基于MSER和SVM以及强种子区域生长的车牌定位
|
摘要
针对传统车牌定位方式的缺陷,给出了将自然场景下文本检测的最大稳定极值区域(MSER,maximally stable extremal regions)算法与支持向量机(SVM,support vector machine)相结合的车牌定位方法。首先利用MSER算子对车牌字符进行初步定位,并依据字符区域的高宽比和占空比剔除明显不是字符的区域,再剔除重叠的字符区域,从而得到候选字符区域,然后将候选字符区域输入训练好的SVM分类器,用来剔除无效的字符区域。最后利用强种子的区域生长法将真实的字符区域聚合,通过求解连通区域的外接矩形,最终提取车牌区域,实现对车牌的精确定位。对比实验结果表明,在不同的自然场景下,该方法比传统的车牌检测算法的定位准确率高2%～3%,其自适应能力和鲁棒性都比较好,具有较高的实用价值。
Faced with the defects of traditional license plate location, this paper proposes a license plate location method that combines the MSER algorithm for text detection under natural scenes with the machine learning algorithm SVM. Firstly, the MSER algorithm is used to preliminarily locate the license plate, and according to the aspect ratio and area duty ratio of the character area, the regions that are obviously not characters are removed, and then the overlapped character areas are removed. Then the candidate character regions are sent to the trained SVM classifier to eliminate invalid character regions. Finally, the real character regions are aggregated by using the region growth method of strong seeds, and the license plate regions are extracted by calculating the outer rectangle of the connected regions, so as to realize the accurate location of the license plate. The experimental results show that, under different natural scenes, the positioning accuracy of this method is 2%～3% higher than that of the traditional detection algorithm, and its adaptive ability and robustness are better, so it has higher practical value.
Faced with the defects of traditional license plate location, this paper proposes a license plate location method that combines the MSER algorithm for text detection under natural scenes with the machine learning algorithm SVM. Firstly, the MSER algorithm is used to preliminarily locate the license plate, and according to the aspect ratio and area duty ratio of the character area, the regions that are obviously not characters are removed, and then the overlapped character areas are removed. Then the candidate character regions are sent to the trained SVM classifier to eliminate invalid character regions. Finally, the real character regions are aggregated by using the region growth method of strong seeds, and the license plate regions are extracted by calculating the outer rectangle of the connected regions, so as to realize the accurate location of the license plate. The experimental results show that, under different natural scenes, the positioning accuracy of this method is 2%～3% higher than that of the traditional detection algorithm, and its adaptive ability and robustness are better, so it has higher practical value.
引文
[1] 周志刚.低对比度图像增强及图像边缘检测方法[D].武汉:华中科技大学,2014:2-6.ZHOU Z G.Low-contrast image enhancement and image edge detection methods[D].Wuhan:Huazhong University of Science and Technology,2014:2-6.(in Chinese)
[2] 侯向宁,刘华春.基于颜色分割与Sobel算子相结合的车牌定位[J].计算机技术与发展,2018,28(8):156-159.HOU X N,LIU H C.License plate location based on color segmentation and sobel operator[J].Computer Technology and Development,2018,28(8):156-159.(in Chinese)
[3] 郭延祥,陈耀武.基于边缘检测和颜色纹理直方图的车牌定位方法[J].计算机科学与探索,2014,8(6):719-726.GUO Y X,CHEN Y W.Vehicle license plate location method based on edge detection and color-texture histogram[J].Journal of Frontiers of Computer Science & Technology,2014,8(6):719-726.(in Chinese)
[4] 梁利利,王兴武.基于Sobel算子边缘检测和数学形态学的车牌定位算法[J].现代电子技术,2015,38(14):98-100.LIANG L L,WANG X W.License plate positioning algorithm based on Sobel operator edge detection and mathematical morphology[J].Modern Electronics Technique,2015,38(14):98-100.(in Chinese)
[5] 孙金岭,庞娟,张泽龙.基于颜色特征和改进 Canny 算子的车牌图像定位[J].吉林大学学报(理学版),2015,53(4):693-697.SUN J L,PANG J,ZHANG Z L.Recognition of vehicle license plate locating based on color feature and improved canny operator[J].Journal of Jilin University (Science Edition),2015,53(4):693-697.(in Chinese)
[6] 赵莉,白猛猛,雷松泽,等.深度学习在车牌定位中的研究[J].计算机应用研究,2018,35(10):3142-3146.ZHAO L,BAI M M,LEI S Z,et al.Vehicle license plate location by deep learning[J].Application Research of Computers,2018,35(10):3142-3146.(in Chinese)
[7] 李俊峰.基于RGB色彩空间自然场景统计的无参考图像质量评价[J].自动化学报,2015,41(9):1601-1615.LI J F.No-reference image quality assessment based on natural scene statistics in RGB color space[J].Acta Automatica Sinica,2015,41(9):1601-1615.(in Chinese)
[8] 刘书华,焦林.基于云计算的汽车车牌自动识别系统[J].西安工程大学学报,2013,27(3):388-391.LIU S H,JIAO L.The research on automobile vehicle license plate recognition based on cloud computing[J].Journal of Xi′an Polytechnic University,2013,27(3):388-391.(in Chinese)
[9] 薛亚娣,阮文惠.基于人工鱼群算法的彩色图像检索方法[J].西安工程大学学报,2016,30(5):651-656.XUE Y D,RUAN W H.Color image retrieval method based on artificial fish algorithm[J].Journal of Xi′an Polytechnic University,2016,30(5):651-656.(in Chinese)
[10] 马爽,陈江宁,卢虎,等.边缘特性筛选与多判定机制下的车牌定位方法[J].计算机工程与应用,2014,50(9):145-149.MA S,CHEN J N,LU H,et al.Approach for license plate location using edge features filter and multi-decision mechanism[J].Computer Engineering and Applications,2014,50(9):145-149.(in Chinese)
[11] 唐有宝,卜巍,邬向前.多层次MSER自然场景文本检测[J].浙江大学学报(工学版),2016,50(6):1134-1140.TANG Y B,BU W,WU X Q.Natural scene text detection based on multi-level MSER[J].Journal of Zhejiang University(Engineering Science),2016,50(6):1134-1140.(in Chinese)
[12] ZHANG X N,GAO X B,TIAN C N.Text detection in natural scene images based on color prior guided MSER[J].Neurocomputing,2018,307:61-71.
[13] THILAGAVATHY A,CHILAMBUCHELVAN A.Fuzzy based edge enhanced text detection algorithm using MSER[J].Cluster Comput,2017:1-7.https://doi.org/10.1007/s/0586-0017-1448-5
[14] 陈祖斌,黄维宁,赵逸夫,等.嵌入式监控影像车牌识别方法[J].吉林大学学报(信息科学版),2015,33(4):463-470.CHEN Z B,HUANG W N,ZHAO Y F,et al.Embedded license plate recognition system of surveillance video[J].Journal of Jilin University(Information Science Edition),2015,33(4):463-470.(in Chinese)
[15] 殷小舟.一种改进的结合K近邻法的SVM分类算法[J].中国图象图形学报,2009,14(11):2299-2303.YIN X Z.An ameliorated SVM classifying algorithm combined with kNN[J].Journal of Image and Graphics,2009,14(11):2299-2303.(in Chinese)
[16] 徐渊,许晓亮,李才年,等.结合SVM分类器与HOG特征提取的行人检测[J].计算机工程,2016,42(1):56-60.XU Y,XU X L,LI C N,et al.Pedestrian detection combining with SVM classifier and HOG feature extraction[J].Computer Engineering,2016,42(1):56-60.(in Chinese)
[17] ZHOU D G,SHAO Y H.Region growing for image segmentation using an extended PCNN model[J].IET Image Processing,2018,12(5):729-737.
[18] FAN J P,ZENG G H,BODY M,et al.Seeded region growing:an extensive and comparative study[J].Pattern Recognition Letters,2005,26(8):1139-1156.
[19] 闫东阳,明冬萍.基于自动多种子区域生长的遥感影像面向对象分割方法[J].工程科学学报,2017,39(11):1735-1742.YAN D Y,MING D P.Object-oriented remote sensing image segmentation based on automatic multiseed region growing algorithm[J].Chinese Journal of Engineering,2017,39(11):1735-1742.(in Chinese)
[20] 李修霞,荆林海,李慧,等.参考1维光谱差异的区域生长种子点选取方法[J].中国图象图形学报,2016,21(9):1256-1264.LI X X,JING L H,LI H,et al.Seed extraction method for seeded region growing based on one-dimensional spectral differences[J].Journal of Image and Graphics,2016,21(9):1256-1264.(in Chinese)
[2] 侯向宁,刘华春.基于颜色分割与Sobel算子相结合的车牌定位[J].计算机技术与发展,2018,28(8):156-159.HOU X N,LIU H C.License plate location based on color segmentation and sobel operator[J].Computer Technology and Development,2018,28(8):156-159.(in Chinese)
[3] 郭延祥,陈耀武.基于边缘检测和颜色纹理直方图的车牌定位方法[J].计算机科学与探索,2014,8(6):719-726.GUO Y X,CHEN Y W.Vehicle license plate location method based on edge detection and color-texture histogram[J].Journal of Frontiers of Computer Science & Technology,2014,8(6):719-726.(in Chinese)
[4] 梁利利,王兴武.基于Sobel算子边缘检测和数学形态学的车牌定位算法[J].现代电子技术,2015,38(14):98-100.LIANG L L,WANG X W.License plate positioning algorithm based on Sobel operator edge detection and mathematical morphology[J].Modern Electronics Technique,2015,38(14):98-100.(in Chinese)
[5] 孙金岭,庞娟,张泽龙.基于颜色特征和改进 Canny 算子的车牌图像定位[J].吉林大学学报(理学版),2015,53(4):693-697.SUN J L,PANG J,ZHANG Z L.Recognition of vehicle license plate locating based on color feature and improved canny operator[J].Journal of Jilin University (Science Edition),2015,53(4):693-697.(in Chinese)
[6] 赵莉,白猛猛,雷松泽,等.深度学习在车牌定位中的研究[J].计算机应用研究,2018,35(10):3142-3146.ZHAO L,BAI M M,LEI S Z,et al.Vehicle license plate location by deep learning[J].Application Research of Computers,2018,35(10):3142-3146.(in Chinese)
[7] 李俊峰.基于RGB色彩空间自然场景统计的无参考图像质量评价[J].自动化学报,2015,41(9):1601-1615.LI J F.No-reference image quality assessment based on natural scene statistics in RGB color space[J].Acta Automatica Sinica,2015,41(9):1601-1615.(in Chinese)
[8] 刘书华,焦林.基于云计算的汽车车牌自动识别系统[J].西安工程大学学报,2013,27(3):388-391.LIU S H,JIAO L.The research on automobile vehicle license plate recognition based on cloud computing[J].Journal of Xi′an Polytechnic University,2013,27(3):388-391.(in Chinese)
[9] 薛亚娣,阮文惠.基于人工鱼群算法的彩色图像检索方法[J].西安工程大学学报,2016,30(5):651-656.XUE Y D,RUAN W H.Color image retrieval method based on artificial fish algorithm[J].Journal of Xi′an Polytechnic University,2016,30(5):651-656.(in Chinese)
[10] 马爽,陈江宁,卢虎,等.边缘特性筛选与多判定机制下的车牌定位方法[J].计算机工程与应用,2014,50(9):145-149.MA S,CHEN J N,LU H,et al.Approach for license plate location using edge features filter and multi-decision mechanism[J].Computer Engineering and Applications,2014,50(9):145-149.(in Chinese)
[11] 唐有宝,卜巍,邬向前.多层次MSER自然场景文本检测[J].浙江大学学报(工学版),2016,50(6):1134-1140.TANG Y B,BU W,WU X Q.Natural scene text detection based on multi-level MSER[J].Journal of Zhejiang University(Engineering Science),2016,50(6):1134-1140.(in Chinese)
[12] ZHANG X N,GAO X B,TIAN C N.Text detection in natural scene images based on color prior guided MSER[J].Neurocomputing,2018,307:61-71.
[13] THILAGAVATHY A,CHILAMBUCHELVAN A.Fuzzy based edge enhanced text detection algorithm using MSER[J].Cluster Comput,2017:1-7.https://doi.org/10.1007/s/0586-0017-1448-5
[14] 陈祖斌,黄维宁,赵逸夫,等.嵌入式监控影像车牌识别方法[J].吉林大学学报(信息科学版),2015,33(4):463-470.CHEN Z B,HUANG W N,ZHAO Y F,et al.Embedded license plate recognition system of surveillance video[J].Journal of Jilin University(Information Science Edition),2015,33(4):463-470.(in Chinese)
[15] 殷小舟.一种改进的结合K近邻法的SVM分类算法[J].中国图象图形学报,2009,14(11):2299-2303.YIN X Z.An ameliorated SVM classifying algorithm combined with kNN[J].Journal of Image and Graphics,2009,14(11):2299-2303.(in Chinese)
[16] 徐渊,许晓亮,李才年,等.结合SVM分类器与HOG特征提取的行人检测[J].计算机工程,2016,42(1):56-60.XU Y,XU X L,LI C N,et al.Pedestrian detection combining with SVM classifier and HOG feature extraction[J].Computer Engineering,2016,42(1):56-60.(in Chinese)
[17] ZHOU D G,SHAO Y H.Region growing for image segmentation using an extended PCNN model[J].IET Image Processing,2018,12(5):729-737.
[18] FAN J P,ZENG G H,BODY M,et al.Seeded region growing:an extensive and comparative study[J].Pattern Recognition Letters,2005,26(8):1139-1156.
[19] 闫东阳,明冬萍.基于自动多种子区域生长的遥感影像面向对象分割方法[J].工程科学学报,2017,39(11):1735-1742.YAN D Y,MING D P.Object-oriented remote sensing image segmentation based on automatic multiseed region growing algorithm[J].Chinese Journal of Engineering,2017,39(11):1735-1742.(in Chinese)
[20] 李修霞,荆林海,李慧,等.参考1维光谱差异的区域生长种子点选取方法[J].中国图象图形学报,2016,21(9):1256-1264.LI X X,JING L H,LI H,et al.Seed extraction method for seeded region growing based on one-dimensional spectral differences[J].Journal of Image and Graphics,2016,21(9):1256-1264.(in Chinese)