基于指数权重局部聚合向量特征的轮毂型号识别
|
摘要
在轮毂型号识别过程中,为了能在大量轮型库中快速识别正确的轮型,提出了基于指数权重局部聚合向量(VLAD)特征的轮型识别方法。VLAD特征是针对BOW特征的改进版,用待分类特征和聚类中心的累积残差代替特征的累加数目,采用四近邻软分配的查找方式,相对于一对一的分配规则具有更好的鲁棒性。最后把得到的VLAD向量进行主成分分析降维,并在降维VLAD的基础上将指数权重和VLAD向量的各数据相乘以削减个别不稳定值,最后通过特征向量的对比来找到最相似图片,识别过程具有非接触、灵活、准确的优点,实验表明在提高图片识别率的同时也具有较好的鲁棒性。
In the process of wheel model identification,in order to quickly identify the correct wheel model in a large number of wheel model data base,a identification algorithm based on index weight VLAD( vector of aggragate locally descriptor) is proposed. The VLAD feature is an improved version of the BOW,using the cumulative residuals between features and cluster centers replace the cumulative number of features,as the searching algorithm we use four nearest neighbor soft allocation,and has better robustness than the one to one allocation rule. Then the PCA is used to reduce dimensions of VLAD obtained above,the multiply index weight and the data of descending dimension VLAD to reduce individual instability value,and finally,the most similar pictures are found by comparing the feature vectors. The experimental results shows that this algorithm has the advantages of non-contact,perform well on flexibility and accuracy,it also improves the identification rate and has good robustness.
In the process of wheel model identification,in order to quickly identify the correct wheel model in a large number of wheel model data base,a identification algorithm based on index weight VLAD( vector of aggragate locally descriptor) is proposed. The VLAD feature is an improved version of the BOW,using the cumulative residuals between features and cluster centers replace the cumulative number of features,as the searching algorithm we use four nearest neighbor soft allocation,and has better robustness than the one to one allocation rule. Then the PCA is used to reduce dimensions of VLAD obtained above,the multiply index weight and the data of descending dimension VLAD to reduce individual instability value,and finally,the most similar pictures are found by comparing the feature vectors. The experimental results shows that this algorithm has the advantages of non-contact,perform well on flexibility and accuracy,it also improves the identification rate and has good robustness.
引文
[1]赵玉良,刘伟军,刘永贤,等.汽车车轮在线识别系统的研究[J].机械设计与制造,2007,(10):164-166.
[2]麻金贺.轮毂在线识别系统的研究[D].秦皇岛:燕山大学,2014.
[3]吴凤和.基于计算机视觉测量技术的图像轮廓提取方法研究[J].计量学报,2007,28(1):18-22.Wu F H. A Study on Contour Extraction Method in Computer Vision Mea surement Technology[J]. Acta Metrologica Sinica,2007,28(1):18-22.
[4]韩卫.基于多特征融合的汽车轮毂识别与分类系统设计[D].南京:南京信息工程大学,2016.
[5]张琴,林清华,康新.基于核密度估计和分形编码算法的图像检索技术研究[J].计量学报,2017,38(3):284-287.Zhang Q, Lin Q H, Kang X. Research on Image Retrieval Based on Kernel Density Estimation and Fractal Coding Algorithm[J]. Acta Metrologica Sinica,2017,38(3):284-287.
[6]杨光,冯涛,秦永左.轮型代码自动识别系统图像处理算法改进研究[J].核电子学与探测技术,2012,32(6):732-735.Yang G,Feng T,Qin Y Z. Study on the Improved Image Processing Algorithms of Wheel Code Automatic Identification System[J]. Nuclear Electronics&Detection Technology,2012,32(6):732-735.
[7]程淑红,管永来,张典范.基于形状匹配及纹理筛选的汽车轮毂型号识别[J].仪器仪表学报,2017,38(9):2299-2306.Chen S H, Guan Y L, Zhang D F. Wheel model identification based on shape recognition and texture filtering[J]. Chinese Journal of Scientific Instrument,2017,38(9):2299-2306.
[8] Zhao C, Chan S S F, Cham W K, et al. Plant identification using leaf shapes-A pattern counting approach[J]. Pattern Recognition,2015,48(10):3203-3215.
[9]陈华,张志娟,刘刚,等.基于局部纹理特性和图像分割的分步立体匹配[J].计量学报,2017,38(1):73-77.Chen H,Zhang Z J,Liu G,et al. A Two-step Stereo Matching Based on the Local Texture Features and Image Segmentation[J]. Acta Metrologica Sinica,2017,38(1):73-77.
[10]颜文,金炜,符冉迪.结合VLAD特征和稀疏表示的图像检索[J].电信科学,2016,32(12):80-85.Yan W,Jin W,Fu R D. Image retrieval based on the feature of VLAD and Sparse Representation[J].Telecommunications Science,2016,32(12):80-85.
[11] Jegou H,Douze M,Schmid C. Packingbag-of-features[C]//IEEE International Conference on Computer Vision. 2010:2357-2364.
[12] Tang M F,Nie F P,Jain R. A graph regularized dimension reduction method for out-of-sample data[J].Neurocomputing,2016,225:58-63.
[13] Ke Y,Sukthankar R. PCA-SIFT:A More Distinctive Representation for Local Image Descriptors[C]//IEEE Computer Society Conference on Computer Vision and Pattern Recongnition. 2004:506-513.
[14] Kim T E,Kim M H. Improving the search accuracy of the VLAD through weighted aggregation of local descriptors[J]. Journal of Visual Communication&Image Representation,2015,31:237-252.
[15] Liu Z,Wang S,Tian Q. Fine-residual VLAD for image retrieval[J]. Neurocomputing,2016,173(3):1183-1191.
[16]周玲.大规模图像检索中局部特征聚合与索引方法研究[D].武汉:华中科技大学,2011.
[17] Kim M. Dual soft assignment clustering algorithm for human action video clustering[J]. Computer Vision&Image Understanding,2017,155:106-112.
[18] Kumar C S,George K K,Ramachandran K I,et al.Weighted cosine distance features for speaker verification[C]//IEEE India Conference. 2016.
[2]麻金贺.轮毂在线识别系统的研究[D].秦皇岛:燕山大学,2014.
[3]吴凤和.基于计算机视觉测量技术的图像轮廓提取方法研究[J].计量学报,2007,28(1):18-22.Wu F H. A Study on Contour Extraction Method in Computer Vision Mea surement Technology[J]. Acta Metrologica Sinica,2007,28(1):18-22.
[4]韩卫.基于多特征融合的汽车轮毂识别与分类系统设计[D].南京:南京信息工程大学,2016.
[5]张琴,林清华,康新.基于核密度估计和分形编码算法的图像检索技术研究[J].计量学报,2017,38(3):284-287.Zhang Q, Lin Q H, Kang X. Research on Image Retrieval Based on Kernel Density Estimation and Fractal Coding Algorithm[J]. Acta Metrologica Sinica,2017,38(3):284-287.
[6]杨光,冯涛,秦永左.轮型代码自动识别系统图像处理算法改进研究[J].核电子学与探测技术,2012,32(6):732-735.Yang G,Feng T,Qin Y Z. Study on the Improved Image Processing Algorithms of Wheel Code Automatic Identification System[J]. Nuclear Electronics&Detection Technology,2012,32(6):732-735.
[7]程淑红,管永来,张典范.基于形状匹配及纹理筛选的汽车轮毂型号识别[J].仪器仪表学报,2017,38(9):2299-2306.Chen S H, Guan Y L, Zhang D F. Wheel model identification based on shape recognition and texture filtering[J]. Chinese Journal of Scientific Instrument,2017,38(9):2299-2306.
[8] Zhao C, Chan S S F, Cham W K, et al. Plant identification using leaf shapes-A pattern counting approach[J]. Pattern Recognition,2015,48(10):3203-3215.
[9]陈华,张志娟,刘刚,等.基于局部纹理特性和图像分割的分步立体匹配[J].计量学报,2017,38(1):73-77.Chen H,Zhang Z J,Liu G,et al. A Two-step Stereo Matching Based on the Local Texture Features and Image Segmentation[J]. Acta Metrologica Sinica,2017,38(1):73-77.
[10]颜文,金炜,符冉迪.结合VLAD特征和稀疏表示的图像检索[J].电信科学,2016,32(12):80-85.Yan W,Jin W,Fu R D. Image retrieval based on the feature of VLAD and Sparse Representation[J].Telecommunications Science,2016,32(12):80-85.
[11] Jegou H,Douze M,Schmid C. Packingbag-of-features[C]//IEEE International Conference on Computer Vision. 2010:2357-2364.
[12] Tang M F,Nie F P,Jain R. A graph regularized dimension reduction method for out-of-sample data[J].Neurocomputing,2016,225:58-63.
[13] Ke Y,Sukthankar R. PCA-SIFT:A More Distinctive Representation for Local Image Descriptors[C]//IEEE Computer Society Conference on Computer Vision and Pattern Recongnition. 2004:506-513.
[14] Kim T E,Kim M H. Improving the search accuracy of the VLAD through weighted aggregation of local descriptors[J]. Journal of Visual Communication&Image Representation,2015,31:237-252.
[15] Liu Z,Wang S,Tian Q. Fine-residual VLAD for image retrieval[J]. Neurocomputing,2016,173(3):1183-1191.
[16]周玲.大规模图像检索中局部特征聚合与索引方法研究[D].武汉:华中科技大学,2011.
[17] Kim M. Dual soft assignment clustering algorithm for human action video clustering[J]. Computer Vision&Image Understanding,2017,155:106-112.
[18] Kumar C S,George K K,Ramachandran K I,et al.Weighted cosine distance features for speaker verification[C]//IEEE India Conference. 2016.