基于改进MRF的冲压件轮廓缺陷图像分割算法  

Image segmentation algorithm on contour defects for stamping part based on improved MRF

作　　者：吕宁 肖剑[2] 高健 欧阳雪峰 罗忠洁 Lyu Ning;Xiao Jian;Gao Jian;Ouyang Xuefeng;Luo Zhongjie(School of Mechanical Engineering,Yangzhou Polytechnic College,Yangzhou 225009,China;School of Automation,Harbin University of Science and Technology,Harbin 150080,China)

机构地区：[1]扬州职业大学机械工程学院,江苏扬州225009 [2]哈尔滨理工大学自动化学院,黑龙江哈尔滨150080

出　　处：《锻压技术》2022年第4期101-109,共9页Forging & Stamping Technology

基　　金：扬州市“绿扬金凤计划”高层次创新创业领军人才引进项目(2021CX044)。

摘　　要：针对冲压件在生产过程中产生的表面缺陷视觉检测问题,提出一种改进的马尔可夫随机场图像分割算法。首先,应用基于像素的马尔可夫随机场算法,获取像素特征,提取基于像素的似然函数。采用随机区域合并算法获得区域特征,提取基于随机区域合并的似然函数。利用最大梯度算法获得图像的边缘特征,提取基于边缘的似然函数,用以恢复随机区域合并过程中丢失的边缘信息。融合3种似然函数,根据能量最小准则,实现图像分割。通过与传统图像分割方法的对比实验,验证了该算法的有效性。实验结果表明,改进算法可实现冲压件图像的精准分割,应用效果较好。For the problem of visual inspection for surface defects of stamping part during the production process, an improved Markov Random Field(MRF) image segmentation algorithm was proposed. First, the pixel-based MRF algorithm was applied to obtain the pixel features and extract the pixel-based likelihood function, and the stochastic region merging algorithm was used to obtain regional features, and the likelihood function based on stochastic region merging was extracted. Then, the edge features of the image was obtained by the maximum gradient algorithm, and the edge-based likelihood function was extracted to restore the edge information lost in the stochastic region merging process. Furthermore, three kinds of likelihood functions were fused, and image segmentation was realized by the minimum energy criterion. Finally, the effectiveness of the algorithm was verified by comparative experiments with traditional image segmentation methods. The experimental results show that the improved algorithm can achieve accurate segmentation of stamping part images, and the application effect is better.

关 键 词：冲压件 视觉检测 马尔可夫随机场 随机区域合并 图像分割 似然函数 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TG84[自动化与计算机技术—计算机科学与技术]