基于几何矩的手势识别算法
|
摘要
手势识别的研究具有广阔的实际应用前景，具体表现在诸多方面如：对语音识别起着辅助作用；利用手势控制VR(Virtual Reality)中的智能化；机器人的示范学习；虚拟现实系统中的多模式接口；可以使聋哑人使用手语和正常人交流等。另外，手语的研究涉及到教学、计算机图形学、机器人运动学、医学等多学科。
本文结合上海市自然科学基金资助课题“手势识别”，从手势图像的预处理、手势特征提取和手势识别等三个方面研究了基于视觉的手势识别的识别算法。提出了一种基于统计分析的手势识别的方法，利用几何矩进行特征提取，并应用到手势识别中去，解决了手势识别过程中手势的旋转，缩放，尺度变换所带来的问题，使得手势识别系统具有很好的稳定性。
在图像预处理阶段，我们主要是对手势字母图像进行平滑、锐化和二值化的处理。平滑主要采用模板操作；锐化过程中使用拉普拉斯算子进行锐化：而二值化则采用的是最大化方差的做法。
特征提取的好坏则直接关系到手势识别的效果。由于手势图像具有旋转、尺度等不确定性，给特征提取带来了诸多困难，而几何矩是一种基于统计分析的算法，本文即将其应用于提取不同手势的特征，被提取的特征可以做到不随图像的旋转、平移的变化而变化，应用到手势识别系统中去表现出有着良好的适应性、稳定性。
在样本库的建立过程中，我们采集了中国手语字母手势，共三套，其中两套作学习用，以形成标准库，第三套则用来测试算法的识别率。
识别过程则是将输入的手势图像进行一些预处理、提取特征，再与标准库中的手势的特征做比对，距离(加权的欧几里德距离)最小的即为匹配手语。
实验中，我们使用了汉语手势字母的全部30个手势，并加入阿拉伯数字10个手势，共40个手势，实验中一次识别率达到86.7％，累积二次识别率即可达93％。
The Research of Gesture-Language can be applied in many fields such as Computer Access Gesture-Language Teaching, TV Bilingual Broadcasting, the search of Virtual Human. The search of Gesture-Language includes the following subjects : Teaching, Computer Graphology, Robot Motion and Physic etc. It is a very meaningful subject. The Search of Gesture Recognition has a wide range of applications such as : the communication between the deaf and the normal, the access recognition of voice recognition ,the control of VR, the study of robot.
This article gives a method of Gesture Recognition based on Statistics. We can apply the Geometric Moment in the Feature Extraction. We can solve the problem of Recognition brought by rotation and scale.
The Gesture Recognition mainly includes the following process: Image preprocessing, feature extraction, pattern recognition.
In the process of image preprocessing there are several image operations such as image smooth, image sharpen and image segment. We use the template operation in the process of image smoothing, laplacian operator hi the process of image sharpening and maximum variance method in the process of Region segment.
Feature Extraction is vital to Gesture Recognition. The uncertainty of rotation, scale of gesture brings many difficulties to the extraction of feature. Geometric Moment is a arithmetic based on statistics. This article applys the arithmetic in the extraction of gesture feature. The feature can remains the same when the image rotating and scaling.
In the process of constructing the sample lib, we collected three sets of Chinese Letter Gesture. Two sets were used to machine learning, the third was used to test the recognition rate the Gesture Recognition System.
The Recognition Process includes the following phases: firstly, image preprocessing; secondly , feature extraction ;finally, matching with the Standard Gesture Lib. The Gesture which has the minimum Distance to the Input Gesture is the
Recognition Result-In the experiment, I used 30 Chinese Pinyin gestures and 10 Arabic number gestures, the cumulate recognition rate is up to 93 %.
Tao Yin
Directed by -.Yuan Ge
本文结合上海市自然科学基金资助课题“手势识别”，从手势图像的预处理、手势特征提取和手势识别等三个方面研究了基于视觉的手势识别的识别算法。提出了一种基于统计分析的手势识别的方法，利用几何矩进行特征提取，并应用到手势识别中去，解决了手势识别过程中手势的旋转，缩放，尺度变换所带来的问题，使得手势识别系统具有很好的稳定性。
在图像预处理阶段，我们主要是对手势字母图像进行平滑、锐化和二值化的处理。平滑主要采用模板操作；锐化过程中使用拉普拉斯算子进行锐化：而二值化则采用的是最大化方差的做法。
特征提取的好坏则直接关系到手势识别的效果。由于手势图像具有旋转、尺度等不确定性，给特征提取带来了诸多困难，而几何矩是一种基于统计分析的算法，本文即将其应用于提取不同手势的特征，被提取的特征可以做到不随图像的旋转、平移的变化而变化，应用到手势识别系统中去表现出有着良好的适应性、稳定性。
在样本库的建立过程中，我们采集了中国手语字母手势，共三套，其中两套作学习用，以形成标准库，第三套则用来测试算法的识别率。
识别过程则是将输入的手势图像进行一些预处理、提取特征，再与标准库中的手势的特征做比对，距离(加权的欧几里德距离)最小的即为匹配手语。
实验中，我们使用了汉语手势字母的全部30个手势，并加入阿拉伯数字10个手势，共40个手势，实验中一次识别率达到86.7％，累积二次识别率即可达93％。
The Research of Gesture-Language can be applied in many fields such as Computer Access Gesture-Language Teaching, TV Bilingual Broadcasting, the search of Virtual Human. The search of Gesture-Language includes the following subjects : Teaching, Computer Graphology, Robot Motion and Physic etc. It is a very meaningful subject. The Search of Gesture Recognition has a wide range of applications such as : the communication between the deaf and the normal, the access recognition of voice recognition ,the control of VR, the study of robot.
This article gives a method of Gesture Recognition based on Statistics. We can apply the Geometric Moment in the Feature Extraction. We can solve the problem of Recognition brought by rotation and scale.
The Gesture Recognition mainly includes the following process: Image preprocessing, feature extraction, pattern recognition.
In the process of image preprocessing there are several image operations such as image smooth, image sharpen and image segment. We use the template operation in the process of image smoothing, laplacian operator hi the process of image sharpening and maximum variance method in the process of Region segment.
Feature Extraction is vital to Gesture Recognition. The uncertainty of rotation, scale of gesture brings many difficulties to the extraction of feature. Geometric Moment is a arithmetic based on statistics. This article applys the arithmetic in the extraction of gesture feature. The feature can remains the same when the image rotating and scaling.
In the process of constructing the sample lib, we collected three sets of Chinese Letter Gesture. Two sets were used to machine learning, the third was used to test the recognition rate the Gesture Recognition System.
The Recognition Process includes the following phases: firstly, image preprocessing; secondly , feature extraction ;finally, matching with the Standard Gesture Lib. The Gesture which has the minimum Distance to the Input Gesture is the
Recognition Result-In the experiment, I used 30 Chinese Pinyin gestures and 10 Arabic number gestures, the cumulate recognition rate is up to 93 %.
Tao Yin
Directed by -.Yuan Ge
引文
1．殷涛，葛元，王林泉 基于几何矩的字母手势识别算法 计算机工程
2．王林泉，章文怡，郑刚．区域特征的乐谱识别系统．软件学报．第5卷第11期 1994年11月
3．王林泉．关于手写汉字识别的研究．计算机研究与发展 1987．24(7)：14-22
4. MING-KUEI HU. Visual pattern recognition by moment invariants. Ire transactions on information theory. 1962.179-187
5．金丰华，秦磊，汪蕙等几何矩不变量在基于内容医学图像检索中的应用．山东生物医学工程．第21卷第2期
6．陈岩松，郑师海，李德华．二维光学几何矩变换．物理学报．第40卷第10期
7 Huttenlocher D P, Klanderman G A, Rucklidge W J. Comparing images using the Hausdorff distance [J]. IEEE Transactions on PAMI, 1993, 15 (9): 850～863.
8 Mukundan. Moment Functions In Image Analysis. World Scientific Publishing Co. Pte. Ltd. 1998
9 Reyman Milonfar, A Moment-Based Variational Approach to Tomographic Reconstruction.IEEE Trans on Image Processing. Vol.5 No.3 1996
10 Hinyannaich H.P. Moments estimation in Radon Space. Pattern Recognition Left..Vol15 No.3. 1994
11 K.R.Castleman.Digital Image Processing. Prentice-Hall.Inc. 1996
12 黄修武．图像录属度及其在人脸识别中的应用．计算机研究和发展Vol．35．．No．11．1998
13 Hong Z.Q. Algebraic feature extraction of image for recognition. Pattem Recognition.1991 14(3)
14 田原．图像点集不变性匹配方法．模式识别与人工智能．Vol．11，No．3．1998
15 马颂德．计算机视觉．科学出版社．1998
16 Tom Malzbender.Fourier Volume Rendering.ACM Trans On Graphics.Vol.12 No.3.1993
17 Reyman Milonfar "A Moment-Based Variational Approach to Tomographic Reconstruction",IEEE Trans on Image Process, Vol.5..No.3.1996
18 章毓晋，刘忠伟基于HIS模型和累积直方图的彩色图像检索第八届全国信号处理学委员会联合论文集，1997，256～260
19 袁昕，朱淼良 基于主色匹配的图像检索系统 计算辅助设计与图形学学报，2000，12(12)：917～921
20 刘忠伟，章毓晋 综合利用颜色和纹理特征的图像检索．Journal of China Institute of Communications，1999，20(5)：36～40
21 王耀明．基于Radon变换的图像矩特征抽取及其在图像识别中的应用．计算机工程．Vol．27．No．2．2001
22 高立志，图像的小波矩．西安交通大学学报．Vol．33．No．5．1995．5
23 姚玉荣．利用小波和矩进行基于形状的图像检索．中国图像图形学 Vol．5(A)，No．3．2000．3
24 边肇祺 模式识别．第二版．北京：清华大学出版社，2000．1
25 李松毅，周瑾丹，张惠 基于矩的数字图像多边形逼近方法．计算机学报 第24卷第4期 2001年4月
26 Chang Pei. Color Image Processing by Using Binaxy Quaternion-Moment-Presering Thresholding Technique. IEEE Trans on Image Processing Vol.8.No.5. 1999
27 Levin D N, Pelizzari C A, Chen G T Y, et al, Retrospective geometric correlation of MR, CT, and PET images[J]. Radiology, 1988,169:817-823
28 Alpert N M, Bradshaw J F, Kenedy D, et al. The pricipal transformation : a method for image registration[J] JNM, 1990,31:1717-1722
29 Bookstein F L.Principal warps:thin-plate splines and the decomposition of deformations[J] IEEE Trans Pattern Anal Mach Intell, 1989,11:567-585
30 Shu HZ ,Luo L M, Yu W X,et al. A new fast method for computeting Legendre moments[J]. Pattern Recognition 200,33:341-348.
31 王秋让，赵荣椿，郑南宁 一种航空照片中的小目标识别方法 中国图像图形学报 第3卷第12期 1998年12月
32 Sahoo P, Wuukins C, yeager J. Threshold Selection Using Renyi's Entropy .Pattern Recognition, 1997,30(1)
33 Cem Yuccer, Kem al O flazer, A Rotation, Scaling, and T ranslation Invariant Pattern Classification System. Pattern Recognition,26,687～710
34 A.K.Jain and B.Chandrasekaran,"Dimensionality and Sample Size Considerations in Pattern Recognition Practice," Handbook of Statistics (Vol.2,pp.835-855),North Holland Publishing Company,1982
35 G, Cybenko,Continuous Valued Neural Network with Two Hidden Layers Are Sufficient, Technical Report, Dept.of Computer Science. Tufts University, Bedford, MA, 1988
36 L. Ott and W.Mendenhall, Understanding Statistics(5th ed), PWS-KENT , Boston, 1990.
37 Kenneth R. Castleman Digital Image Processing Prentice-Hall International, Inc 1998
38 Donald O.Tanguay, Jr .Hidden Markov Models for Gesture Recognition , Donald Tanguay, 1995
39 Chan Wah Ng, Surendra Ranganath Real-time gesture recognition system and application, Image and Vision Computing 20(2002)993-1007
40 任海兵，祝远新，林学阂 基于视觉手势识别的研究-综述 电子学报 第2期 2000年2月
41 方志刚 计算机手势输入及其在人机交互技术中的应用 小型微型计算机系统 第20卷 第6期限 1999年6月
42 甘俊英，张有为 基于不变矩特征和神经网络的人脸识别模型
43 杨静 丘江 王岩飞 线性不变矩及其在图像识别中的应用算法研究 光子学报 第32卷第3期
2．王林泉，章文怡，郑刚．区域特征的乐谱识别系统．软件学报．第5卷第11期 1994年11月
3．王林泉．关于手写汉字识别的研究．计算机研究与发展 1987．24(7)：14-22
4. MING-KUEI HU. Visual pattern recognition by moment invariants. Ire transactions on information theory. 1962.179-187
5．金丰华，秦磊，汪蕙等几何矩不变量在基于内容医学图像检索中的应用．山东生物医学工程．第21卷第2期
6．陈岩松，郑师海，李德华．二维光学几何矩变换．物理学报．第40卷第10期
7 Huttenlocher D P, Klanderman G A, Rucklidge W J. Comparing images using the Hausdorff distance [J]. IEEE Transactions on PAMI, 1993, 15 (9): 850～863.
8 Mukundan. Moment Functions In Image Analysis. World Scientific Publishing Co. Pte. Ltd. 1998
9 Reyman Milonfar, A Moment-Based Variational Approach to Tomographic Reconstruction.IEEE Trans on Image Processing. Vol.5 No.3 1996
10 Hinyannaich H.P. Moments estimation in Radon Space. Pattern Recognition Left..Vol15 No.3. 1994
11 K.R.Castleman.Digital Image Processing. Prentice-Hall.Inc. 1996
12 黄修武．图像录属度及其在人脸识别中的应用．计算机研究和发展Vol．35．．No．11．1998
13 Hong Z.Q. Algebraic feature extraction of image for recognition. Pattem Recognition.1991 14(3)
14 田原．图像点集不变性匹配方法．模式识别与人工智能．Vol．11，No．3．1998
15 马颂德．计算机视觉．科学出版社．1998
16 Tom Malzbender.Fourier Volume Rendering.ACM Trans On Graphics.Vol.12 No.3.1993
17 Reyman Milonfar "A Moment-Based Variational Approach to Tomographic Reconstruction",IEEE Trans on Image Process, Vol.5..No.3.1996
18 章毓晋，刘忠伟基于HIS模型和累积直方图的彩色图像检索第八届全国信号处理学委员会联合论文集，1997，256～260
19 袁昕，朱淼良 基于主色匹配的图像检索系统 计算辅助设计与图形学学报，2000，12(12)：917～921
20 刘忠伟，章毓晋 综合利用颜色和纹理特征的图像检索．Journal of China Institute of Communications，1999，20(5)：36～40
21 王耀明．基于Radon变换的图像矩特征抽取及其在图像识别中的应用．计算机工程．Vol．27．No．2．2001
22 高立志，图像的小波矩．西安交通大学学报．Vol．33．No．5．1995．5
23 姚玉荣．利用小波和矩进行基于形状的图像检索．中国图像图形学 Vol．5(A)，No．3．2000．3
24 边肇祺 模式识别．第二版．北京：清华大学出版社，2000．1
25 李松毅，周瑾丹，张惠 基于矩的数字图像多边形逼近方法．计算机学报 第24卷第4期 2001年4月
26 Chang Pei. Color Image Processing by Using Binaxy Quaternion-Moment-Presering Thresholding Technique. IEEE Trans on Image Processing Vol.8.No.5. 1999
27 Levin D N, Pelizzari C A, Chen G T Y, et al, Retrospective geometric correlation of MR, CT, and PET images[J]. Radiology, 1988,169:817-823
28 Alpert N M, Bradshaw J F, Kenedy D, et al. The pricipal transformation : a method for image registration[J] JNM, 1990,31:1717-1722
29 Bookstein F L.Principal warps:thin-plate splines and the decomposition of deformations[J] IEEE Trans Pattern Anal Mach Intell, 1989,11:567-585
30 Shu HZ ,Luo L M, Yu W X,et al. A new fast method for computeting Legendre moments[J]. Pattern Recognition 200,33:341-348.
31 王秋让，赵荣椿，郑南宁 一种航空照片中的小目标识别方法 中国图像图形学报 第3卷第12期 1998年12月
32 Sahoo P, Wuukins C, yeager J. Threshold Selection Using Renyi's Entropy .Pattern Recognition, 1997,30(1)
33 Cem Yuccer, Kem al O flazer, A Rotation, Scaling, and T ranslation Invariant Pattern Classification System. Pattern Recognition,26,687～710
34 A.K.Jain and B.Chandrasekaran,"Dimensionality and Sample Size Considerations in Pattern Recognition Practice," Handbook of Statistics (Vol.2,pp.835-855),North Holland Publishing Company,1982
35 G, Cybenko,Continuous Valued Neural Network with Two Hidden Layers Are Sufficient, Technical Report, Dept.of Computer Science. Tufts University, Bedford, MA, 1988
36 L. Ott and W.Mendenhall, Understanding Statistics(5th ed), PWS-KENT , Boston, 1990.
37 Kenneth R. Castleman Digital Image Processing Prentice-Hall International, Inc 1998
38 Donald O.Tanguay, Jr .Hidden Markov Models for Gesture Recognition , Donald Tanguay, 1995
39 Chan Wah Ng, Surendra Ranganath Real-time gesture recognition system and application, Image and Vision Computing 20(2002)993-1007
40 任海兵，祝远新，林学阂 基于视觉手势识别的研究-综述 电子学报 第2期 2000年2月
41 方志刚 计算机手势输入及其在人机交互技术中的应用 小型微型计算机系统 第20卷 第6期限 1999年6月
42 甘俊英，张有为 基于不变矩特征和神经网络的人脸识别模型
43 杨静 丘江 王岩飞 线性不变矩及其在图像识别中的应用算法研究 光子学报 第32卷第3期