交互式分割及目标定位在视频场景与虚拟图像融合中的研究
|
摘要
图像分割和目标跟踪技术是计算机视觉领域的重要的研究方向,在模式识别、图像编辑、智能监控等领域有着广泛的应用。我们把图像分割和目标跟踪应用到视频真实场景与虚拟图像融合问题之中,研究并改进了相关算法,并针对该问题设计了一套解决方案。论文以实现逼真的图像融合效果为目的,提出了以交互式分割获得场景中背景目标的边界,以8参数透视模型为基础估计目标的透视变换参数,并以高分辨率重采样进行图像融合的算法思想。
在交互式分割环节,论文采用了基于最大流的Graph Cuts方法,讨论了通过分水岭变换、区域合并进行预分割,在预分割区域上构造能量函数,并且在用最大流算法最小化能量函数时,提出了通过减小待分割图像的有效分割范围来降低流网络的复杂度,以加快Graph Cuts分割速度的方法。
在参数估计环节,论文详细分析了在灰度不变性条件下建立代价函数,并运用Levenberg-Marquardt算法来估计透视参数的方法,并在此基础上提出了一种减少迭代计算量的LM改进算法。同时本文采用由粗至精的图像金字塔数据结构,在每层金字塔上对透视参数逐步求精,进一步减少了计算量并避免了误匹配。除此之外,为了保证在目标快速运动时算法的鲁棒性,本文提出了将卡尔曼滤波、Mean Shift和基于透视模型的参数估计相结合的目标跟踪算法,实验表明此算法可以在目标快速运动时仍能准确估计透视参数。
最后本文采用了一种将输出图像坐标变换到输入图像坐标,并且在高分辨率输入图像上重采样的图像融合方法。并且在融合时考虑到光照影响,对输入图像生成阴影并控制阴影的显示范围,实现了更逼真的融合效果。
Image segmentation and object tracking are important research task in the field of computer vision, they are wildly used in areas such as pattern recognition, image editing, intelligent surveillance. We use image segmentation and object tracking to realize the image registration of true video scene and the virtual image. For more vivid effect, we propose a whole completed solution with three steps such as using interactive segmentation to gain the boundary of object, estimating 8-parameter of the perspective transformation, image registration by high resolution resampling.
In the step of interactive segmentation, this paper expatiates and improves the Graph Cuts method based on Max-flow algorithm. By watershed and region merging, also some method of reducing the valid segmentation range of image, we construct the flow network less complicated for accelerating the algorithm.
In the step of parameter estimation, this paper labor the method of using the Levenberg-Marquardt algorithm to estimate transformation parameter by minimizing the cost function based on brightness invariant, meanwhile the "coarse-to-fine" strategy is adopted for computational reason and avoiding incorrect matching. After that, we introduce an improved LM algorithm to reduce the computational magnitude. Besides that, we propose a novel object tracking algorithm based on Kalman filter, Mean Shift and perspective transformation. Experimental results show that the algorithm is reliable and efficient even in the circumstance of fast object moving.
In the end this paper adopt a novel image registration method based on high resolution resampling, meanwhile consider the effect of light, we generate the shadow for the virtual input image and make it in the area of the object's boundary which obtained in first step, then the final effect of image registration may as real as possible.
在交互式分割环节,论文采用了基于最大流的Graph Cuts方法,讨论了通过分水岭变换、区域合并进行预分割,在预分割区域上构造能量函数,并且在用最大流算法最小化能量函数时,提出了通过减小待分割图像的有效分割范围来降低流网络的复杂度,以加快Graph Cuts分割速度的方法。
在参数估计环节,论文详细分析了在灰度不变性条件下建立代价函数,并运用Levenberg-Marquardt算法来估计透视参数的方法,并在此基础上提出了一种减少迭代计算量的LM改进算法。同时本文采用由粗至精的图像金字塔数据结构,在每层金字塔上对透视参数逐步求精,进一步减少了计算量并避免了误匹配。除此之外,为了保证在目标快速运动时算法的鲁棒性,本文提出了将卡尔曼滤波、Mean Shift和基于透视模型的参数估计相结合的目标跟踪算法,实验表明此算法可以在目标快速运动时仍能准确估计透视参数。
最后本文采用了一种将输出图像坐标变换到输入图像坐标,并且在高分辨率输入图像上重采样的图像融合方法。并且在融合时考虑到光照影响,对输入图像生成阴影并控制阴影的显示范围,实现了更逼真的融合效果。
Image segmentation and object tracking are important research task in the field of computer vision, they are wildly used in areas such as pattern recognition, image editing, intelligent surveillance. We use image segmentation and object tracking to realize the image registration of true video scene and the virtual image. For more vivid effect, we propose a whole completed solution with three steps such as using interactive segmentation to gain the boundary of object, estimating 8-parameter of the perspective transformation, image registration by high resolution resampling.
In the step of interactive segmentation, this paper expatiates and improves the Graph Cuts method based on Max-flow algorithm. By watershed and region merging, also some method of reducing the valid segmentation range of image, we construct the flow network less complicated for accelerating the algorithm.
In the step of parameter estimation, this paper labor the method of using the Levenberg-Marquardt algorithm to estimate transformation parameter by minimizing the cost function based on brightness invariant, meanwhile the "coarse-to-fine" strategy is adopted for computational reason and avoiding incorrect matching. After that, we introduce an improved LM algorithm to reduce the computational magnitude. Besides that, we propose a novel object tracking algorithm based on Kalman filter, Mean Shift and perspective transformation. Experimental results show that the algorithm is reliable and efficient even in the circumstance of fast object moving.
In the end this paper adopt a novel image registration method based on high resolution resampling, meanwhile consider the effect of light, we generate the shadow for the virtual input image and make it in the area of the object's boundary which obtained in first step, then the final effect of image registration may as real as possible.
引文
[1]廖志军.交互式图像分割算法.中国科学院研究生院硕士学位论文,2006,10-15
[2]赵荣椿,迟耀斌,朱重光.图像分割技术进展.中国体视学与图像分析,1998,3(2):121-128
[3]李旭超,朱善安.图像分割中的马尔可夫随机场方法综述.中国图象图形学报,2007:12(5):789-499
[4]Sonka M,Hlavac V,Boyle R.图像处理、分析与机器视觉.第二版.人民邮电出版社
[5]Forsyth D A,Ponce J.计算机视觉-一种现代方法。电子工业出版社,2004
[6]Malik J,Belongie S,Leung T,Shi J.Contour and texture analysis for image segmentation.International Journal of Computer Vision,2001,43(1):7-27
[7]孙涵,任明武.基于链码的分水岭变换算法.中国图象图形学报,2004,9(9):1025-1031
[8]刘颖.基于分层模板匹配的运动目标跟踪算法研究.吉林大学硕士学位论文,2007:21-29
[9]于丹,汤井田,徐大宏,罗东礼.基于最大流的交互式目标提取算法.计算机工程与应用,2007,43(5):246-248
[10]Rother C,Kolmogorov V,Blake A."GrabCut" -Interactive Foreground Extraction using Iterated Graph Cuts.ACM Transactions on Graphics,Special Issue:Proceeding of the 2004.SIGGRAPH Conference,23(3):309-314
[11]杨晓峰。基于仿射变换模型的图像目标定位跟踪方法。华中科技大学硕士学位论文,2005.14-32
[12]曾文锋,李树山,王江安.基于仿射变换模型的图像配准中的平移、旋转和缩放.红外与激光工程,2001,30(1):18-21
[13]曹邦兴.LM算法在地下水动态预测中的应用研究。广西水利水电,2007(3):4-6
[14]金勇俊,李言俊,张科.一种透视变换图像金字塔匹配改进算法.计算机工程与应用,2007,43(24):78-81
[15]朱胜利.Mean Shift及相关算法在视频跟踪中的研究.浙江大学博士学位论文,2006:13-41
[16]邓红梅,吴四夫.基于相位相关算法的研究与实现.信息技术,2005,(4):19-21
[17]肖秦琨,雷斌.基于卡尔曼滤波的摄像头目标跟踪.西安工业学院学报,2006,26(1):1-4
[18]Carpenter J,Clifford P,Fearnhead P.An improved particle filter for non-linear problems, IEEE Proceedings on Radar and Sonar Navigation,1999,146:2-7
[19]陈艳琴 罗大庸.基于Kalman滤波和Mean Shift算法的人眼实时跟踪.模式识别与人工智能,2004,12(2):173-177
[20]刘翔.基于光流场算法的运动估计方法研究.信号与信息处理,2006,36(4):17-20
[21]范春宇.视频跟踪技术研究.西安电子科技大学硕士学位论文,2007,14-21
[22]Mortensen E.N,Barrett W.A.Interactive segmentation with intelligent scissors.Graphical Models and Image Processing,60(5):349-384
[23]Chuang Y.Y,Curless B,Salesin D.H,Szeliski R.A Bayesian approach to digital matting.Proceeding of the IEEE Computer Vision and Pattern Recognition,2001,264-271
[24].Ruzon M,Tomasi C.Alpha estimation in natural images.Proceeding of IEEE Computer Vision and Pattern Recognition,2000,18-25
[25]Greig D,Porteous B,Seheult A.Exact maximum a posterior estimation for binary images.Journal of the Royal Statistical Society,Series B,1989,52(2):271-279
[26]Blake A,Rother C,Brown M,Perez P,and Torr P.Interactive Image Segmentation Using an Adaptive GMMRF Model.Proceeding ECCV,2004,428-441
[27]王飞,罗东礼,汤井田,徐大宏,杨晓利.一种基于图的交互式目标分割算法.计算机工程与应用,2006,24:65-68
[28]Cormen T H,Liserson C E,Rivest R L,Stein R C.算法导论.第2版。机械工业出版社,2006
[29]Kolmogorov V,Zabih R.What Energy Functions Can Be Minimized via Graph Cuts?IEEE Computer Society,2004,26(2):147-159
[30]Kolmogorov V.Graph based algorithms for scene reconstruction from two or more views.A Dissertation Presented to the Faculty of the Graduate School of Comell University,2004,97-113
[31]Boykov Y,Kolmogorov V.An Experimental Comparison of Min-Cut/Max-Flow Algorithms for Energy Minimization in Vision.IEEE transactions on pattern analysis and machine intelligence.2004,26(9):1124-1137
[32]Kostas H,Serafim N E,Nicos M.Hybrid Image Segmentation Using Watersheds and Fast Region Merging.IEEE Transaction on image processing,1998,7(12):1684-1699
[33]王秋瑞.序列图像的接图实现.哈尔滨工业大学硕士学位论文,2006,32-25
[34]章权兵,罗斌,韦穗,杨尚骏.基于仿射变换模型的图像特征点集配准方法研究.中国图象图形学报,2003,8(10):1121-1125
[35]巴彤.齐次坐标透视变换的几何意义.电力学刊,1990,5(1):13-23
[36]杨晓峰,张桂林.一种基于仿射变换模型的目标跟踪算法.计算机与数字工程,2005, 3(12):30-34
[37]庞博.图像插值技术及其在超分辨率图像重建中的应用.合肥工业大学硕士学位论文,2005,23-25
[38]伏燕军,杨坤涛,邹文栋,何兴道.基于Levenberg-Marquardt算法的图像拼接.激光杂志,2007,28(5):46-47
[39]Yizong Cheng.Mean Shift mode Seeking and Clustering.IEEE Transactions on Pattem analysis and Machine Intelligence,1995,17(8):6-10
[40]李涛.基于Mean Shift算法和Particle Filter算法的目标跟踪.南京理工大学硕士学位论文,2007,6-13
[41]Manku G S,Jain P,Aggarwal A,Kumar L.Object Tracking using Affine Structure for Point Correspondences.IEEE,1997,704-708
[42]Comaniciu D,Ramesh V,Meer P.Kenel-based object tracking.IEEE Transactions.on Pattem Analysis and Machine Intelligenee,2003,25(2):564-577
[43]Comaniciu D,Ramesh V,Meer P.Real-Time Tracking of Non-Rigid Objects using Mean Shift.IEEE,2000
[44]张旭光,赵恩良,王延杰.基于Mean-shift的灰度目标跟踪新算法.光学技术,2007,33(2):226-228
[45]常发亮,刘雪,王华杰.基于均值漂移与卡尔曼滤波的目标跟踪算法。计算机工程与应用,2007,43(12):50-52
[46]Boykov Y Y,Jolly M P.Interactive Graph Cuts for Optimal Boundary & Region Segmentation of Objects in N-D Images.IEEE,2001,105-112
[47]Boykov Y,Kolmogorov V.Computing Geodesics and Minimal Surfaces via Graph Cuts.International conference on computer vision,2003,1-8
[48]杜啸晓,杨新,施鹏飞.一种新的基于区域和边界的图象分割方法.中国图象图形学报,2001,6(8):755-759
[49]王兵,赵荣椿,薛笑荣.视频序列中运动参数估计方法研究。计算机应用研究,2004.8:43-45
[50]宋建中.粒子滤波跟踪方法研究.中国科学院研究生院博士学位论文,2004,13-17
[51]张立华,徐文立.基于p2-不变量的透视变换下的点模式匹配方法。中国图象图形学报,2000,5(11):949-952
[2]赵荣椿,迟耀斌,朱重光.图像分割技术进展.中国体视学与图像分析,1998,3(2):121-128
[3]李旭超,朱善安.图像分割中的马尔可夫随机场方法综述.中国图象图形学报,2007:12(5):789-499
[4]Sonka M,Hlavac V,Boyle R.图像处理、分析与机器视觉.第二版.人民邮电出版社
[5]Forsyth D A,Ponce J.计算机视觉-一种现代方法。电子工业出版社,2004
[6]Malik J,Belongie S,Leung T,Shi J.Contour and texture analysis for image segmentation.International Journal of Computer Vision,2001,43(1):7-27
[7]孙涵,任明武.基于链码的分水岭变换算法.中国图象图形学报,2004,9(9):1025-1031
[8]刘颖.基于分层模板匹配的运动目标跟踪算法研究.吉林大学硕士学位论文,2007:21-29
[9]于丹,汤井田,徐大宏,罗东礼.基于最大流的交互式目标提取算法.计算机工程与应用,2007,43(5):246-248
[10]Rother C,Kolmogorov V,Blake A."GrabCut" -Interactive Foreground Extraction using Iterated Graph Cuts.ACM Transactions on Graphics,Special Issue:Proceeding of the 2004.SIGGRAPH Conference,23(3):309-314
[11]杨晓峰。基于仿射变换模型的图像目标定位跟踪方法。华中科技大学硕士学位论文,2005.14-32
[12]曾文锋,李树山,王江安.基于仿射变换模型的图像配准中的平移、旋转和缩放.红外与激光工程,2001,30(1):18-21
[13]曹邦兴.LM算法在地下水动态预测中的应用研究。广西水利水电,2007(3):4-6
[14]金勇俊,李言俊,张科.一种透视变换图像金字塔匹配改进算法.计算机工程与应用,2007,43(24):78-81
[15]朱胜利.Mean Shift及相关算法在视频跟踪中的研究.浙江大学博士学位论文,2006:13-41
[16]邓红梅,吴四夫.基于相位相关算法的研究与实现.信息技术,2005,(4):19-21
[17]肖秦琨,雷斌.基于卡尔曼滤波的摄像头目标跟踪.西安工业学院学报,2006,26(1):1-4
[18]Carpenter J,Clifford P,Fearnhead P.An improved particle filter for non-linear problems, IEEE Proceedings on Radar and Sonar Navigation,1999,146:2-7
[19]陈艳琴 罗大庸.基于Kalman滤波和Mean Shift算法的人眼实时跟踪.模式识别与人工智能,2004,12(2):173-177
[20]刘翔.基于光流场算法的运动估计方法研究.信号与信息处理,2006,36(4):17-20
[21]范春宇.视频跟踪技术研究.西安电子科技大学硕士学位论文,2007,14-21
[22]Mortensen E.N,Barrett W.A.Interactive segmentation with intelligent scissors.Graphical Models and Image Processing,60(5):349-384
[23]Chuang Y.Y,Curless B,Salesin D.H,Szeliski R.A Bayesian approach to digital matting.Proceeding of the IEEE Computer Vision and Pattern Recognition,2001,264-271
[24].Ruzon M,Tomasi C.Alpha estimation in natural images.Proceeding of IEEE Computer Vision and Pattern Recognition,2000,18-25
[25]Greig D,Porteous B,Seheult A.Exact maximum a posterior estimation for binary images.Journal of the Royal Statistical Society,Series B,1989,52(2):271-279
[26]Blake A,Rother C,Brown M,Perez P,and Torr P.Interactive Image Segmentation Using an Adaptive GMMRF Model.Proceeding ECCV,2004,428-441
[27]王飞,罗东礼,汤井田,徐大宏,杨晓利.一种基于图的交互式目标分割算法.计算机工程与应用,2006,24:65-68
[28]Cormen T H,Liserson C E,Rivest R L,Stein R C.算法导论.第2版。机械工业出版社,2006
[29]Kolmogorov V,Zabih R.What Energy Functions Can Be Minimized via Graph Cuts?IEEE Computer Society,2004,26(2):147-159
[30]Kolmogorov V.Graph based algorithms for scene reconstruction from two or more views.A Dissertation Presented to the Faculty of the Graduate School of Comell University,2004,97-113
[31]Boykov Y,Kolmogorov V.An Experimental Comparison of Min-Cut/Max-Flow Algorithms for Energy Minimization in Vision.IEEE transactions on pattern analysis and machine intelligence.2004,26(9):1124-1137
[32]Kostas H,Serafim N E,Nicos M.Hybrid Image Segmentation Using Watersheds and Fast Region Merging.IEEE Transaction on image processing,1998,7(12):1684-1699
[33]王秋瑞.序列图像的接图实现.哈尔滨工业大学硕士学位论文,2006,32-25
[34]章权兵,罗斌,韦穗,杨尚骏.基于仿射变换模型的图像特征点集配准方法研究.中国图象图形学报,2003,8(10):1121-1125
[35]巴彤.齐次坐标透视变换的几何意义.电力学刊,1990,5(1):13-23
[36]杨晓峰,张桂林.一种基于仿射变换模型的目标跟踪算法.计算机与数字工程,2005, 3(12):30-34
[37]庞博.图像插值技术及其在超分辨率图像重建中的应用.合肥工业大学硕士学位论文,2005,23-25
[38]伏燕军,杨坤涛,邹文栋,何兴道.基于Levenberg-Marquardt算法的图像拼接.激光杂志,2007,28(5):46-47
[39]Yizong Cheng.Mean Shift mode Seeking and Clustering.IEEE Transactions on Pattem analysis and Machine Intelligence,1995,17(8):6-10
[40]李涛.基于Mean Shift算法和Particle Filter算法的目标跟踪.南京理工大学硕士学位论文,2007,6-13
[41]Manku G S,Jain P,Aggarwal A,Kumar L.Object Tracking using Affine Structure for Point Correspondences.IEEE,1997,704-708
[42]Comaniciu D,Ramesh V,Meer P.Kenel-based object tracking.IEEE Transactions.on Pattem Analysis and Machine Intelligenee,2003,25(2):564-577
[43]Comaniciu D,Ramesh V,Meer P.Real-Time Tracking of Non-Rigid Objects using Mean Shift.IEEE,2000
[44]张旭光,赵恩良,王延杰.基于Mean-shift的灰度目标跟踪新算法.光学技术,2007,33(2):226-228
[45]常发亮,刘雪,王华杰.基于均值漂移与卡尔曼滤波的目标跟踪算法。计算机工程与应用,2007,43(12):50-52
[46]Boykov Y Y,Jolly M P.Interactive Graph Cuts for Optimal Boundary & Region Segmentation of Objects in N-D Images.IEEE,2001,105-112
[47]Boykov Y,Kolmogorov V.Computing Geodesics and Minimal Surfaces via Graph Cuts.International conference on computer vision,2003,1-8
[48]杜啸晓,杨新,施鹏飞.一种新的基于区域和边界的图象分割方法.中国图象图形学报,2001,6(8):755-759
[49]王兵,赵荣椿,薛笑荣.视频序列中运动参数估计方法研究。计算机应用研究,2004.8:43-45
[50]宋建中.粒子滤波跟踪方法研究.中国科学院研究生院博士学位论文,2004,13-17
[51]张立华,徐文立.基于p2-不变量的透视变换下的点模式匹配方法。中国图象图形学报,2000,5(11):949-952
© 2004-2018 中国地质图书馆版权所有 京ICP备05064691号 京公网安备11010802017129号 地址：北京市海淀区学院路29号 邮编：100083 电话：办公室：(+86 10)66554848；文献借阅、咨询服务、科技查新：66554700 |