结合多尺度及密集特征图融合的阴影检测方法

doi:10.3969/j.issn.1000-1158.2021.05.05

摘要
图/表
参考文献(18)
相关文章 (15)

全文: PDF (2941 KB) HTML (1 KB)
输出: BibTeX | EndNote (RIS)

摘要为了提高图像中阴影检测的准确性,提出一种利用深度神经网络实现阴影检测的方法。首先,构造了一种密集特征图融合结构,将不同卷积层产生的特征图进行融合;其次,针对图像中阴影的多种尺度特征,设计了一种串并联结合的扩张卷积结构提取图像中阴影多尺度特征;最后,将串并联结合的扩张卷积结构和密集特征图融合结构进行结合,设计出一种端到端的Dilated Dense Fusion-Unet网络实现阴影检测功能。实验结果表明,所提方法在SBU和UCF阴影检测数据集上的阴影检测结果及量化评估均优于已有代表性的阴影检测方法,在2个数据集上的准确率分别提高5.8%和6.5%,平衡误差率分别降低2.2%和0.5%。

	服务

	把本文推荐给朋友
	加入我的书架
	加入引用管理器
	E-mail Alert
	RSS
	作者相关文章
	张世辉
	张笑维
	李贺
	张笑笑
	牛景春
	陈琦

关键词 ：计量学, 图像处理, 阴影检测, 端到端, 多尺度特征, 扩张卷积, 密集特征图

Abstract：In order to improve the accuracy of shadow detection in the image, a shadow detection method utilizing deep neural network is proposed. Firstly, a dense feature map fusion structure is proposed to fuse the feature maps generated by different convolutional layers. Secondly, a serial-parallel dilated convolution structure is designed to extract the multi-scale feature in the original image aiming to the scale variant phenomena in shadow detection task. Finally, combining the dense feature map fusion structure and serial-parallel dilated convolution structure, an end-to-end dilated dense fusion-unet is constructed to detect shadow. Experimental results demonstrate that the shadow detection results and quantitative evaluation of the proposed method on the SBU and UCF shadow detection datasets outperform the existing representative shadow detection methods, the accuracy on the two datasets increased by 5.8% and 6.5%, and the balance error rate decreased by 2.2% and 0.5%, respectively. The ablation study verifies the structure rationality of the proposed dilated dense fusion-unet.

Key words： metrology image processing shadow detection end-to-end multi-scale feature dilated convolution dense feature map

收稿日期: 2019-07-01 发布日期: 2021-05-24

PACS:

TB96

基金资助:国家自然科学基金(61379065);河北省自然科学基金(F2019203285)

作者简介: 张世辉(1973-),男,河北赞皇人,燕山大学教授,博士生导师,主要从事视觉信息处理、模式识别方面的研究。Email: sshhzz@ysu.edu.cn

引用本文:

张世辉,张笑维,李贺,张笑笑,牛景春,陈琦. 结合多尺度及密集特征图融合的阴影检测方法[J]. 计量学报, 2021, 42(5): 570-576.
ZHANG Shi-hui,ZHANG Xiao-wei,LI He,ZHANG Xiao-xiao,NIU Jing-chun,CHEN Qi. Shadow Detection Method Combining Multi-Scale and Dense Feature Map Fusion. Acta Metrologica Sinica, 2021, 42(5): 570-576.

链接本文:

http://jlxb.china-csm.org:81/Jwk_jlxb/CN/10.3969/j.issn.1000-1158.2021.05.05 或 http://jlxb.china-csm.org:81/Jwk_jlxb/CN/Y2021/V42/I5/570

［1］Okabe T, Sato I, Sato Y. Attached shadow coding: Estimating surface normals from shadows under unknown reflectance and lighting conditions［C］// International Conference on Computer Vision. Kyoto, Japan, 2009: 1693-1700.
［2］Jean-Franois L, Efros A A, Narasimhan S G. Estimating natural illumination from a single outdoor image［C］// International Conference on Computer Vision. Kyoto, Japan, 2009: 183-190.
［3］毛翠丽, 卢荣胜, 董敬涛, 等. 相移条纹投影三维形貌测量技术综述［J］. 计量学报, 2018, 39(5): 628-640.
Mao C L, Lu R S, Dong J T, et al. Overview of the 3D profilometry of phase shifting fringe projection［J］. Acta Metrologica Sinica, 2018, 39(5): 628-640.
［4］Panagopoulos A, Samaras D, Paragios N. Robust shadow and illumination estimation using a mixture model［C］// Computer Vision and Pattern Recognition. Miami, Florida, United States, 2009: 651-657.
［5］Panagopoulos A, Wang C, Samaras D, et al. Simultaneous cast shadows, illumination and geometry inference using hypergraphs［J］. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2013, 35(2): 437-449.
［6］范亚兵, 黄桂平, 姚思远, 等. 三目立体摄影测量系统中相机标定技术研究［J］. 计量学报, 2015, 36(2): 132-135.
Fan Y B, Huang G P, Yao S Y, et al. Study on the techniques of camera calibration in the trinocular stereo photogrammetry system［J］. Acta Metrologica Sinica, 2015, 36(2): 132-135.
［7］Junejo I N. Foroosh H. Estimating geo-temporal location of stationary cameras using shadow trajectories［C］// European Conference on Computer Vision. Marseille, France, 2008: 318-331.
［8］Karsch K, Hedau V, Forsyth D. Rendering synthetic objects into legacy photographs［C］// ACM SIGGRAPH Asia. Hong Kong, China, 2011.
［9］Zhu J, Samuel K G G, Masood S Z, et al. Learning to recognize shadows in monochromatic natural images［C］// Computer Vision and Pattern Recognition. San Francisco, California, United States, 2010: 223-230.
［10］Jean-Franois L, Efros A A, Narasimhan S G. Detecting ground shadows in outdoor consumer photographs［C］// European Conference on Computer Vision. Heraklion, Crete, Greece, 2010: 322-335.
［11］Guo R, Dai Q, Hoiem D. Single-image shadow detection and removal using paired regions［C］// Computer Vision and Pattern Recognition, Colorado Springs, Coloradoa, United States, 2011, 2033-2040.
［12］Vicente T F Y, Hoai M, Samaras D. Leave-one-out kernel optimization for shadow detection and removal［J］. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2018, 40(3): 682-695.
［13］Khan S H, Bennamoun M, Sohel F, et al. Automatic feature learning for robust shadow detection［C］// Computer Vision and Pattern Recognition. Columbus, Ohio, United States, 2014: 1939-1946.
［14］Khan S H, Bennamoun M, Sohel F, et al. Automatic shadow detection and removal from a single image［J］. IEEE Transactions on Pattern Analysis and Machine Intelligence. 2016, 38(3): 431-446.
［15］Vicente T F Y, Hou L, Yu C P. Large-Scale training of shadow detectors with noisily-annotated shadow examples［C］// European Conference on Computer Vision. Amsterdam, Netherlands, 2016: 816-832.
［16］Hosseinzadeh S, Shakeri M, Zhang H. Fast shadow detection from a single image using a patched convolutional neural network ［C］// International Conference on Intelligent Robots and Systems. Madrid, Spain, 2018: 3124-3129.
［17］Nguyen V, Vicente T F Y, Zhao M. Shadow detection with conditional generative adversarial networks［C］// International Conference on Computer Vision. Venice, Italy, 2017: 4520-4528.
［18］Ronneberger O, Fischer P, Brox T. U-Net: Convolutional networks for biomedical image segmentation［C］// Medical Image Computing and Computer-Assisted Intervention. Munich, Free State of Bavaria, Germany, 2015: 234-241.