一种基于Sobel与K-means的边缘检测方法  

An edge detection method based on the Sobel and K-means

作　　者：杨昆[1,2] 张明新[1,2] 先晓兵[2] 郑金龙[2] 聂盼红[2] 

机构地区：[1]中国矿业大学计算机科学与技术学院,江苏徐州221008 [2]常熟理工学院计算机科学与工程学院,江苏常熟215500

出　　处：《光学技术》2014年第5期394-398,共5页Optical Technique

基　　金：国家自然科学基金项目(61173130)

摘　　要：传统边缘检测算子处理结果为边缘锐化的梯度图像,需人为确定阈值获取二值边缘图像,容易造成边缘信息丢失。应用增加了卷积模板的Sobel算子,使用K-means聚类算法基于梯度直方图自适应获取阈值,并分割梯度图像得到二值化边缘,最后对边缘细化与连接。通过使用最大类间方差法检验阈值与边缘检测结果对比分析,该方法自适应梯度阈值定位准确,所得边缘信息丰富度、定位精度、连续性均优于改进Sobel算子,与最佳阈值Canny算子检测结果基本相同,适用于机器视觉均匀稳定照明环境下获取图像的边缘检测。The result of the image processed by traditional edge detection operator is an edge sharpening gradient image.In order to get the binary edge image,it requires determining the threshold artificially.The loss of edge information is caused.So it uses the Sobel operator with increase convolution mask to get the gradient image.The threshold value through the steps K-means algorithm based on gradient histogram is got,the binary edge image is obtained by segmenting gradient image which uses the acquired threshold value,the edge is thinned and connected.The Otsu algorithm is used to check the correctness ofthreshold value and different edge detection approaches are compared with improved method.It obtains reasonable threshold value and has a better performance than improved Sobel operator and the result is close to Canny operator with optimal threshold.It can be applied in the field of machine vision.

关 键 词：边缘检测 索贝尔 K均值 阈值分割 细化连接 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]