半监督学习下复杂背景图像边缘检测仿真  

Simulation of Edge Detection of Complex Background Images Under Semi-Supervised Learning

作　　者：倪波 柯亨进 蔡贤涛[2] NI Bo;KE Heng-jin;CAI Xian-tao(Computer School,Hubei Polytechnic University,Huangshi Hubei 435003,China;Computer School,Wuhan University,Wuhan Hubei 430072,China)

机构地区：[1]湖北理工学院计算机学院,湖北黄石435003 [2]武汉大学计算机学院,湖北武汉430072

出　　处：《计算机仿真》2023年第12期269-272,320,共5页Computer Simulation

基　　金：湖北省中青年创新团队计划(T201927);湖北省自然科学基金(2022CFB524)。

摘　　要：为了有效避免图像边缘检测过程中出现边缘间断或者伪边缘的情况,提出一种半监督学习的复杂背景图像边缘检测算法。设定特征筛选规则,增强复杂背景图像敏感区域,提取图像特征,通过灰度共生矩阵和Gabor滤波提取图像高频与中低频纹理特征。利用半监督学习对图像样本展开训练,实现复杂背景图像分类。采用双边滤波对完成分类的图像预处理,通过最大类间方差法展开阈值分割,引入形态学思想将分割后的图像分别展开膨胀和腐蚀,获取形态学梯度图,计算梯度图像和原始阈值图像的交并集获取具有精确边缘的灰度图,实现复杂背景图像边缘检测。实验结果表明,所提算法可以获取高精度的边缘检测结果,且不会出现边缘间断和伪边缘的问题。In order to effectively avoid edge discontinuity or false edges in the process of edge detection,this pa⁃per put forward an edge detection algorithm for complex background images based on semi-supervised learning.First⁃ly,we determined the feature filtering rule to enhance sensitive regions of complex background images,thus extrac⁃ting image features and the high-frequency and low-frequency texture features of the image through grayscale co-oc⁃currence matrix and Gabor filtering.Secondly,we used semi-supervised learning to train the image samples and thus to realize the image classification of complex backgrounds.Moreover,we used the bilateral filter to preprocess the classified images,and segmented the threshold by the Otsu method.Meanwhile,we expanded and eroded the seg⁃mented images respectively according to the morphological idea,thus obtaining morphological gradient images.Final⁃ly,we calculated the intersection set of gradient images and original threshold images,and then obtained the grayscale image with accurate edge,thus achieving the edge detection for complex background images.Experimental re⁃sults show that the proposed algorithm can obtain high-precision edge detection results without edge discontinuities and false edges.

关 键 词：半监督学习 复杂背景图像 边缘检测 滤波 形态学 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]