仿视网膜采样的二进制描述子

doi:10.11959/j.issn.1000-436x.2019021

通信学报 ›› 2019, Vol. 40 ›› Issue (1): 15-23.doi: 10.11959/j.issn.1000-436x.2019021

仿视网膜采样的二进制描述子

袁庆升¹,靳国庆¹,张冬明¹,包秀国¹

¹ 国家计算机网络应急技术处理协调中心，北京 100029
² 中国科学院计算技术研究所，北京 100190
³ 中国科学院大学网络空间安全学院，北京 100049
⁴ 中国科学院信息工程研究所，北京 100193

修回日期:2018-12-01 出版日期:2019-01-01 发布日期:2019-02-03
作者简介:袁庆升（1980- ），男，山东济南人，中国科学院信息工程研究所博士生，国家计算机网络应急技术处理协调中心副高级工程师，主要研究方向为多媒体大数据处理、网络与信息安全。|靳国庆（1988- ），男，山东单县人，博士，中国科学院计算技术研究所助理研究员，主要研究方向为多媒体内容检索、模式识别等。|张冬明（1977- ），男，江苏盐城人，博士，国家计算机网络应急技术处理协调中心研究员、硕士生导师，主要研究方向为多媒体内容检索、模式识别、视频编码等。|包秀国（1963- ），男，江苏如皋人，博士，国家计算机网络应急技术处理协调中心教授级高级工程师、博士生导师，主要研究方向为网络与信息安全。
基金资助:
国家自然科学基金资助项目(61672495);国家自然科学基金资助项目(61273247);国家重点研发专项基金资助项目(2016YFB0801203);国家重点研发专项基金资助项目(2016YFB0801200)

Retina-imitation sampling based binary descriptor

Qingsheng YUAN¹,Guoqing JIN¹,Dongming ZHANG¹,Xiuguo BAO¹

¹ National Computer Network Emergency Response Technical Team/Coordination Center of China,Beijing 100029,China
² Institute of Computing Technology,Chinese Academy of Sciences,Beijing 100190,China
³ School of Cyber Security,University of Chinese Academy of Sciences,Beijing 100049,China
⁴ Institute of Information Engineering,Chinese Academy of Sciences,Beijing 100193,China

Revised:2018-12-01 Online:2019-01-01 Published:2019-02-03
Supported by:
The National Natural Science Foundation of China(61672495);The National Natural Science Foundation of China(61273247);The National Key Research and Development Program of China(2016YFB0801203);The National Key Research and Development Program of China(2016YFB0801200)

摘要/Abstract

摘要：

现有二进制描述子生成采用随机点对或均匀采样方式，顽健性弱、计算复杂。针对这一问题，提出了一种模仿人眼视网膜特性的采样模式（RBS），首先通过设计采样密度、多尺度光滑、视野重叠等采样方法来模仿视网膜神经节细胞层（ganglion cell layer），也称为视神经层，将光信号转换为视信息的方式，再通过对典型数据学习来选择特征点对，最后使用区块均值代替单像素点计算点对比较值，生成顽健的紧致二进制描述子。在Mikolajczyk提出的数据集上进行了实验，实验结果表明，128 bit的RBS-128相对于512 bit的FREAK和BRISK正确率分别提升16.4%和5.3%。

关键词: 二进制描述子, 仿视网膜采样, 神经节细胞, 比较点对

Abstract:

The existing binary descriptors，generated from random or uniform point pairs sampling，suffer from low robustness and high computation.A novel sampling method,named RBS (retina-imitation based sampling),was proposed， which combines different densities sampling,multi-scale smoothing and reception field overlapping to imitate the converting from light signal to vision of ganglion cells of human retina cells,and further selects most discriminative comparison pairs based on learning on training data.Finally，compact binary descriptor was generated based on comparisons between the neighbor mean instead of singe sampled point.The experimental results show the RBS-128 with 128 bit outperforms FREAK and BRSIK with 512 bit about 16.4% and 5.3% in precision on the dataset provided by Mikolajczyk.

Key words: binary descriptor, retina-imitation based sampling, ganglion cell, comparison pairs

中图分类号:

TP37

袁庆升,靳国庆,张冬明,包秀国. 仿视网膜采样的二进制描述子[J]. 通信学报, 2019, 40(1): 15-23.

Qingsheng YUAN,Guoqing JIN,Dongming ZHANG,Xiuguo BAO. Retina-imitation sampling based binary descriptor[J]. Journal on Communications, 2019, 40(1): 15-23.

图/表 8

图1

图2

图3

图4

图5

图6

表1

不同描述子的正确率对比"

图像	变换	SIFT	ORB	BRISK	BRIEF	FREAK	RBS-32	RBS-64	RBS-128	RBS-160
	1-2	85.3%	64.1%	65.9%	56.9%	58.4%	62.3%	72.5%	79.7%	74.7%
	1-3	84.5%	63.1%	64.9%	58.6%	58.6%	60.4%	71.3%	79.7%	72.7%
wall	1-4	77.4%	53.2%	55.9%%	49.3%	48.5%	47.5%	63.1%	71.4%	64.9%
	1-5	66.3%	44.0%	46.7%%	36.7%	37.6%	41.1%	50.6%	59.7%	53.1%
	1-6	45.1%	30.1%	34.0%	23.0%	25.5%	30.1%	32.0%	33.2%	32.4%
	平均	71.7%	50.9%	53.5%	44.9%	45.7%	48.3%	57.9%	64.7%	59.5%
	1-2	77.2%	59.5%	60.5%	54.9%	57.6%	60.3%	62.4%	66.9%	62.9%
	1-3	62.6%	49.3%	50.4%	45.6%	48.8%	50.7%	53.1%	55.2%	54.0%
graffiti	1-4	45.3%	36.5%	36.5%	36.5%	33.9%	32.0%	34.7%	37.6%	35.4%
	1-5	39.5%	27.5%	29.6%	25.6%	24.8%	24.0%	22.9%	24.0%	22.6%
	1-6	22.2%	15.7%	16.8%	13.9%	10.4%	9.3%	12.3%	12.0%	11.8%
	平均	49.4%	37.7%	38.8%	35.3%	35.1%	35.3%	37.1%	39.1%	37.3%
	1-2	94.9%	93.6%	93.3%	91.9%	90.9%	90.5%	91.6%	93.2%	91.4%
	1-3	92.7%	83.8%	82.9%	81.7%	79.1%	75.9%	78.3%	81.7%	77.6%
ubc	1-4	89.2%	77.7%	78.6%	72.8%	71.9%	73.6%	76.8%	79.1%	76.6%
	1-5	85.8%	73.6%	77.4%	74.8%	73.6%	72.5%	73.0%	75.4%	73.7%
	1-6	83.9%	71.9%	74.5%	70.7%	68.1%	69.9%	71.0%	72.2%	70.7%
	平均	89.3%	80.1%	81.3%	78.4%	76.7%	76.5%	78.2%	80.3%	78.0%
	1-2	74.6%	57.6%	59.7%	52.0%	53.1%	58.7%	61.6%	65.1%	63.9%
	1-3	61.4%	51.5%	49.6%	44.0%	45.6%	48.5%	50.7%	52.3%	51.1%
boat	1-4	42.4%	33.1%	34.9%	36.0%	32.0%	29.3%	32.3%	33.9%	32.4%
	1-5	34.5%	24.8%	26.7%	23.2%	21.3%	20.3%	22.4%	22.9%	22.6%
	1-6	19.3%	14.1%	14.7%	11.2%	10.1%	7.5%	11.5%	11.5%	11.8%
	平均	46.4%	36.2%	37.1%	33.3%	32.4%	32.9%	35.7%	37.1%	36.4%
	1-2	91.5%	74.9%	77.9%	73.3%	71.2%	74.1%	77.6%	82.4%	76.6%
	1-3	90.1%	73.3%	76.0%	68.5%	67.7%	70.4%	75.7%	81.3%	76.6%
tree	1-4	85.1%	68.8%	71.2%	64.0%	64.8%	66.9%	72.3%	77.6%	74.7%
	1-5	79.5%	62.4%	64.8%	57.6%	59.7%	61.3%	66.7%	68.8%	66.8%
	1-6	69.0%	53.1%	56.0%	48.8%	50.4%	52.3%	54.9%	57.1%	54.0%
	平均	83.0%	66.5%	69.2%	62.5%	62.8%	65.0%	69.4%	73.4%	69.8%
	1-2	93.3%	82.7%	84.6%	78.7%	78.2%	79.8%	84.3%	87.0%	83.5%
	1-3	90.4%	77.7%	79.8%	73.1%	73.9%	75.0%	77.4%	81.4%	76.6%
leuven	1-4	87.2%	73.1%	74.2%	69.7%	68.9%	70.7%	73.9%	76.9%	73.7%
	1-5	84.6%	69.9%	71.5%	67.6%	66.2%	67.8%	71.8%	74.5%	72.7%
	1-6	81.2%	66.2%	66.0%	64.6%	61.4%	64.1%	66.0%	68.6%	64.9%
	平均	87.4%	73.9%	75.2%	70.7%	69.7%	71.5%	74.7%	77.7%	74.3%
平均	—	71.2%	57.6%	59.2%	54.2%	53.7%	54.9%	58.8%	62.1%	59.2%

表1

表2

参考文献 20

[1]	CALONDER M , LEPETIT V , STRECHA C ,et al. Brief:binary robust independent elementary features[C]// European Conference on Computer Vision. 2010: 778-792.
[2]	RUBLEE E , RABAUD V , KONOLIGE K ,et al. ORB:an efficient alternative to SIFT or SURF[C]// International Conference on Computer Vision. 2011: 2564-2571.
[3]	LEUTENEGGER S , CHLI M , SIEGWART R Y . BRISK:binary robust invariant scalable keypoints[C]// International Conference on Computer Vision. 2011: 2548-2555.
[4]	ALAHI A , ORTIZ R , VANDERGHEYNST P . Freak:fast retina keypoint[C]// Computer Vision and Pattern Recognition. 2012: 510-517.
[5]	TOLA E , LEPETIT V , FUA P . Daisy:an efficient dense descriptor applied to wide-baseline stereo[J]. Pattern Analysis and Machine Intelligence, 2010,32(5): 815-830.
[6]	李兵, 刘磊波, 魏志强 ,等. 一种具有强实时性、强顽健性的图像匹配算法[J]. 软件学报, 2014,25(7): 1583-1592.
	LI B , LIU L , WEI Z Q . A strong Robust real-time image matching algorithm[J]. Journal of Software, 2014,25(7): 1583-1592.
[7]	朱英宏, 李俊山, 郭莉莎 ,等. 基于LBP的尺度不变特征的描述和匹配算法[J]. 计算机辅助设计与图形学学报, 2011,23(10): 1758-1763.
	ZHU Y H , LI J S , GUO L S ,et al. A local invariant feature description and matching algorithm based on LBP[J]. Journal of Computer-Aided Design ＆ Computer Graphics, 2011,23(10): 1758-1763.
[8]	卢鸿波, 孙愿, 张志敏 . 采用旋转匹配的二进制局部描述子[J]. 中国图象图形学报, 2013,18(10): 1315-1321.
	LU H B , SUN Y , ZHANG Z M . Binary local descriptor based on rotative matching[J]. Journal of Image and Graphics, 2013,18(10): 1315-1321.
[9]	惠国保, 李东波, 童一飞 . 挖掘图像补丁特征信息增强二进制描述子独特性[J]. 计算机辅助设计与图形学学报, 2014,26(9): 1419-1429.
	HUI G B , LI D B , TONG Y F . A discriminative binary descriptor built on further mining marginal information[J]. Journal of Computer-Aided Design ＆ Computer Graphics, 2014,26(9): 1419-1429.
[10]	FAN B , KONG Q , TRZCINSKI T ,et al. Receptive fields selection for binary feature description[J]. IEEE Transactions on Image Processing, 2014,23(6): 2583-2595.
[11]	TRZCINSKI T , CHRISTOUDIAS M , FUA P ,et al. Boosting binary keypoint descriptors[C]// Computer Vision and Pattern Recognition. 2013: 2874-2881.
[12]	KRIZHEVSKY A , SUTSKEVER I , HINTON G E . ImageNet classification with deep convolutional neural networks[C]// Neural Information Processing Systems, 2012: 1097-1105.
[13]	XIA R , PAN Y , LAI H ,et al. Supervised hashing for image retrieval via image representation learning[C]// The 28th AAAI Conference on Artificial Intelligence. 2014: 2156-2162.
[14]	LIN K , YANG H F , HSIAO J H ,et al. Deep learning of binary hash codes for fast image retrieval[C]// IEEE Conference on Computer Vision and Pattern Recognition Workshops. 2015: 27-35.
[15]	ZHAO F , HUANG Y , WANG L ,et al. Deep semantic ranking based hashing for multi-label image retrieval[C]// Computer Vision and Pattern Recognition. 2015: 1556-1564.
[16]	GARWAY-HEATH D F , CAPRIOLI J , FITZKE F W ,et al. Scaling the hill of vision:the physiological relationship between light sensitivity and ganglion cell numbers[J]. Investigative Ophthalmology and Visual Science, 2000,41(7): 1774-1782.
[17]	MIKOLAJCZYK K , SCHMID C . A performance evaluation of local descriptors[J]. Pattern Analysis and Machine Intelligence, 2005,27(10): 1615-1630.
[18]	ROSTEN E , DRUMMOND T . Machine learning for high-speed corner detection[C]// European Conference on Computer Vision. 2006: 430-443.
[19]	MAIR E , HAGER G D , BURSCHKA D ,et al. Adaptive and generic corner detection based on the accelerated segment test[C]// European Conference on Computer Vision. 2010: 183-196.
[20]	LOWE D G , . Object recognition from local scale-invariant features[C]// The Seventh IEEE International Conference on IEEE. 1999: 1150-1157.

仿视网膜采样的二进制描述子

Retina-imitation sampling based binary descriptor

在线阅读

PDF下载

可视化

摘要/Abstract

引用本文

使用本文

图/表 8

参考文献 20

相关文章 1

Metrics

推荐阅读 0

比较内容	描述子总个数	总时间/ms	平均时间/μs
SIFT	2 415	3145	1302
ORB	1 516	76	50
BRISK	1 752	79	45
BRIEF	1 345	65	48
FREAK	1 428	51	36
RBS-32	1 456	15	10
RBS-64	1 456	30	21
RBS-128	1 456	60	41