基于不同空间结构Hadamard融合的图像边缘提取算法  

Image Edge Extraction Algorithm Based on Hadamard Fusion of Different Spatial Structures

作　　者：邵东[1] 刘志广[2] SHAO Dong;LIU Zhi-guang(Dalian Neusoft Information University, Dalian 16023, China;Dalian University of Technology, Dalian 116024, China)

机构地区：[1]大连东软信息学院,大连116023 [2]大连理工大学,大连116024

出　　处：《包装工程》2018年第17期208-214,共7页Packaging Engineering

基　　金：国家虚拟重点实验示范中心对外资助项目(LN09BYO35NJ005);辽宁省教育科学规划项目(ZX2013SK006);辽宁省高等学校优秀人才支持计划(LR2014003)

摘　　要：目的针对图像边缘提取算法中噪声对边缘的影响,易导致边缘定位精度不高,出现虚假边缘与漏检等不足,设计一种不同空间结构Hadamard融合的图像边缘提取方案。方法首先,通过计算像素与相邻点之间的方差来分析像素的结构,得到边缘点的最大概率分布矩阵(MPDM),利用MPDM来表示候选边缘集。其次,通过分析邻域点之间的亮度,计算像素与其4个相邻像素之间的最大和最小差值,得到相应的差异矩阵,并引入Logistic回归分析对2种矩阵归一化处理,得到一个权重矩阵(WM)。然后,通过Hadamard乘积模型将MPDM与WM进行融合,从而设计边缘分割阈值函数。最后,通过比较WM和分割阈值,去掉非边缘点,检测出真实图像边缘。结果实验表明,与当前边缘提取方法对比,文中方法能够有效抑制噪声,得到的边缘清晰、完整,边缘细化度与平滑度良好,在客观评价FOM与ROC中具有更大的优势。结论所提算法具有良好的边缘提取精度,在图像处理与包装条码领域具有良好的应用价值。The work aims to design an image edge extraction scheme based on Hadamard fusion of different spatial structures, regarding the effect of noise on the edge in the image edge extraction algorithm, thus easily leading to such defects as low edge location accuracy, false edges and missing detection. Firstly, by calculating the variance between pixels and adjacent points, the structure of pixels was analyzed, and the maximum probability distribution matrix（MPDM） of edge points was obtained, representing candidate edge sets by means of MPDM. Secondly, by analyzing ness between neighborhood points, the maximum and minimum differences between pixels and their 4 adjacent pixels were calculated, and the maximum and minimum difference matrices were obtained. The Logistic regression analysis was introduced to normalize the two matrices, and a weight matrix（WM） was obtained. Then, MPDM and WM were fused by Hadamard product model, and thus the edge segmentation threshold function was designed. Finally, the edge of the real image was detected by removing the non edge points based on the comparison of the WM and the segmentation threshold. The experiment showed that, compared with the current edge extraction method, the proposed method could effectively suppress the noise, and get clear and complete edges, good edge thinning degree and smoothness, and achieve greater advantages in the FOM and ROC objective evaluation. The proposed algorithm has good edge extraction accuracy, and good application value in image processing and packaging barcode.

关 键 词：图像边缘提取 HADAMARD 最大概率分布矩阵 权重矩阵 LOGISTIC回归分析 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]