基于卷积神经网络的多尺度Logo检测算法

doi:10.11959/j.issn.2096-109x.2020026

摘要/Abstract

摘要：

针对自然场景图像中多尺度Logo的检测需求，提出了一种基于卷积神经网络的多尺度Logo检测算法。该算法基于两阶段目标检测的实现思路，通过构建特征金字塔并采取逐层预测的方式实现多尺度候选区域的生成，通过融合卷积神经网络中的多层特征图以增强特征的表达能力。在FlickrLogos-32数据集上的实验结果显示，相比基线方法，所提算法能够提升生成候选区域的召回率，并且在保证大中尺度 Logo 检测精度的前提下，提升小尺度Logo的检测性能，验证了所提算法的优越性。

关键词: Logo检测, 卷积神经网络, 多尺度, 区域生成网络, 特征融合

Abstract:

Aiming at the requirements for multi-scale Logo detection in natural scene images,a multi-scale Logo detection algorithm based on convolutional neural network was proposed.The algorithm was based on the realization of two-stage object detection.By constructing feature pyramids and adopting layer-by-layer prediction,multi-scale region proposals were generated.The multi-layer feature maps in convolutional neural networks were fused to enhance the feature representation.The experimental results on the FlickrLogos-32 dataset show that compared with the baseline,the proposed algorithm can improve the recall rate of region proposals,and can improve the performance of small Logo detection while ensuring the accuracy of large and middle Logo,proving the superiority of the proposed algorithm.

Key words: Logo detection, convolutional neural network, multi-scale, region proposal network, feature fusion

中图分类号:

TP391

江玉朝,吉立新,高超,李邵梅. 基于卷积神经网络的多尺度Logo检测算法[J]. 网络与信息安全学报, 2020, 6(2): 116-124.

Yuchao JIANG,Lixin JI,Chao GAO,Shaomei LI. Multi-scale Logo detection algorithm based on convolutional neural network[J]. Chinese Journal of Network and Information Security, 2020, 6(2): 116-124.

图/表 11

图1

表1

图2

图3

图4

表2

Flickrlogos-32中官方Logo以及所有Logo对象高宽比的近似统计　Table 2 Approximate statistics of the aspect ratios of the official Logo and all Logo objects in FlickrLogos-32"

目标数据集			高宽比
目标数据集	$\frac{1}{3}$	$\frac{1}{2}$	1	2	3
官方Logo（32）	6	6	20	0	0
FlickrLogos-32（3 404）	315	775	1 853	416	45

表2

图5

表3

图6

表4

图7

参考文献 27

[1]	SU H , GONG S , ZHU X . Scalable deep learning logo detection[C]// International Conference on Computer Vision Workshop. 2017: 270-279.
[2]	GAO Y , WANG F , LUAN H ,et al. Brand data gathering from live social media streams[C]// International Conference on Multimedia Retrieval. 2014:169.
[3]	PAN C , YAN Z , XU X ,et al. Vehicle logo recognition based on deep learning architecture in video surveillance for intelligent traffic system[C]// IET International Conference on Smart and Sustainable City. 2013: 123-126.
[4]	ZHANG S , WEN L , BIAN X ,et al. Single-shot refinement neural network for object detection[C]// IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 2018.
[5]	REN S , HE K , GIRSHICK R ,et al. Faster R-CNN:towards real-time object detection with region proposal networks[C]// International Conference on Neural Information Processing Systems. 2015.
[6]	LIU W , ANGUELOV D , ERHAN D ,et al. SSD:single shot multibox detector[C]// European Conference on Computer Vision. 2016.
[7]	LUO W , LI Y , URTASUN R ,et al. Understanding the effective receptive field in deep convolutional neural networks[C]// Conference and Workshop on Neural Information Processing Systems. 2016.
[8]	ZHOU B , KHOSLA A , LAPEDRIZA A ,et al. Object detectors emerge in deep scene CNNs[J]. Computer Science, 2014
[9]	ROMBERG S , PUEYO L G , LIENHART R ,et al. Scalable logo recognition in real-world images[C]// ACM International Conference on Multimedia Retrieval. 2011:25.
[10]	HANG S , ZHU X , GONG S . Deep learning logo detection with data expansion by synthesising context[C]// Applications of Computer Vision. 2017.
[11]	HU P , RAMANAN D . Finding tiny faces[C]// Computer Vision ＆Pattern Recognition. 2017.
[12]	EGGERT C , WINSCHEL A , ZECHA D ,et al. Saliency-guided selective magnification for company logo detection[C]// IEEE 23rd International Conference on Pattern Recognition (ICPR). 2016.
[13]	MENG Z , FAN X , XIN C ,et al. Detecting small signs from large images[C]// IEEE International Conference on Information Reuse and Integration. 2017.
[14]	ZHANG S , ZHU X , LEI Z ,et al. S3FD:single shot scale-invariant face detector[C]// International Conference on Computer Vision. 2017.
[15]	CAO G , XIE X , YANG W ,et al. Feature-fused SSD:fast detection for small objects[C]// IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 2017.
[16]	LIN T Y , DOLLáR , PIOTR , GIRSHICK R ,et al. Feature pyramid networks for object detection[C]// IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 2017.
[17]	IANDOLA F N , SHEN A , GAO P ,et al. DeepLogo:hitting logo recognition with the deep neural network hammer[J]. arXiv preprint arXiv:1510.02131, 2015
[18]	GIRSHICK R , . Fast R-CNN[C]// International Conference on Computer Vision. 2015.
[19]	HOI S C H , WU X , LIU H ,et al. LOGO-Net:large-scale deep logo detection and brand recognition with deep region-based convolutional networks[J]. IEEE Transactions on Pattern Analysis ＆ Machine Intelligence, 2015,46(5): 2403-2412.
[20]	EGGERT C , WINSCHEL A , LIENHART R . On the benefit of synthetic data for company logo detection[C]// The 23rd ACM International Conference on Multimedia. 2015.
[21]	EGGERT C , ZECHA D , BREHM S ,et al. Improving small object proposals for company logo detection[J]. arXiv preprint arXiv:1704.08881, 2017
[22]	GIRSHICKR , DONAHUE J , DARRELL T ,et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]// 2014 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 2014.
[23]	KONG T , YAO A , CHEN Y ,et al. HyperNet:towards accurate region proposal generation and joint object detection[C]// 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 2016.
[24]	LIU W , RABINOVICH A , BERG A C . ParseNet:looking wider to see better[J]. Computer Science, 2015
[25]	LONG J , SHELHAMER E , DARRELL T . Fully convolutional networks for semantic segmentation[J]. IEEE Transactions on Pattern Analysis ＆ Machine Intelligence, 2014,39(4): 640-651.
[26]	HE K , GKIOXARI G , DOLLAR P ,et al. Mask R-CNN[J]. IEEE Transactions on Pattern Analysis ＆ Machine Intelligence, 2017,(99):1.
[27]	OLIVEIRA , GON?ALO , FRAZ?O ,et al. Automatic graphic logo detection via fast region-based convolutional networks[C]// International Joint Conference on Neural Networks. 2016: 985-991.

预测层	步长	理论感受野	锚点尺度	锚点高宽比
Conv4_3	8	108	32	1/3
Conv5_3	16	228	64	1/2
Conv_fc7	32	340	128	1
Conv6_2	64	564	256	2
Conv7_2	128	724	512

	Adidas	Aldi	Apple	Becks	BMW	Carls	Chim	Coke
算法	Corona	DHL	Erdi	Esso	Fedex	Ferra	Ford	Fost	mAP
	Google	Guin	Hein	HP	Milka	Nvid	Paul	Pepsi
	Ritt	Shell	Sing	Starb	Stel	Texa	Tsin	Ups
	78.3	88.7	80.7	86.1	84.2	77.4	83.1	76.0
本文算法	97.6	80.6	90.3	94.9	83.2	94.0	92.0	90.4	85.7
	99.7	87.5	79.1	76.0	66.8	69.1	94.1	79.6
	86.2	71.0	95.3	99.5	87.4	93.7	86.0	93.7
	68.1	79.1	84.5	72.3	86.4	68.0	78.0	73.3
基线方法	90.9	77.4	90.9	88.6	71.1	91.0	98.3	86.2	81.1
Faster R-CNN^[10]	98.0	90.7	81.3	67.0	54.5	64.0	90.9	59.6
	81.0	57.3	97.9	99.5	86.7	90.4	87.5	85.8
	61.6	67.2	84.9	72.5	70.0	49.6	71.9	33.0
DeepLogo^[17]	92.9	53.5	80.1	88.8	61.3	90.0	84.2	79.7	74.4
	85.2	89.4	57.8	58.4	34.6	50.3	98.6	34.2
	63.0	57.4	94.2	95.9	82.2	87.4	84.3	81.5
	—	—	—	—	—	—	—	—
BD-FRCN-M^[27]	—	—	—	—	—	—	—	—	73.5
	—	—	—	—	—	—	—	—
	—	—	—	—	—	—	—	—

目标检测网络的输入特征层	检测结果mAP
Conv3_3+Conv4_3+Conv5_3	85.7
Conv3_3	83.2
Conv4_3	84.5
Conv5_3	84.1