1.School of Computer Science,Wuhan University,Wuhan 430072,Hubei,China
2.Shenzhen Institute of Wuhan University,Shenzhen 518000,Guangzhou,China
Show less
文章历史+
Received
Published
2019-01-04
2020-12-24
Issue Date
2026-07-23
PDF (6126K)
摘要
为了解图像分割领域的研究现状,对图像分割方法进行了系统性梳理,首先按照基于阈值、边缘、区域、聚类、图论及特定理论等6类方法介绍传统图像分割方法;然后介绍基于深度学习的分割方法,并探讨了几种常用的分割网络模型,包括全卷积网络(full convolutional network,FCN)、金字塔场景解析网络(pyramid scene parsing network,PSPNet)、DeepLab、Mask R-CNN;最后在图像分割的常用数据集上对同类方法进行了性能比较和分析。
Abstract
In order to understand the current research status in the field of image segmentation, the image segmentation methods are systematically sorted out. Firstly, traditional image segmentation methods are introduced according to 6 types of methods based on thresholds, edges, regions, clusters, graph theory, and specific theories. Then the segmentation methods based on deep learning are introduced, and several commonly used segmentation network models are discussed, including full convolutional network (FCN), pyramid scene parsing network (PSPNet), DeepLab, and Mask R-CNN. Finally, the performance comparison and analysis of similar methods are performed on the commonly used datasets for image segmentation.
最早的DeepLab模型[30]如图14所示,输入图像经过带有多孔(atrous)卷积层的DCNN处理后,得到粗略的评分图,双线性内插值上采样后引入全连接条件随机场(conditional random fields,CRF)[30]作为后处理,充分考虑全局信息,对目标边缘像素点进行更准确地分类,排除噪声干扰,从而提升分割精度。
DeepLab-v3[32]如图15,输入图像经过卷积池化后,图像尺寸缩小了4倍,再依次经过3个Block模块[39](Block1~ Block3)进行卷积、线性整流函数(rectified linear unit, ReLU)、池化处理,图像依次缩小8、16、16倍,然后经过Block4处理后进入ASPP模块,ASPP通过融合不同多孔卷积(插孔数rate=6、12、18)处理后,与1×1卷积层、全局池化层进行整合,得到缩小16倍的特征图,再进行分类预测得到分割图。DeepLab-v3+模型[33]采取编解码结构,如图16,将DeepLab-v3模型作为编码部分,对图像进行处理后输出DCNN中浅层特征图和经过ASPP融合卷积后的特征图,并将两者作为解码部分的输入。进入解码模块,先对输入的浅层特征图卷积,再与经过上采样的ASPP特征图进行融合,然后经过卷积、上采样操作输出原始尺寸大小的分割图,实现端到端的语义分割。DeepLab-v3+模型对不同目标(人、马、狗等)的分割效果如图17所示,其中第1、3列表示输入图像,2、4列表示对应的分割结果。由图17可知,分割后的图像中能够明显区分出前景目标和背景,目标边缘轮廓清晰,说明该模型能够实现细粒度的分割。
3.4 Mask R-CNN
Mask R-CNN[34]是He等基于Faster R-CNN[40]提出的用于图像分割的深度卷积网络,在进行目标检测的同时实现高质量的分割。Mask R-CNN框架如图18,第一阶段,首先用区域建议网络(region propo-sal networks,RPN)[40]提取出候选目标的边界框,然后对边界框里面的内容(regions of interest,RoI)进行RoIAlign[34]处理,将RoI划分为mm的子区域;第二阶段,与预测类和边界框回归任务并行,增加了为每个RoI输出二分类掩码的分支,可理解为用FCN对每个RoI进行分割,以像素到像素的方式预测分割掩码。
OTSUN. A threshold selection method from gray-level histograms [J]. IEEE Transactions on Systems, Man, and Cybernetics, 1979, 9(1):62-66. DOI: 10.1109/TSMC.1979.4310076 .
[2]
PUN T. A new method for grey-level picture thresholding using the entropy of the histogram [J]. Signal Processing, 1980,2(3):223-237. DOI: 10.1016/0165-1684(80)90020-1 .
[3]
YEN J C, CHANGF J, CHANGS. A new criterion for automatic multilevel thresholding [J]. IEEE Transactions on Image Processing, 1995, 4(3):370-378.
[4]
KHANJ F, BHUIYANS M A, ADHAMIR R. Image segmentation and shape analysis for road-sign detection[J]. IEEE Transactions on Intelligent Transportation Systems, 2011, 12(1):83-96. DOI: 10.1109/ TITS.2010.2073466 .
[5]
ROSENFELDA. The max Roberts operator is a Hueckel-type edge detector[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 1981, 3(1):101-103. DOI: 10.1109/TPAMI.1981.4767056 .
[6]
LANGY, ZHENGD. An improved Sobel edge detection operator [C]// IEEE International Conference on Computer Science & Information Technology. New York: IEEE Press, 2010: 67-71. DOI: 10.1109/ ICCSIT.2010.5563693 .
[7]
YANGL, WUX Y, ZHAOD W, et al. An improved Prewitt algorithm for edge detection based on noised image[C]// 2011 4th International Congress on Image and Signal Processing. New York: IEEE Press, 2011:1197-1200.
LIE S, ZHUS L, ZHUB S, et al. An adaptive edge-detection method based on the Canny operator [C]//2009 International Conference on Environmental Science and Information Application Technology. New York: IEEE Press, 2009: 465-469. DOI: 10.1109/ESIAT.2009.49 .
[10]
ZHANGY J. An Overview of Image and Video Segmentation in the Last 40 Years [EB/OL]. [2018-02-10]. isxn=97815914 07539.
[11]
PHAMD L, XUC Y, PRINCEJ L. A survey of current methods in medical image segmentation [J]. Annual Review of Biomedical Engineering, 2000,2(1):315-337.
[12]
TREMEAUA, BORELN. A region growing and merging algorithm to color segmentation [J]. Pattern Recognition, 1997, 30(7): 1191-1203. DOI: 10.1016/S0031-3203(96)00147-1 .
[13]
CHENGY .Mean shift,mode seeking,and clustering [J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 1995, 17(8): 790-799. DOI: 10.1109/34.400568 .
[14]
FUKUNAGAK, HOSTETLERL D .The estimation of the gradient of a density function,with applications in pattern recognition [J]. IEEE Transactions on Information Theory, 1975, 21(1): 32-40. DOI:10.1109/TIT.1975.1055330.
[15]
SHEIKHY A, KHANE A, KANADET .Mode-seeking by Medoidshifts [C]//2007 IEEE 11th International Conference on Computer Vision. New York : IEEE Press, 2007:1-8.DOI: 10.1109/ ICCV.2007.4408978 .
[16]
LEVINSHTEINA, STEREA, KUTULAKOSK N,et al .TurboPixels: Fast superpixels using geometric flows [J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2009, 31(12): 2290-2297. DOI: 10.1109/TPAMI.2009.96 .
[17]
RACHANTA, SHAJIA, SMITHK,et al .SLIC superpixels compared to state-of-the-art superpixel methods [J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2012, 34(11): 2274-2282. DOI:10.1109/TPAMI.2012.120.
[18]
BOYKOVY Y, JOLLYM P. Interactive graph cuts for optimal boundary region segmentation of objects in N-D images [C]//Proceedings Eighth IEEE International Conference on Computer Vision. New York: IEEE, 2001: 105-112. DOI: 10.1109/ICCV.2001.937505 .
[19]
ROTHERC, KOLMOGOROVV, BLAKEA, et al. “GrabCut”: Interactive foreground extraction using iterated graph cuts [J]. International Conference on Computer Graphics and Interactive Techniques, 2004, 23(3): 309-314. DOI: 10.1145/1015706.1015720 .
[20]
TANGM, GORELICKL, VEKSLERO, et al. GrabCut in One Cut [C]//Proceedings of the 2013 IEEE International Conference on Computer Vision. New York: IEEE Press, 2013: 1769-1776.
[21]
MATHERONG. Random Sets and Integral Geometry[M]. New York: Wiley Press,1975.
[22]
CRESPOJ, W.SCHAFERR, SERRAJ, et al. The flat zone approach: A general low-level region merging segmentation method [J]. Signal Processing, 1997, 62(1): 37-60.
[23]
HOLLANDJ H. Genetic algorithms and the optimal allocation of trials [J]. SIAM Journal on Computing, 1973, 2(2): 88-105. DOI: 10.1137/0202009 .
[24]
LIUH H, CHENZ H, CHENX H, et al. Multiresolution medical image segmentation based on wavelet transform [C]// 2005 IEEE Engineering in Medicine and Biology 27th Annual Conference. New York : IEEE Press, 2006: 3418-3421. DOI: 10.1109/IEMBS.2005.1617212 .
[25]
YANGX, CHUNGA C S, JIANY. An active contour model for image segmentation based on elastic interaction[J]. Journal of Computational Physics, 2006, 219(1): 455-476. DOI: 10.1016/j.jcp.2006.03.026 .
[26]
SAHAP K, UDUPAJ K. Relative fuzzy connectedness among multiple objects: Theory, algorithms, and applications in image segmentation [J]. Computer Vision and Image Understanding, 2001, 82(1): 42-56. DOI: 10.1006/cviu.2000.0902 .
[27]
ROYP, GOSWAMIS, CHAKRABORTYS, et al. Image segmentation using rough set theory: A review [J]. International Journal of Rough Sets & Data Analysis, 2014, 1(2):62-74. DOI: 10.1006/cviu.2000.0902 .
[28]
LONGJ, SHELHAMERE, DARRELLT. Fully convolutional networks for semantic segmentation[C]//Proceedings of the 2015 IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2015: 3431-3440. DOI: 10.1109/CVPR.2015.7298965 .
[29]
ZHAOH S, SHIJ P, QIX J, et al. Pyramid scene parsing network [C]//Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2017: 2881-2890. DOI: 10.1109/CVPR.2017.660 .
[30]
CHENL C, PAPANDREOUG, KOKKINOSI, et al. Semantic Image Segmentation with Deep Convolutional Nets and Fully Connected CRFs [EB/OL]. [2018-08-09].
[31]
CHENL C, PAPANDREOUG, KOKKINOSI, et al. Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected CRFs [J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2018, 40(4): 834-848. DOI:10.1109/TPAMI.2017.2699184 .
[32]
CHENL C, PAPANDREOUG, SCHROFFF, et al. Rethinking Atrous Convolution for Semantic Image Segmentation [EB/OL]. [2018-05-09].
[33]
CHENL C, ZHUY, PAPANDREOUG, et al. Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation [EB/OL]. [2018-08-09].
[34]
HEK M, GKIOXARIG, DOLLÁRP, et al. Mask R-CNN [C]//2017 IEEE International Conference on Computer Vision (ICCV). New York: IEEE Press, 2017: 2980-2988. DOI: 10.1109/ICCV.2017.322 .
[35]
EVERINGHAMM, ESLAMIS M A, GOOL LVAN, et al. The PASCAL visual object classes challenge: A retrospective [J]. International Journal of Computer Vision, 2015, 111(1): 98-136. DOI: 10.1007/s11263-014-0733-5 .
[36]
LINT Y, MAIREM, BELONGIES, et al. Microsoft COCO: Common objects in context [C]//European Conference on Computer Vision (LNCS 8639). Berlin: Springer, 2014: 740-755. DOI: 10.1007/978-3-319-10602-1_48 .
[37]
CHENY P, LIJ N, XIAOH X, et al. Dual path networks [C]//Proceedings of the 31st International Conference on Neural Information Processing Systems. New York: ACM, 2017: 4467-4475.
[38]
CHENGJ, SUNY, MENGQ H. A dense semantic mapping system based on CRF-RNN network[C]//2017 18th International Conference on Advanced Robotics (ICAR). New York: IEEE Press, 2017: 589-594. DOI: 10.1109/ICAR.2017.8023671 .
[39]
HEK M, ZHANGX Y, RENS Q, et al. Deep residual learning for image recognition [C]// 2016 IEEE Conference on Computer Vision & Pattern Recognition. Washington D C: IEEE Computer Society, 2016:770-778.
[40]
RENS Q, HEK M, GIRSHICKR, et al. Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks[EB/OL]. [2018-08-08].
[41]
LIY, QIH Z, DAIJ, et al. Fully convolutional instance-aware semantic segmentation[C]//Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition. New York : IEEE Press, 2017: 2359-2367.
[42]
DAIJ, HEK, SUNJ. Instance-aware semantic segmentation via multi-task network cascades [C]//Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2016: 3150-3158.
[43]
CORDTSM, OMRANM, RAMOSS, et al. The Cityscapes dataset for semantic urban scene understanding[C]//Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition. New York : IEEE Press, 2016: 3213-3223.