活动轮廓模型及在人脸部轮廓检测中的应用
|
摘要
近些年来,计算机图像的检测和分割在图像处理中起着越来越重要的作用。由于物体形状的多样性以及图像质量的不同,传统的图像检测和分割方法,如边缘检测、阈值方法等,用来提取轮廓边界时可能完全失效,有时必须在分割结果中去除无效的对象边界。活动轮廓模型是利用能量最小化函数在复杂图象中寻找目标边缘的一种新方法,在国内外都受到了广泛关注。本文对活动轮廓模型进行了详细的研究,并将研究结果应用于人脸的轮廓检测,为一下步人脸识别奠定了基础。
本文首先介绍了在计算机中图像预处理和边缘检测常用的一些方法,比较了这些算法的优缺点。然后后引入了活动轮廓模型,并深入研究了活动轮廓模型进行边缘检测的原理,给出了它的物理解释和实现方法。在基于活动轮廓模型的图像边缘检测中,搜索范围小和深度凹陷区域的分割是难点,为此本文对活动轮廓模型不同的外力场进行了研究,并对它们的轮廓收敛性能进行了分析,在此基础上提出了一个新的改进模型;它先采用距离snake使初始轮廓逼近目标物体边缘,然后结合NGVF使收敛进入目标物体的凹陷部分,最后应用有限差分实现了改进模型的数值计算;实验表明该模型具有较大的捕获区且能快速收敛到物体边缘的凹陷部分,提取出感兴趣目标的轮廓。
接下来针对梯度矢量流活动轮廓模型同样存在着滤除噪声和准确定位的矛盾,提出利用三次B样条小波与GVF活动轮廓模型进行融合,得到一种去噪功能好的多尺度活动轮廓方法。它先在大尺度下对图像进行滤波,用GVF Snake搜索到一个抗噪性能好定位不太准的目标轮廓;然后逐步减少尺度,并在前一次的基础上进行GVFSnake搜索,最后得到搜索轮廓。
文章最后用改进的模型检测出人脸和嘴唇的轮廓,它异于传统的检测方法,具有去噪功能好、边缘轮廓线连续的特点。全文的实验在MATLAB的环境下实现,得到了各种不同的边缘提取的效果。实验结果表明,本文算法在目标轮廓连续边缘检测过程中,不仅降低了对初始位置的敏感性,还提高了搜索效率,并且对真实图象的边缘提取也有较好的效果。
Recently, Digital image detection and segmentation have been playing an increasingly important role in image Processing. Due to both the tremendous variability of object shapes and the variation in image quality, when applying classical image detection and segmentation techniques such as edge detection and threshold etc, these techniques will be either failed completely or required some kind of post-processing step to remove invalid object boundaries in the detection and segmentation result. Active contour model is a new method which can achieve the object boundaries with the energy minimum function and have been paid attention extensively at home and overseas. This dissertation studied the active contour models at large and used them to extract object boundaries in face detection, which is the base of face recognitions.
At first, some common methods of image pretreatment and edge detection were introduced in this dissertation, and then their strongpoint and shortcoming were compared. After discussing the parametric active contour model theory thoroughly, a reality-based interpretation of elasticity theory was given. Then the active contour models were recommended in image detection. The active contour is poor to converge on concave boundary and capture a narrow range, which are difficult to locate the object. Several exterior force fields of the active contour models were studied, which convergent features were compared then a new improved method was presented. This model includes tow stages. The first we made use of distance snake near to the object boundaries, while in the second the NGVF would drive the contour into the concave region. The finite difference method is applied to do numerical implementations of the new model. The experimental results showed the new model has a large capture range, can move a snake into the boundary concaves and is able to obtain the interested object contour precisely.
Then there is a conflict among noise filtration and precise location when use the gradient vector flow active contour model to detect the edges. To filtrate the noise and get an excellent method, this thesis integrates cubic B-spline wavelet and GVF snake. First, we filtrated the image noise with large scale, then drive the GVF snake to an aimed contour that noise filtration well but location bad. Second, reducing the scale, and take the former contour as the GVF snake to drive.
At last, this thesis uses the new improved active contour model to detect the face and tip contour. This method different from traditional detection methods, and have advantages in noise filtration and edge connection. The whole experiments were implemented using MATLAB code, and some results of object boundaries in image detection and segmentation are prominent. That indicated, when object boundaries were continuous, this new algorithm could cut down the sensitivity of initialized location and enhance the efficiency of convergent speed. Meanwhile the new algorithm would be applied to the real image and the effect is well.
本文首先介绍了在计算机中图像预处理和边缘检测常用的一些方法,比较了这些算法的优缺点。然后后引入了活动轮廓模型,并深入研究了活动轮廓模型进行边缘检测的原理,给出了它的物理解释和实现方法。在基于活动轮廓模型的图像边缘检测中,搜索范围小和深度凹陷区域的分割是难点,为此本文对活动轮廓模型不同的外力场进行了研究,并对它们的轮廓收敛性能进行了分析,在此基础上提出了一个新的改进模型;它先采用距离snake使初始轮廓逼近目标物体边缘,然后结合NGVF使收敛进入目标物体的凹陷部分,最后应用有限差分实现了改进模型的数值计算;实验表明该模型具有较大的捕获区且能快速收敛到物体边缘的凹陷部分,提取出感兴趣目标的轮廓。
接下来针对梯度矢量流活动轮廓模型同样存在着滤除噪声和准确定位的矛盾,提出利用三次B样条小波与GVF活动轮廓模型进行融合,得到一种去噪功能好的多尺度活动轮廓方法。它先在大尺度下对图像进行滤波,用GVF Snake搜索到一个抗噪性能好定位不太准的目标轮廓;然后逐步减少尺度,并在前一次的基础上进行GVFSnake搜索,最后得到搜索轮廓。
文章最后用改进的模型检测出人脸和嘴唇的轮廓,它异于传统的检测方法,具有去噪功能好、边缘轮廓线连续的特点。全文的实验在MATLAB的环境下实现,得到了各种不同的边缘提取的效果。实验结果表明,本文算法在目标轮廓连续边缘检测过程中,不仅降低了对初始位置的敏感性,还提高了搜索效率,并且对真实图象的边缘提取也有较好的效果。
Recently, Digital image detection and segmentation have been playing an increasingly important role in image Processing. Due to both the tremendous variability of object shapes and the variation in image quality, when applying classical image detection and segmentation techniques such as edge detection and threshold etc, these techniques will be either failed completely or required some kind of post-processing step to remove invalid object boundaries in the detection and segmentation result. Active contour model is a new method which can achieve the object boundaries with the energy minimum function and have been paid attention extensively at home and overseas. This dissertation studied the active contour models at large and used them to extract object boundaries in face detection, which is the base of face recognitions.
At first, some common methods of image pretreatment and edge detection were introduced in this dissertation, and then their strongpoint and shortcoming were compared. After discussing the parametric active contour model theory thoroughly, a reality-based interpretation of elasticity theory was given. Then the active contour models were recommended in image detection. The active contour is poor to converge on concave boundary and capture a narrow range, which are difficult to locate the object. Several exterior force fields of the active contour models were studied, which convergent features were compared then a new improved method was presented. This model includes tow stages. The first we made use of distance snake near to the object boundaries, while in the second the NGVF would drive the contour into the concave region. The finite difference method is applied to do numerical implementations of the new model. The experimental results showed the new model has a large capture range, can move a snake into the boundary concaves and is able to obtain the interested object contour precisely.
Then there is a conflict among noise filtration and precise location when use the gradient vector flow active contour model to detect the edges. To filtrate the noise and get an excellent method, this thesis integrates cubic B-spline wavelet and GVF snake. First, we filtrated the image noise with large scale, then drive the GVF snake to an aimed contour that noise filtration well but location bad. Second, reducing the scale, and take the former contour as the GVF snake to drive.
At last, this thesis uses the new improved active contour model to detect the face and tip contour. This method different from traditional detection methods, and have advantages in noise filtration and edge connection. The whole experiments were implemented using MATLAB code, and some results of object boundaries in image detection and segmentation are prominent. That indicated, when object boundaries were continuous, this new algorithm could cut down the sensitivity of initialized location and enhance the efficiency of convergent speed. Meanwhile the new algorithm would be applied to the real image and the effect is well.
引文
[1]Valentin D,bdi H,O' Toole A J.Connection is Models of Face Processing.[J]Pattern Recognition.1994,27:1209-1230.
[2]Gunn S R,Nixon M S.Global and local contours for head boundary extraction[J].International Journal of Computer Vision,1998.30(1):43-54.
[3]Kass M,Withkin A,Terzopoulos D.Snakes:Active contour models.Proceeding of 1~(st)International Conference on Computer Vision[C],London,1987:321-331.
[4]W.R.Pratt,Digital image processing[J],NEWYORK:wiley,1997
[5]J.F.Canny,A computational approach to edge detection[J],IEEE Trans on PAM1,1985:8(6):679-698
[6]S.Mallat,Multifrequency channel decompositions of images and wavelet models[J],IEEE Transaction in Acoustic Speech and Signal Processing,vol.1989(37) 2091-2110
[7]S.Mallat,S.Zhong,Characterization of signals from multiscale edges[J],Pattern Analysis and Machine Intelligence,IEEE Transactions on Pattern Analysis and Machine Intelligence,1992,14(7):710-732.
[8]佚名,基于数学形态学的RTR图像边缘检[J],中国图形图像网,2006
[9]Yateen Chitre,Atam P.D hawan,M-Band Wavelet Discrimination of Natural Textures[J].Pattern Reconition.1999,(32):773-78
[10]R.M.Rao,J.S.Bundonis,H.H.Szu,Three-Scale Wavele[J]t Transforms,SPIE,1998,3(3):91-32
[11]倪虹霞.基于小波变换和中值滤波的遥感图像预处理,[硕士学位论文]吉林大学,2005
[12]杜会斌.基于小波变换的静止图像编码技术研究,[硕士学位论文]山东大学,2006
[13]秦安.医学图像感兴趣区的内嵌小波编码.[硕士学位论文]第一军医大学,2004
[14]李培华,张田文.主动轮廓线模型(蛇模型)综述[J].软件学报,2000,11(6):751-757.
[15]Caselles V,Catte F,CoilT,et.al.A geometric model for active contours in image processing[J]. Numeri7cal Mathematic, 1993, 66(1): 1-31.
[16] Caselles V, Kimmel R., Sapiro G. Geodesic active contours[J]. Internat.J. Comput. 1997,10(1): 61—79.
[17] Cohen L. D, Cohen I, Finite-element methods for active contour models and balloons for 2-D and 3-D images[J]. IEEE Trans. Pattern Anal.Machine Intell. 1993, (15):1131-1147.
[18]XuC, Prince J. L. Snakes, shapes, and gradient vector flow[J]. IEEE Trans.Image Process. 1998, (7) : 359-369
[19] Zeyun Yu, Bajaj C. Image segmentation using gradient vector diffusion and region merging[J]. In Proceedings: 16th International Conference on Pattern Recognition, 2002, (5): 2941—2944.
[20] AminiA A, Weymouth T E, Jain R C. Using dynamic programming for solving variational problems in vision[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 1990, 12(9):855—867
[21] Seok-Woo Jang. Shaking snakes using color edge for contour extraction[J]. Proceedings: International Conference on Image Processing, 2002, 9(2): 817—820.
[22] Osher S, Sethian J. A. Fronts propagating with curvaturen dependent speed:Algorithms based on Hamilton - Jacobi formulations[J]. Comput. Phys.1988, (79):12—49.
[23] Sethian J A. Level Set methods and Fast M arching Methods: Evolving interfaces in computational geometry, fluid mechanics, computer vision,and materials science[M]. Cambridge. United Kingdom : Cam bridge University Press, 1999.
[24] V Caselles, R Kirmnel, G Sapiro. Geodesic active contours[J]. Fifth International Conference on Computer Vision, 1995, (3):694—699
[25] Mansoufi A. R, Konrad J. Multiple motion segmentation with level sets[J].IEEE Transactions on Image Processing, 2003,2(12): 201—220.
[26] Chan T F. Active contours without edges[J]. IEEE Transactions on Image Processing, 2001, 2(10):266—277
[27] L. D. Cohen, I. Cohen, Finite-element methods for active contour models and balloons for 2-D and 3-D images.[J]IEEE Trans.Patt.Anal.Mach.Intell,1993,15(11):1131-1147
[28]H.Delingette,Simplex meshes:a general representation for 3 D shape reconstruction.[J]Tech.Rep.TR2214,INRIA,Sophia-Antipolis,France,1994.
[29]D.MacDonald,D.Avis,and A.C.Evans.Multiple surface identification and matching in magnetic resonance images.[J]Visualization in Biomedical Computing.1994,(2359):160-169
[30]Marr D c,Hidreth E.Theory of edge diction.[J]ProcR Soc,1980,B207:187-217
[31]A.K.Jain,Fundamentals of Digital Image Processing.[J]Englewood Clis,N J:Prentice-Hall,1989.
[32]Ning Jifeng et al,NGVF:An imporved external force field for active contour mode..[J]Pattern Recogintion Letters(2006),doi:10.1016/j.patrec.2006.06.014.
[33]Canny J.A computational approach to edge detection.[J]Proc R Soc,1980,B207:187-217
[34]姜杭毅,蔡元龙.用于边缘检测的Laplace-样条算子.[J]中国科学.A辑,1989,(10):1113-1120
[35]Mallat S,Huang W L.Singularity detection and processing with wavelets.[J]IEEE Trans,IT-38,1992,2:617-643
[36]Unser M.Aldroubi A,Eden M.On the asymptotic convergence of B-spline wavelets to Gobor functions.[J]IEEE Trans,IT-38,1992,2:864-872
[37]Rafael C.Gonzalez,Richard E.Woods.Digital Image Processing Second Edition[M].电子工业出版社,2003,3
[38]Xie X,Mirmehdi M.RAGS:Region-Aided Geometric Snake.[J]IEEE Transactions on Image Processing.2004,5(13):640-652
[39]P.M.Morse and H.Feshbach,Methods of Theoretical Physics.New York:McGraw-Hill,1953
[40]李天庆,张毅,刘志,胡东成.Snake模型综述[J].计算机工程.2005,31(9):1-3.
[2]Gunn S R,Nixon M S.Global and local contours for head boundary extraction[J].International Journal of Computer Vision,1998.30(1):43-54.
[3]Kass M,Withkin A,Terzopoulos D.Snakes:Active contour models.Proceeding of 1~(st)International Conference on Computer Vision[C],London,1987:321-331.
[4]W.R.Pratt,Digital image processing[J],NEWYORK:wiley,1997
[5]J.F.Canny,A computational approach to edge detection[J],IEEE Trans on PAM1,1985:8(6):679-698
[6]S.Mallat,Multifrequency channel decompositions of images and wavelet models[J],IEEE Transaction in Acoustic Speech and Signal Processing,vol.1989(37) 2091-2110
[7]S.Mallat,S.Zhong,Characterization of signals from multiscale edges[J],Pattern Analysis and Machine Intelligence,IEEE Transactions on Pattern Analysis and Machine Intelligence,1992,14(7):710-732.
[8]佚名,基于数学形态学的RTR图像边缘检[J],中国图形图像网,2006
[9]Yateen Chitre,Atam P.D hawan,M-Band Wavelet Discrimination of Natural Textures[J].Pattern Reconition.1999,(32):773-78
[10]R.M.Rao,J.S.Bundonis,H.H.Szu,Three-Scale Wavele[J]t Transforms,SPIE,1998,3(3):91-32
[11]倪虹霞.基于小波变换和中值滤波的遥感图像预处理,[硕士学位论文]吉林大学,2005
[12]杜会斌.基于小波变换的静止图像编码技术研究,[硕士学位论文]山东大学,2006
[13]秦安.医学图像感兴趣区的内嵌小波编码.[硕士学位论文]第一军医大学,2004
[14]李培华,张田文.主动轮廓线模型(蛇模型)综述[J].软件学报,2000,11(6):751-757.
[15]Caselles V,Catte F,CoilT,et.al.A geometric model for active contours in image processing[J]. Numeri7cal Mathematic, 1993, 66(1): 1-31.
[16] Caselles V, Kimmel R., Sapiro G. Geodesic active contours[J]. Internat.J. Comput. 1997,10(1): 61—79.
[17] Cohen L. D, Cohen I, Finite-element methods for active contour models and balloons for 2-D and 3-D images[J]. IEEE Trans. Pattern Anal.Machine Intell. 1993, (15):1131-1147.
[18]XuC, Prince J. L. Snakes, shapes, and gradient vector flow[J]. IEEE Trans.Image Process. 1998, (7) : 359-369
[19] Zeyun Yu, Bajaj C. Image segmentation using gradient vector diffusion and region merging[J]. In Proceedings: 16th International Conference on Pattern Recognition, 2002, (5): 2941—2944.
[20] AminiA A, Weymouth T E, Jain R C. Using dynamic programming for solving variational problems in vision[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 1990, 12(9):855—867
[21] Seok-Woo Jang. Shaking snakes using color edge for contour extraction[J]. Proceedings: International Conference on Image Processing, 2002, 9(2): 817—820.
[22] Osher S, Sethian J. A. Fronts propagating with curvaturen dependent speed:Algorithms based on Hamilton - Jacobi formulations[J]. Comput. Phys.1988, (79):12—49.
[23] Sethian J A. Level Set methods and Fast M arching Methods: Evolving interfaces in computational geometry, fluid mechanics, computer vision,and materials science[M]. Cambridge. United Kingdom : Cam bridge University Press, 1999.
[24] V Caselles, R Kirmnel, G Sapiro. Geodesic active contours[J]. Fifth International Conference on Computer Vision, 1995, (3):694—699
[25] Mansoufi A. R, Konrad J. Multiple motion segmentation with level sets[J].IEEE Transactions on Image Processing, 2003,2(12): 201—220.
[26] Chan T F. Active contours without edges[J]. IEEE Transactions on Image Processing, 2001, 2(10):266—277
[27] L. D. Cohen, I. Cohen, Finite-element methods for active contour models and balloons for 2-D and 3-D images.[J]IEEE Trans.Patt.Anal.Mach.Intell,1993,15(11):1131-1147
[28]H.Delingette,Simplex meshes:a general representation for 3 D shape reconstruction.[J]Tech.Rep.TR2214,INRIA,Sophia-Antipolis,France,1994.
[29]D.MacDonald,D.Avis,and A.C.Evans.Multiple surface identification and matching in magnetic resonance images.[J]Visualization in Biomedical Computing.1994,(2359):160-169
[30]Marr D c,Hidreth E.Theory of edge diction.[J]ProcR Soc,1980,B207:187-217
[31]A.K.Jain,Fundamentals of Digital Image Processing.[J]Englewood Clis,N J:Prentice-Hall,1989.
[32]Ning Jifeng et al,NGVF:An imporved external force field for active contour mode..[J]Pattern Recogintion Letters(2006),doi:10.1016/j.patrec.2006.06.014.
[33]Canny J.A computational approach to edge detection.[J]Proc R Soc,1980,B207:187-217
[34]姜杭毅,蔡元龙.用于边缘检测的Laplace-样条算子.[J]中国科学.A辑,1989,(10):1113-1120
[35]Mallat S,Huang W L.Singularity detection and processing with wavelets.[J]IEEE Trans,IT-38,1992,2:617-643
[36]Unser M.Aldroubi A,Eden M.On the asymptotic convergence of B-spline wavelets to Gobor functions.[J]IEEE Trans,IT-38,1992,2:864-872
[37]Rafael C.Gonzalez,Richard E.Woods.Digital Image Processing Second Edition[M].电子工业出版社,2003,3
[38]Xie X,Mirmehdi M.RAGS:Region-Aided Geometric Snake.[J]IEEE Transactions on Image Processing.2004,5(13):640-652
[39]P.M.Morse and H.Feshbach,Methods of Theoretical Physics.New York:McGraw-Hill,1953
[40]李天庆,张毅,刘志,胡东成.Snake模型综述[J].计算机工程.2005,31(9):1-3.