摘要
边缘检测是图像处理工作的关键步骤之一,目前边缘检测模型基于卷积神经网络(CNNs)搭建编码-解码网络。由于现有编码网络提取特征能力有限,且忽视了神经元之间复杂的信息流向,本文模拟视网膜、外侧膝状体(LGN)和腹侧通路(“what”通路)前端V1区、V2区、V4区的生物视觉机制,搭建全新的编码网络和解码网络。编码网络模拟视网膜-LGN-V1-V2的信息传递机制,充分提取图像中的特征信息;解码网络模拟V4区的信息整合功能,设计邻近融合网络以整合编码网络的特征预测,实现特征的充分融合。该神经网络模型在BSDS500数据集和NYUD-V2数据集上进行了实验。结果表明,本文搭建的编码-解码方法的F值(ODS)为0.820,相比于LRCNet提高了0.49%。
Edge detection is a key step in image processing.In recent years,edge detection has built an encoding-decoding network based on Convolutional Neural Networks(CNNs),and has achieved good results.Among them,the coding network is usually built based on classic networks such as VGG16,and researchers more focus on the design of the decoding network.Considering that the existing coding network has limited ability to extract features and ignores the complex information flow between neurons,this study simulates the biological vision mechanism of the retina,the lateral genicu‐late body(LGN),and the front end of the ventral pathway("what"pathway),including V1,V2,and V4,to build a new encoding network and decoding network.In this paper,the encoding network simulates the information transfer mechanism of the retina-LGN-V1-V2 to fully extract the feature information in the image;the decoding network simulates the information integration function of the V4 area,and the adjacent fusion module is designed to integrate the feature prediction of the encoding network to realize the full integration of feature information.This neural network model has performed a large number of experiments on the BSDS500 dataset and NYUD-V2 dataset,and the results have been significantly improved compared with competitors in recent years.Through comparative experiments,the F value(ODS)of the encoding-decoding method built in this paper is 0.820,which is about 0.49%higher than that of LRCNet.
作者
潘盛辉
王蕤兴
林川
PAN Shenghui;WANG Ruixing;LIN Chuan(School of Electrical,Electronic and Computer Science,Guangxi University of Science and Technology,Liuzhou 545006,China)
出处
《广西科技大学学报》
2022年第2期60-68,共9页
Journal of Guangxi University of Science and Technology
基金
国家自然科学基金项目(61866002)
广西自然科学基金项目(2020GXNSFDA297006,2018GXNSFAA138122,2015GXNSFAA139293)
广西科技大学研究生教育创新计划项目(GKYC202005)资助。
