图像数据中干扰性强的高频特征向量进行分离。该方法利用非下采样轮廓波变换对图像进行预分解,再对高频分量进行K-奇异值分解字典学习,用得到的学习字典对图像进行稀疏表示,根据稀疏系数对应的字典原子活跃度将图像分解为高、低频两个部分。并且对Canny边缘检测算法进行改进,利用双稀疏方法分解复杂图像求得低频部分,再对较纯净的低频图像进行Canny边缘检测。仿真实验表明,双稀疏方法的稀疏分解效率更高,结合了双稀疏的Canny边缘检测结果更清晰、完整。In the edge detection of complex images,how to eliminate the influence of non-coherent factors has always been the focus and difficulty of research.To solve the above problems,a double sparse decomposition method is proposed to separate the high-frequency feature vectors with strong interference in the image data.This method uses the Nonsubsampled Contourlet transform to pre-decompose the image,and then performs K-Singular Value Decomposition dictionary learning on the high-frequency compo-nents,and uses the obtained learning dictionary to sparse the image.According to the dictionary atom activity corresponding to the sparse coefficient.The image is decomposed into high and low frequency parts.And the Canny edge detection algorithm is im-proved,using the double sparse method to decompose the complex image to obtain the low frequency part,and then the Canny edge detection is performed on the purer low frequency image.Simulation experiments show that the sparse decomposition efficiency of the double sparse method is higher,and the Canny edge detection result combined with the double sparse method is clearer and more complete.

关 键 词：稀疏表示 学习字典 K-奇异值分解 轮廓波变换 边缘检测 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]