Android手机上图像分类技术的研究
摘要
近几年,随着智能手机的飞速发展,人们对手机功能的需求越来越多样化。其中,智能手机一个重要的功能是拍照。利用手机拍摄下来的照片可应用于制作头像、社交网络分享等诸多领域。在智能手机上实现图像场景分类技术,将有助于手机在拍照时自动设置光圈快门等参数,提高拍摄照片的质量。
本论文面向智能终端上的前沿需求,研究了Android手机上图像分类的技术,实现了Android上的图像场景分类系统。首先,研究了基于颜色、纹理、形状和空间特征的图像分类技术,根据Android手机处理速度有限,内存空间较小的特点,最终选择了抽取图片颜色及纹理信息的CEDD特征描述符作为图片特征提取方式,这就使本系统在处理速度和质量上保持了较好的平衡。在系统中,用Android手机拍摄照片后,抽取图像CEDD特征值并通过“一对一”SVM分类方法最终将照片分为风景、人物、夜间、逆光、微距、文本六个类别。在照片分类处理问题上,对libsvm预测功能进行了引入分类训练模型和投票选择部分的改写,并移植到Android手机上
本论文详细介绍了Android上图像场景分类系统的实现过程。系统共分为界面模块、NDK接口模块、特征提取模块和分类判断模块四个部分,是分别在Android操作系统上通过Java与C++编程环境实现的,通过研究Android NDK的开发方法,将C++文件编译为动态链接库与系统上层应用建立连接,在Android系统上成功实现了CEDD特征抽取及SVM分类工作。在本系统中,用户拍摄照片后,经过系统分类处理,就能够获得当前拍摄照片所属类别的结果。
最后,对实现的图像场景分类系统性能进行了测试。测试结果表明系统分类效果良好。
In recent years, with the rapid development of smart phones, the need of smart phone's function is becoming more and more diverse. Taking pictures is an important function among them. The photos taken by smart phones can be used in avatar, social network sharing and other areas. Image classification can help smart phone to set parameters such as aperture and shutter automatically, improve the quality of pictures.
In order to meet the demand, this paper researches the technology of image classification on smartphones, implementing the Android image scene classification system. This paper introduces image classification technologies based on color, texture, shape and spatial features. Due to Android phone's processing speed and memory limit, CEDD which combines the color and texture features was choosed as the method of extracting image information, it keeps a good balance between features' size and retrieval quality. The system extacts CEDD descriptors after taking picture, then classifies the photo into one of the six categories whitch include landscape, people, night, backlit, macro and text by one to one SVM method. In the classification prcessing, this paper rewrite the code of training model and voting in predict function from libsvm, then move it to Android mobile phone.
This paper introduces the implementation process of classification system. The system has four modules, including interface module, NDK interface module, feature extracting module and classification jugement module. They work on Android operating system with the programming environment of Java and C. By study of the development of the Android NDK, this paper achieve the feature extraction and SVM classification by establishing a connection between the system upper application and dynamic link library from compiled C files. In this system, the user will get the result of classification through processing after taking a picture.
Finally, the performance of the implementation of the image scene classification system was tested. The results show that the system has a good performance.
本论文面向智能终端上的前沿需求,研究了Android手机上图像分类的技术,实现了Android上的图像场景分类系统。首先,研究了基于颜色、纹理、形状和空间特征的图像分类技术,根据Android手机处理速度有限,内存空间较小的特点,最终选择了抽取图片颜色及纹理信息的CEDD特征描述符作为图片特征提取方式,这就使本系统在处理速度和质量上保持了较好的平衡。在系统中,用Android手机拍摄照片后,抽取图像CEDD特征值并通过“一对一”SVM分类方法最终将照片分为风景、人物、夜间、逆光、微距、文本六个类别。在照片分类处理问题上,对libsvm预测功能进行了引入分类训练模型和投票选择部分的改写,并移植到Android手机上
本论文详细介绍了Android上图像场景分类系统的实现过程。系统共分为界面模块、NDK接口模块、特征提取模块和分类判断模块四个部分,是分别在Android操作系统上通过Java与C++编程环境实现的,通过研究Android NDK的开发方法,将C++文件编译为动态链接库与系统上层应用建立连接,在Android系统上成功实现了CEDD特征抽取及SVM分类工作。在本系统中,用户拍摄照片后,经过系统分类处理,就能够获得当前拍摄照片所属类别的结果。
最后,对实现的图像场景分类系统性能进行了测试。测试结果表明系统分类效果良好。
In recent years, with the rapid development of smart phones, the need of smart phone's function is becoming more and more diverse. Taking pictures is an important function among them. The photos taken by smart phones can be used in avatar, social network sharing and other areas. Image classification can help smart phone to set parameters such as aperture and shutter automatically, improve the quality of pictures.
In order to meet the demand, this paper researches the technology of image classification on smartphones, implementing the Android image scene classification system. This paper introduces image classification technologies based on color, texture, shape and spatial features. Due to Android phone's processing speed and memory limit, CEDD which combines the color and texture features was choosed as the method of extracting image information, it keeps a good balance between features' size and retrieval quality. The system extacts CEDD descriptors after taking picture, then classifies the photo into one of the six categories whitch include landscape, people, night, backlit, macro and text by one to one SVM method. In the classification prcessing, this paper rewrite the code of training model and voting in predict function from libsvm, then move it to Android mobile phone.
This paper introduces the implementation process of classification system. The system has four modules, including interface module, NDK interface module, feature extracting module and classification jugement module. They work on Android operating system with the programming environment of Java and C. By study of the development of the Android NDK, this paper achieve the feature extraction and SVM classification by establishing a connection between the system upper application and dynamic link library from compiled C files. In this system, the user will get the result of classification through processing after taking a picture.
Finally, the performance of the implementation of the image scene classification system was tested. The results show that the system has a good performance.
引文
[1]刘亚丽.Android软件前景浅析[J].科技创新导报2011.25：16-20.
[2]李弼程,彭天强,彭波等.智能图像处理技术[M].北京：电子工业出版社.2004：209—-210.
[3]易茜.采用彩色CCD图像的加工表面烧伤度研究[学位论文].南京：南京航空航天大学.2009.
[4]王娟,孔兵,贾巧丽.基于颜色特征的图像检索技术[J].计算机系统应用.2011.7：160-164.
[5]唐寿成.图像分类方法的比较及应用[学位论文].北京：北京邮电大学.2011.
[6]谢毓湘,吴玲达,栾悉道.基于内容的图像检索技术研究[J].计算机工程与应用.2002.1：35-38.
[7]Manjunath B S,Ma W Y.Texture features for browsing and retrieval of image data. IEEE Transactions on Pattern Analysis and Machine Intelligence.1996
[8]高程程,惠晓威.基于灰度共生矩阵的纹理特征提取[J].计算机系统应用.2010.6：69-71.
[9]Peleg S, Naor J, Hartleycet al. Multiple conclution texture allalysis and classification[J]. IIIEE Transactions on PatternAnalysis and Machille InteUigence,1984,6(4):518-523.
[10]ISO/IEC/JTC1/SC29/WG11:MPEG-7 XM Document:MPE-G-7 Visual Part Experimentation Model Version 10.0, MPEG Document N4063, Singapore (2001)
[11]杨楠.基于内容的商品图像分类技术研究[学位论文].大连：大连理工大学.2011.
[12]Paul E, Utgoff.Incremental induction of decision trees[J]Maehine Learning, 1989,4(2):161-186.
[13]徐义蜂,陈春明,徐云青.一种改进的k-均值聚类算法[J].计算机应用与软件.2008.3：275-277.
[14]李雪梅.数字识别技术探究.[J].电力学报.2007.3：348-351.
[15]Weston J,Watkins C. Multi-class support vector machines. Royal Holloway College, "rech Rep:SCD-lm-98-04,1998.
[16]P. Xu, A. K. Chan. Support vector machine for multi-class signal classification with unbalanced samples. Proceedings of the International Joint Conference on Neural Networks 2003. Portland,pp.l 116-1119,2003.
[17]秦磊,高文.基于内容相关性的场景图像分类方法[J].计算机研究与发展.2009.46(7)：1198—1205.
[18]王胜正.基于两种色彩空间的颜色选择方法.[J].计算机应用与软件.2004：2：35-39.
[19]Chatzichristofis, S., Boutalis, Y.:A Hybrid Scheme for Fast and Accurate Image Retrieval Based on Color Descriptors. In:De Mallorca, P. (ed.) IASTED International Conference.
[20]Konstantinidis, K., Gasteratos, A., Andreadis, I.:Image Retrieval Based on Fuzzy Color Histogram Processing. Optics Communications 248(4-6),15,375-386 (2005)
[21]Mertzios, B., Tsirikolias, K.:Coordinate Logic Filters:Theory and Applications Nonlinear Image Processing. In:Mitra, S., Sicuranza, G. (eds.), ch.11, Academic Press, London (2004), ISBN:0125004516
[22]Savvas A. Chatzichristofis, Yiannis S. Boutalis.FCTH:Fuzzy Color and Texture Histogram Alow Level Feature for Accurate Image Retrieval [J].Department of Electrical & Computer Engineering Democritus University of Thrace.67100,2008.
[23]Zimmerman, H.J.:Fuzzy Sets, Decision Making and Expert Systems. Kluwer Academic Publ. Boston (1987)
[24]江嘉明.基于自然图片的人脸检测识别系统的实现[J].大连理工大学：大连理工大学.2008
[25]Won, C.S., Park, D.K., Park, S.-J.:Efficient Use of MPEG-7 Edge Histogram Descriptor. ETRI Journal 24 (2002)
[26]Savvas A. Chatzichristofis, Yiannis S. Boutalis. CEDD:Color and Edge Directivity Descriptor. A Compact Descriptor for Image Indexing and Retrieval [J]. Department of Electrical & Computer Engineering Democritus University of Thrace.67100,2009.
[27]Tom M.Mitchell,著.曾花军,张银奎,等.译.机器学习[M].北京：机械工业出版社,2003.
[28]周芳.智能监控系统中运动目标识别技术的研究[学位论文].武汉：武汉科技大学.2011.
[29]吴崇明,王晓丹,白冬婴,等.利用KKT条件与类边界包向量的SVM增量学 习算法[J].计算机工程与设计.2010.8：145-146.
[30]邬啸,魏延,吴瑕.基于混合函数的支持向量机[J].重庆理工大学学报(自然科学).2011.10：66-70.
[31]李琳,张晓龙.基于RBF核的SVM学习算法的优化计算[J].计算机工程与应用.2006.29：190-192.
[32]王巍.基于先验知识的支持向量机图像分割算法研究[学位论文].西安：西安工业大学.2009.
[33]金智义,张戟.基于Android平台的串口通信实现[J].电脑知识与技术.2011.13：2983-2985.
[34]刘洁彬.面向实时监控的流媒体播放器的设计与实现[学位论文].北京：北京邮电大学.2010.
[35]龚伟.基于gdb的嵌入式系统调试器的设计与实现[学位论文].成都：电子科技大学.2006.
[36]曹欣.半虚拟化技术分析与研究[学位论文].浙江：浙江大学.2008.
[37]尹红梅.在Delphi7.0中使用DLL[J].现代电子技术.2006.17：69-71.
[38]柯家海.摄影曝光的理论初探[J].中国现代教育装备.2010.9：47-49.
[2]李弼程,彭天强,彭波等.智能图像处理技术[M].北京：电子工业出版社.2004：209—-210.
[3]易茜.采用彩色CCD图像的加工表面烧伤度研究[学位论文].南京：南京航空航天大学.2009.
[4]王娟,孔兵,贾巧丽.基于颜色特征的图像检索技术[J].计算机系统应用.2011.7：160-164.
[5]唐寿成.图像分类方法的比较及应用[学位论文].北京：北京邮电大学.2011.
[6]谢毓湘,吴玲达,栾悉道.基于内容的图像检索技术研究[J].计算机工程与应用.2002.1：35-38.
[7]Manjunath B S,Ma W Y.Texture features for browsing and retrieval of image data. IEEE Transactions on Pattern Analysis and Machine Intelligence.1996
[8]高程程,惠晓威.基于灰度共生矩阵的纹理特征提取[J].计算机系统应用.2010.6：69-71.
[9]Peleg S, Naor J, Hartleycet al. Multiple conclution texture allalysis and classification[J]. IIIEE Transactions on PatternAnalysis and Machille InteUigence,1984,6(4):518-523.
[10]ISO/IEC/JTC1/SC29/WG11:MPEG-7 XM Document:MPE-G-7 Visual Part Experimentation Model Version 10.0, MPEG Document N4063, Singapore (2001)
[11]杨楠.基于内容的商品图像分类技术研究[学位论文].大连：大连理工大学.2011.
[12]Paul E, Utgoff.Incremental induction of decision trees[J]Maehine Learning, 1989,4(2):161-186.
[13]徐义蜂,陈春明,徐云青.一种改进的k-均值聚类算法[J].计算机应用与软件.2008.3：275-277.
[14]李雪梅.数字识别技术探究.[J].电力学报.2007.3：348-351.
[15]Weston J,Watkins C. Multi-class support vector machines. Royal Holloway College, "rech Rep:SCD-lm-98-04,1998.
[16]P. Xu, A. K. Chan. Support vector machine for multi-class signal classification with unbalanced samples. Proceedings of the International Joint Conference on Neural Networks 2003. Portland,pp.l 116-1119,2003.
[17]秦磊,高文.基于内容相关性的场景图像分类方法[J].计算机研究与发展.2009.46(7)：1198—1205.
[18]王胜正.基于两种色彩空间的颜色选择方法.[J].计算机应用与软件.2004：2：35-39.
[19]Chatzichristofis, S., Boutalis, Y.:A Hybrid Scheme for Fast and Accurate Image Retrieval Based on Color Descriptors. In:De Mallorca, P. (ed.) IASTED International Conference.
[20]Konstantinidis, K., Gasteratos, A., Andreadis, I.:Image Retrieval Based on Fuzzy Color Histogram Processing. Optics Communications 248(4-6),15,375-386 (2005)
[21]Mertzios, B., Tsirikolias, K.:Coordinate Logic Filters:Theory and Applications Nonlinear Image Processing. In:Mitra, S., Sicuranza, G. (eds.), ch.11, Academic Press, London (2004), ISBN:0125004516
[22]Savvas A. Chatzichristofis, Yiannis S. Boutalis.FCTH:Fuzzy Color and Texture Histogram Alow Level Feature for Accurate Image Retrieval [J].Department of Electrical & Computer Engineering Democritus University of Thrace.67100,2008.
[23]Zimmerman, H.J.:Fuzzy Sets, Decision Making and Expert Systems. Kluwer Academic Publ. Boston (1987)
[24]江嘉明.基于自然图片的人脸检测识别系统的实现[J].大连理工大学：大连理工大学.2008
[25]Won, C.S., Park, D.K., Park, S.-J.:Efficient Use of MPEG-7 Edge Histogram Descriptor. ETRI Journal 24 (2002)
[26]Savvas A. Chatzichristofis, Yiannis S. Boutalis. CEDD:Color and Edge Directivity Descriptor. A Compact Descriptor for Image Indexing and Retrieval [J]. Department of Electrical & Computer Engineering Democritus University of Thrace.67100,2009.
[27]Tom M.Mitchell,著.曾花军,张银奎,等.译.机器学习[M].北京：机械工业出版社,2003.
[28]周芳.智能监控系统中运动目标识别技术的研究[学位论文].武汉：武汉科技大学.2011.
[29]吴崇明,王晓丹,白冬婴,等.利用KKT条件与类边界包向量的SVM增量学 习算法[J].计算机工程与设计.2010.8：145-146.
[30]邬啸,魏延,吴瑕.基于混合函数的支持向量机[J].重庆理工大学学报(自然科学).2011.10：66-70.
[31]李琳,张晓龙.基于RBF核的SVM学习算法的优化计算[J].计算机工程与应用.2006.29：190-192.
[32]王巍.基于先验知识的支持向量机图像分割算法研究[学位论文].西安：西安工业大学.2009.
[33]金智义,张戟.基于Android平台的串口通信实现[J].电脑知识与技术.2011.13：2983-2985.
[34]刘洁彬.面向实时监控的流媒体播放器的设计与实现[学位论文].北京：北京邮电大学.2010.
[35]龚伟.基于gdb的嵌入式系统调试器的设计与实现[学位论文].成都：电子科技大学.2006.
[36]曹欣.半虚拟化技术分析与研究[学位论文].浙江：浙江大学.2008.
[37]尹红梅.在Delphi7.0中使用DLL[J].现代电子技术.2006.17：69-71.
[38]柯家海.摄影曝光的理论初探[J].中国现代教育装备.2010.9：47-49.