基于混合分割的图像边缘提取算法研究  

Research on Image Edge Extraction Algorithm Based on Hybrid Segmentation

作　　者：叶勇健[1] 朱荣钊 YE Yongjian;ZHU Rongzhao(Information Technology College,Xiamen Huatian International Vocation Institute,Xiamen 361102,China;School of Computer and Information Engineering,Hubei University,Wuhan 430062,China)

机构地区：[1]厦门华天涉外职业技术学院信息技术学院,福建厦门361102 [2]湖北大学计算机与信息工程学院,湖北武汉430062

出　　处：《安阳师范学院学报》2023年第5期20-25,共6页Journal of Anyang Normal University

基　　金：福建省教育厅教育科学研究课题(项目编号:JA15892)。

摘　　要：卷积神经网络在图像语义分割与边缘提取中得到广泛研究,但是在实际应用中存在传统识别器无法抹除域间差异所产生的误差问题。研究针对鉴别器网络性能对整体图像分割存在结果权重较高的问题,提出基于空洞卷积的域识别网络,并将其应用于图像边缘提取中。结果表明,所提出的模型可以在不增加额外训练参数的前提下明显增大感受视野,有效地提升图像分割与边缘提取性能,在GTA5与SYNTHINA公开数据上mIoU分别为44.1%和44.9%。Convolutional neural network is widely used in image semantic segmentation and edge extraction,but there is a problem that traditional recognizers cannot erase the errors caused by the differences between domains.A domain recognition network based on hollow convolution is proposed to address the issue of high result weights in overall image segmentation due to the performance of discriminator networks,and it is applied to image edge extraction.The results show that the proposed model can significantly increase the receptive field of view and improve the performance of image segmentation and edge extraction without adding additional training parameters,with mIoU of 44.1%and 44.9%on GTA5 and SYNTHINA public data,respectively.

关 键 词：分割与边缘提取 迁移学习 空洞卷积 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]