超分辨率图像的重建
|
摘要
在遥感、医学和公安等应用领域中,经常需要高分辨率图像提供更多的细节和信息。在获取图像的过程中有许多因素,如运动,系统噪声等,会导致图像质量的下降。另一方面,由于成像设备本身存在很多限制,使图像的分辨率不能够满足应用的要求。单纯通过改进硬件系统的性能提高图像的分辨率,在技术方面很难实现突破,成本也会急剧增加。而基于信号处理的超分辨率图像重建技术既能有效提高图像的空间分辨率,又不需要提高成本支出。因此,超分辨率图像重建技术为提高图像分辨率提供了一条有效途径。
超分辨率图像重建技术是指将多幅变形、模糊、有噪、频谱混叠的低分辨率降质图像(或视频序列)融合估计出一幅高分辨率图像的技术。其核心的设计思想是：利用同一场景的多幅低分辨率图像间的相对互补信息,将它们融合到单幅高分辨率图像中,获得一幅高画质的图像。
本文在分析了图像降质模型的基础上,针对重建过程中的图像配准以及重建算法两个关键问题展开了研究。
考虑到低分辨率序列间的亚像素配准精度直接影响到超分辨率重建的效果,在图像配准这一关键环节,本文对泰勒级数法和相位相关模板匹配方法进行了实验和对比。针对相位相关法中的位移方向问题,本文对位移的方向性给出了一种合理的定义,实验证明由此定义得到的位移量能够满足配准精度的要求。
此外,本文对空域中常用的三种图像重建方法进行了深入研究和算法实现。在基于最大后验概率(MAP)的重建方法中,详细推导了退化过程的构造,解决了低分辨率图像产生过程中引入的散焦模糊问题,对正则化参数的作用进行了详细的实验分析。在基于凸集投影(POCS)的重建方法中,本文设计了用梯形低通滤波器来消除重建结果边缘振荡的现象,达到了较为理想的重建效果。最后对基于规范化卷积的重建方法进行了理解和实验测试。
In most electronic imaging applications, images with high resolution (HR) are often desired and required. In the process of acquiring the image, there are many factors lead to declines in image quality, such as sports, system noise. At the same time, the limit of imaging device makes the resolution of the image meet the application requirements. It is expensive and difficult to increase the current resolution level by improving hardware performance. Therefore, super-resolution image reconstruction is an effective way to improve the spatial resolution of the image.
Super-Resolution (SR) Image Reconstruction is the technology of reconstructing a frame of image with high resolution from a group of warped, blurred and noised Low-Resolution (LR) Images or video sequence about the same scene. It mainly integrates the relative movement information of the same scene's multiple low-resolution into the single high-resolution image, and obtains a high-quality image.
This thesis studies the two key issues of the reconstruction process on the basis of the mechanism of degraded image, image registration and reconstruction algorithm.
For the basic part of super-resolution image reconstruction, image registration, in order to the accurate registration, the thesis describes and contrast two registration methods, namely, Taylor series method and phase related to template matching. Focusing on the displacement directions of the phase correlation method, the thesis makes provision of the directional shift so as to get the correct amount of displacement.
According to the different theoretical system, the thesis discusses and realizes the three airspace commonly kinds of image reconstruction method. In the Maximum a Posteriori (MAP) method, by the relationship between the displacements of low-resolution sequence, construct degraded matrix of every point of low-resolution images using Gaussian function, as well as do a specific analysis about the regularization parameter. With the ringing effect of high-resolution reconstruction image based on the Projection onto Convex Sets (POCS) algorithm included, this thesis proposes a trapezoidal filtering method. The experimental tests show that this algorithm can significantly improve the quality of the reconstruction image. Finally Make understand and do the experiments about the reconstruction based on normal convolution (NC).
超分辨率图像重建技术是指将多幅变形、模糊、有噪、频谱混叠的低分辨率降质图像(或视频序列)融合估计出一幅高分辨率图像的技术。其核心的设计思想是：利用同一场景的多幅低分辨率图像间的相对互补信息,将它们融合到单幅高分辨率图像中,获得一幅高画质的图像。
本文在分析了图像降质模型的基础上,针对重建过程中的图像配准以及重建算法两个关键问题展开了研究。
考虑到低分辨率序列间的亚像素配准精度直接影响到超分辨率重建的效果,在图像配准这一关键环节,本文对泰勒级数法和相位相关模板匹配方法进行了实验和对比。针对相位相关法中的位移方向问题,本文对位移的方向性给出了一种合理的定义,实验证明由此定义得到的位移量能够满足配准精度的要求。
此外,本文对空域中常用的三种图像重建方法进行了深入研究和算法实现。在基于最大后验概率(MAP)的重建方法中,详细推导了退化过程的构造,解决了低分辨率图像产生过程中引入的散焦模糊问题,对正则化参数的作用进行了详细的实验分析。在基于凸集投影(POCS)的重建方法中,本文设计了用梯形低通滤波器来消除重建结果边缘振荡的现象,达到了较为理想的重建效果。最后对基于规范化卷积的重建方法进行了理解和实验测试。
In most electronic imaging applications, images with high resolution (HR) are often desired and required. In the process of acquiring the image, there are many factors lead to declines in image quality, such as sports, system noise. At the same time, the limit of imaging device makes the resolution of the image meet the application requirements. It is expensive and difficult to increase the current resolution level by improving hardware performance. Therefore, super-resolution image reconstruction is an effective way to improve the spatial resolution of the image.
Super-Resolution (SR) Image Reconstruction is the technology of reconstructing a frame of image with high resolution from a group of warped, blurred and noised Low-Resolution (LR) Images or video sequence about the same scene. It mainly integrates the relative movement information of the same scene's multiple low-resolution into the single high-resolution image, and obtains a high-quality image.
This thesis studies the two key issues of the reconstruction process on the basis of the mechanism of degraded image, image registration and reconstruction algorithm.
For the basic part of super-resolution image reconstruction, image registration, in order to the accurate registration, the thesis describes and contrast two registration methods, namely, Taylor series method and phase related to template matching. Focusing on the displacement directions of the phase correlation method, the thesis makes provision of the directional shift so as to get the correct amount of displacement.
According to the different theoretical system, the thesis discusses and realizes the three airspace commonly kinds of image reconstruction method. In the Maximum a Posteriori (MAP) method, by the relationship between the displacements of low-resolution sequence, construct degraded matrix of every point of low-resolution images using Gaussian function, as well as do a specific analysis about the regularization parameter. With the ringing effect of high-resolution reconstruction image based on the Projection onto Convex Sets (POCS) algorithm included, this thesis proposes a trapezoidal filtering method. The experimental tests show that this algorithm can significantly improve the quality of the reconstruction image. Finally Make understand and do the experiments about the reconstruction based on normal convolution (NC).
引文
[1]Tsai R Y, Huang A K. Multiframe image restoration and registration[J]. In Asvanceds in Computer Vision and Image Processing,1984:317-339.
[2]TOM B. C, KATSAGGELOS A K, GALATSANSOS N P. Reconstruction of a High Resolution Image from Registration and Restoration of Low Resolution Images[C]. ICIP Proceediys Image processing. IEEE International Conference,1994:553-557.
[3]S.P.Kim, N. K. Bose, H. M. Valenzuela. Recursive Reconstruction of High Resolution Image from Noisy Undersampled Multiframes[J]. IEEE Trans. on Speech Signal Processing,1990,38 (6):1013-1027.
[4]苏秉华,金伟其,牛丽红等.超分辨率图像复原及其进展[J].光学技术,2001,27(1)：6-9.
[5]孔玲莉,黄华,齐春,刘美娟.图像超分辨率研究的最新进展[J].光学技术,2004,30(3)：374-377.
[6]S.P.Kim, Su Wenyu. Recursive high resolution reconstruction of blurred multiframe images[J]. IEEE Transactions on Image Processing,1993,2 (4):534-539.
[7]侯莉.图像超分辨率重建算法研究[D].燕山大学,2007.
[8]袁建华.超分辨率重建中若干问题的研究[D].中国科学院电子学研究所,2006.
[9]闫华.基于重构的超分辨率技术研究[D].山东：山东大学,2007.
[10]张艳.空间域超分辨率图像重建技术研究[D].中国人民解放军信息工程大学,2007.
[11]邵文泽.基于图像建模理论的多幅图像正则化超分辨率重建算法研究[D].江苏：南京理工大学,2008.
[12]张平.基于序列图像重构的超分辨率成像技术研究[D].浙江：浙江大学,2003.
[13]甘亚莉,涂丹,李国辉.超分辨率图像配准方法研究[J].计算机应用研究,2007,24(8)：215-217.
[14]禹晶.序列图像超分辨率重建方法的研究[D].北京：北京工业大学,2005.
[15]黄婧.基于图像配准的超分辨率重建[D].黑龙江：哈尔滨工业大学,2006.
[16]P.Adelson. Model based techniques for image registration and three dimensional scene analysis from image sequences. Proc. Of 27th Asilaomar IEEE Conf. on Signals. Systems and Computers pp.1993:90-94.
[17]邹文艺,李庆辉,丁洪萍.基于预滤波迭代梯度法的运动估计[J].光学技术,2004,30(5)：538-540.
[18]刘良云,李英才,相里斌.超分辨率图像重构技术的仿真实验研究[J].中国图像图形学报,2001,6(7)：629-635.
[19]潘兵,续伯钦,李克景.梯度算子选择对基于梯度的亚像素位移算法的影响[J].光学技术, 2005,31(1)：26-30.
[20]李中科,吴乐南.基于霍夫变换和相位相关的图像配准方法[J].信号处理,2004,20(2)166-169.
[21]Murat Balci, Hassan Foroosh. Subpixel estimation of shifts directly in the fourier domain[J].IEEE Transactions on Image Processing,2006,15 (7):123-124.
[22]Chun havitta yatera, S.Chitsobhuk, O.Tongprasert, K.Image. Registration using Hough Transform and Phase Correlation[J]. Advanced Communication Technology.2006. ICACT 2006. The 8th International Conference Volume 2,20-22 Feb.2006 Page (s):973-977.
[23]Kenneth R.Castleman. Digital image processing [M].北京:电子工业出版社,2007.
[24]唐玎,涂丹,甘亚莉.基于相位相关和模板匹配的亚像素图像配准参数估计[J].微电子学与计算机,2007.
[25]Shaun S Gleason. Subpixel measurement of image features based on paraboloid surface fit[C]. Machine Vision Systems Inegration in Industry, SPIE Machine Vision Systems Integration in Industry,1990:135-144.
[26]侯成刚,赵明涛,屈梁生.基于二次曲面拟合的亚像素图像匹配算法[J].计量学,1997,18(3)：227-231.
[27]吕海霞.自动图像配准技术研究[D].西安：西北工业大学,2007.
[28]沈焕锋,李平湘,张良培.一种自适应正则MAP超分辨率重建方法[J].武汉大学学报信息科学版,2006,31(11)：949-952.
[29]钱浩.基于MAP的超分辨率重建算法研究[D].西安：西安电子科技大学,2006.
[30]刘晓天.基于MAP技术的图像超分辨率复原研究与实现[D].国防科学技术大学,2004.
[31]韩玉兵,吴乐南,张冬青.基于正则化处理的超分辨率重建[J].电子与信息学报,2007,29(7)：1713-1716.
[32]Sina Farsiu, M. Fast and Robust Multiframe Super Resolution[J]. IEEE TRANSACTIONS ON IMAGE PROCESSING,2004.
[33]钦桂勤.超分辨率遥感影像重建技术的研究[D].中国人民解放军信息工程大学,2002.
[34]P Cheeseman, B Kanefsky, R Kruft, et al. Super-Resolved Surface Reconstruction From Multiple Images[J]. NASA Technical Report,1994.
[35]邓乾国,游志胜.基于凸集投影算法的超分辨率图像重建技术[J].成都信息工程学院学报,2005,20(6)：708-711.
[36]陈光盛.空域超分辨率图像融合算法研究[D].湖南：湖南大学,2006.
[37]陈爱萍.基于图像配准的高分辨率重构技术研究[D].西安：西安电子科技大学,2007.
[38]Salina, F V Mascarenhas, N Cruvinel. A Comparison of POCS Algorithms for Tomographic Reconstruction under Noise and Limited View[J]. In:Proc of Computer Graphics and Image Processing. Symposium,2002,342-346.
[39]Ozkan M K, Tekalp A M, Sezan M I. POCS_based restoration of space-vayring blurred images[J]. IEEE Trans. On Image processing,1994,3 (7):450-45.
[40]A.S.CARASSO. The Apex Method in Image Sharpening and The Use of Low Exponent Le'vy Stable Law[J]. Society for Industrial and Applied Mathematics.2002.
[41]章毓晋.图像工程(上册)图像处理第2版[M].清华大学出版社,2006.
[42]Tuan Q. Pham, Lucas J. van Vliet, and Klamer Schutte. Robust Fusion of Irregularly Sampled Data Using Adaptive Normalized Convolution[J]. Hindawi Publishing Corporation EURASIP Journal on Applied Signal Processing,2006,1-12.
[43]王怀野,张科,李言俊.各向异性滤波在红外图像处理中的应用[J].红外与毫米波学报,2005,24(2)：109-113.
[2]TOM B. C, KATSAGGELOS A K, GALATSANSOS N P. Reconstruction of a High Resolution Image from Registration and Restoration of Low Resolution Images[C]. ICIP Proceediys Image processing. IEEE International Conference,1994:553-557.
[3]S.P.Kim, N. K. Bose, H. M. Valenzuela. Recursive Reconstruction of High Resolution Image from Noisy Undersampled Multiframes[J]. IEEE Trans. on Speech Signal Processing,1990,38 (6):1013-1027.
[4]苏秉华,金伟其,牛丽红等.超分辨率图像复原及其进展[J].光学技术,2001,27(1)：6-9.
[5]孔玲莉,黄华,齐春,刘美娟.图像超分辨率研究的最新进展[J].光学技术,2004,30(3)：374-377.
[6]S.P.Kim, Su Wenyu. Recursive high resolution reconstruction of blurred multiframe images[J]. IEEE Transactions on Image Processing,1993,2 (4):534-539.
[7]侯莉.图像超分辨率重建算法研究[D].燕山大学,2007.
[8]袁建华.超分辨率重建中若干问题的研究[D].中国科学院电子学研究所,2006.
[9]闫华.基于重构的超分辨率技术研究[D].山东：山东大学,2007.
[10]张艳.空间域超分辨率图像重建技术研究[D].中国人民解放军信息工程大学,2007.
[11]邵文泽.基于图像建模理论的多幅图像正则化超分辨率重建算法研究[D].江苏：南京理工大学,2008.
[12]张平.基于序列图像重构的超分辨率成像技术研究[D].浙江：浙江大学,2003.
[13]甘亚莉,涂丹,李国辉.超分辨率图像配准方法研究[J].计算机应用研究,2007,24(8)：215-217.
[14]禹晶.序列图像超分辨率重建方法的研究[D].北京：北京工业大学,2005.
[15]黄婧.基于图像配准的超分辨率重建[D].黑龙江：哈尔滨工业大学,2006.
[16]P.Adelson. Model based techniques for image registration and three dimensional scene analysis from image sequences. Proc. Of 27th Asilaomar IEEE Conf. on Signals. Systems and Computers pp.1993:90-94.
[17]邹文艺,李庆辉,丁洪萍.基于预滤波迭代梯度法的运动估计[J].光学技术,2004,30(5)：538-540.
[18]刘良云,李英才,相里斌.超分辨率图像重构技术的仿真实验研究[J].中国图像图形学报,2001,6(7)：629-635.
[19]潘兵,续伯钦,李克景.梯度算子选择对基于梯度的亚像素位移算法的影响[J].光学技术, 2005,31(1)：26-30.
[20]李中科,吴乐南.基于霍夫变换和相位相关的图像配准方法[J].信号处理,2004,20(2)166-169.
[21]Murat Balci, Hassan Foroosh. Subpixel estimation of shifts directly in the fourier domain[J].IEEE Transactions on Image Processing,2006,15 (7):123-124.
[22]Chun havitta yatera, S.Chitsobhuk, O.Tongprasert, K.Image. Registration using Hough Transform and Phase Correlation[J]. Advanced Communication Technology.2006. ICACT 2006. The 8th International Conference Volume 2,20-22 Feb.2006 Page (s):973-977.
[23]Kenneth R.Castleman. Digital image processing [M].北京:电子工业出版社,2007.
[24]唐玎,涂丹,甘亚莉.基于相位相关和模板匹配的亚像素图像配准参数估计[J].微电子学与计算机,2007.
[25]Shaun S Gleason. Subpixel measurement of image features based on paraboloid surface fit[C]. Machine Vision Systems Inegration in Industry, SPIE Machine Vision Systems Integration in Industry,1990:135-144.
[26]侯成刚,赵明涛,屈梁生.基于二次曲面拟合的亚像素图像匹配算法[J].计量学,1997,18(3)：227-231.
[27]吕海霞.自动图像配准技术研究[D].西安：西北工业大学,2007.
[28]沈焕锋,李平湘,张良培.一种自适应正则MAP超分辨率重建方法[J].武汉大学学报信息科学版,2006,31(11)：949-952.
[29]钱浩.基于MAP的超分辨率重建算法研究[D].西安：西安电子科技大学,2006.
[30]刘晓天.基于MAP技术的图像超分辨率复原研究与实现[D].国防科学技术大学,2004.
[31]韩玉兵,吴乐南,张冬青.基于正则化处理的超分辨率重建[J].电子与信息学报,2007,29(7)：1713-1716.
[32]Sina Farsiu, M. Fast and Robust Multiframe Super Resolution[J]. IEEE TRANSACTIONS ON IMAGE PROCESSING,2004.
[33]钦桂勤.超分辨率遥感影像重建技术的研究[D].中国人民解放军信息工程大学,2002.
[34]P Cheeseman, B Kanefsky, R Kruft, et al. Super-Resolved Surface Reconstruction From Multiple Images[J]. NASA Technical Report,1994.
[35]邓乾国,游志胜.基于凸集投影算法的超分辨率图像重建技术[J].成都信息工程学院学报,2005,20(6)：708-711.
[36]陈光盛.空域超分辨率图像融合算法研究[D].湖南：湖南大学,2006.
[37]陈爱萍.基于图像配准的高分辨率重构技术研究[D].西安：西安电子科技大学,2007.
[38]Salina, F V Mascarenhas, N Cruvinel. A Comparison of POCS Algorithms for Tomographic Reconstruction under Noise and Limited View[J]. In:Proc of Computer Graphics and Image Processing. Symposium,2002,342-346.
[39]Ozkan M K, Tekalp A M, Sezan M I. POCS_based restoration of space-vayring blurred images[J]. IEEE Trans. On Image processing,1994,3 (7):450-45.
[40]A.S.CARASSO. The Apex Method in Image Sharpening and The Use of Low Exponent Le'vy Stable Law[J]. Society for Industrial and Applied Mathematics.2002.
[41]章毓晋.图像工程(上册)图像处理第2版[M].清华大学出版社,2006.
[42]Tuan Q. Pham, Lucas J. van Vliet, and Klamer Schutte. Robust Fusion of Irregularly Sampled Data Using Adaptive Normalized Convolution[J]. Hindawi Publishing Corporation EURASIP Journal on Applied Signal Processing,2006,1-12.
[43]王怀野,张科,李言俊.各向异性滤波在红外图像处理中的应用[J].红外与毫米波学报,2005,24(2)：109-113.