摘要
引入三支决策,通过分步约简的方式,改进纹身图像检索算法。将待检索图像及图像库中所有图像灰度化,比较图像库中各图像与待检索图像对应像素点灰度值的差异,统计差异值小于某阈值的像素点在各图像中所占比例,据此以三支决策从图像库中筛选出相似图像并剔除不相似图像;提取其余图像的尺度不变特征变换(scale invariant feature transform, SIFT)特征并与待检索图像进行特征点匹配,再次利用三支决策选出匹配正确的特征点并剔除匹配错误的特征点,提取其余特征点邻域色调-饱和度-亮度(hue saturation value, HSV)空间的颜色特征,结合SIFT特征再次进行特征点匹配,由匹配点的多少确定相似性大小,进而实现检索。在包含3 579幅纹身图像的自建图像库中所进行的检索实验结果显示,改进算法比尺度不变特征变换算法和融合局部颜色特征算法的查准率和查全率皆有所提升,且平均检索时间更短。
The method of three-way decision is introduced to improve the algorithm of tattoo image retrieval through step-by-step reduction. After all images are convert into grayscale, the gray value difference of the corresponding pixels of all images in the image library from the image to be retrieved is compared, and the proportion of the pixels whose difference is less than a certain threshold in each image is figured out, on which the three-way decision is used for the first time to select or eliminate some images from the image library. The scale invariant feature transform(SIFT) features of the other images are extracted, Euclidean distance is used to match the feature points, and the three-way decision is used for the second time to select or eliminate some feature points. As for the remaining ones, the color features in their neighborhood hue saturation value(HSV) space are extracted, and the feature points are matched agan by combining the SIFT feature. By the number of matching points, the size of similarity is determined, and then the retrieval is realized. The retrieval experiment results in a self-built image library containing 357 9 tattoo images show that, the accuracy and recall of the improved algorithm are higher than those of the scale invariant feature transform algorithm and the fusion local color feature algorithm, and the average retrieval time of the improved algorithm is shorter.
引文
[1] NGAN M,QUINN G W,GROTHER P.Tattoo Recognition Technology-Challenge (Tatt-C):Outcomes and Recommendations[R/OL].(2016-09-06)[2018-10-01].http://dx.doi.org/10.6028/NIST.IR.8078.
[2] JAIN A K,LEE J E,JIN R.Tattoo-ID:Automatic Tattoo Image Retrieval for Suspect and Victim Identification[C/OL]//IP H HS,AU O C,LEUNG H,et al.Advances in Multimedia Information Processing - PCM 2007:Lecture Notes in Computer Science (vol 4810).Berlin:Springer,2007:256-265[2018-10-01].http://dx.doi.org/10.1007/978-3-540-77255-2_28.
[3] ACTON S T,ROSSI A.Matching and Retrieval of Tattoo Images:Active Contour CBIR and Glocal Image Features[C/OL]//2008 IEEE Southwest Symposium on Image Analysis and Interpretation.USA NM Santa Fe:IEEE,2008:21-24[2018-10-01].http://dx.doi.org/10.1109/SSIAI.2008.4512275.
[4] LEE J E,JIN R,JAIN A K.Image Retrieval in Forensics:Tattoo Image Database Application[J/OL].IEEE Multimedia,2012,19(1):40-49[2018-10-01].http://dx.doi.org/10.1109/MMUL.2011.59.
[5] LEE J E,JIN R,JAIN A K.Rank-based distance metric learning:An application to image retrieval[C/OL]//2008 IEEE Conference on Computer Vision and Pattern Recognition.USA AK Anchorage:IEEEIEEE,2008:1-8[2018-10-01].http://dx.doi.org/10.1109/CVPR.2008.4587389.
[6] 兰蓉,贾世英.基于纹理与颜色特征融合的刑侦图像检索算法[J/OL].西安邮电大学学报,2016,21(2):57-62[2018-10-01].http://dx.doi.org/10.13682/j.issn.2095-6533.2016.02.011.
[7] LOWE D G.Object recognition from local scale-invariant features[C/OL]//Proceedings of the Seventh IEEE International Conference on Computer Vision.Kerkyra:IEEE,1999:1150-1157[2018-10-01].http://dx.doi.org/10.1109/ICCV.1999.790410.
[8] LOWE D G.Distinctive Image Features from Scale-Invariant Keypoints[J/OL].International Journal of Computer Vision.2004,60(2):91-110[2018-10-01].https://link.spring er.com/article/10.1023/B:VISI.0000029664.99615.94.
[9] LEE J E,JAIN A K,JIN R.Scars,marks and tattoos(SMT):Soft biometric for suspect and victim identification[C/OL]//2008 Biometrics Symposium.USA FL Tampa:IEEE,2008:1-8[2018-10-01].http://dx.doi.org/10.1109/BSYM.2008.4655515.
[10] JAIN A K,LEE J E,JIN R,et al.Content-based image retrieval:an application to tattoo images[C/OL]//2009 16th IEEE International Conference on Image Processing (ICIP).Egypt Cairo:IEEE,2009:2709-2712[2018-10-01].http://dx.doi.org/10.1109/ICIP.2009.5414140.
[11] KIM J,PARRA A,YUE J,et al.Robust local and global shape context for tattoo image matching[C/OL]//2015 IEEE International Conference on Image Processing (ICIP).QC Quebec City:IEEE,2015:2194-2198[2018-10-01].http://dx.doi.org/10.1109/ICIP.2015.7351190.
[12] MANGER D.Large-Scale Tattoo Image Retrieval[C/OL]//2012 Ninth Conference on Computer and Robot Vision.ON Toronto:IEEE,2012:454-459[2018-10-01].http://dx.doi.org/10.1109/CRV.2012.67.
[13] KIM J,LI H,YUE J,DELP E J.Tattoo image retrieval for region of interest[C/OL]//2016 IEEE Symposium on Technologies for Homeland Security (HST).USA MA Waltham:IEEE,2016:1-6[2018-10-01].http://dx.doi.org/10.1109/THS.2016.7568954.
[14] 刘盾,梁德翠.广义三支决策与狭义三支决策[J/OL].计算机科学与探索,2017,11(3):502-510[2018-10-01].http://www.cnki.com.cn/Article/CJFDTotal-KXTS201703019.htm.DOI:http://dx.doi.org/10.3778/j.issn.1673-9418.1605042.
[15] 于洪,王国胤,姚一豫.决策粗糙集理论研究现状与展望[J/OL].计算机学报,2015(8):1628-1639.[2018-10-01].http://www.cqvip.com/QK/90818X/201508/665892812.html.DOI:10.11897/SP.J.1016.2015.01628.
[16] 孙光民,王晨阳.一种基于改进SIFT的图像检索算法[J/OL].国外电子测量技术,2016(8):32-37[2018-10-01].http://www.cqvip.com/QK/94755X/201608/670040533.html.DOI:10.3969/j.issn.1002-8978.2016.08.008.
[17] 张铮,王艳平,薛桂香.数字图像处理与机器视觉[M/OL].北京:人民邮电出版社,2010:101-139[2018-10-01].http://www.bookask.com/book/1883206.html.