## Joint MAP registration and high-resolution image estimation using a sequence of undersampled images (1997)

### Cached

### Download Links

- [www.doc.ic.ac.uk]
- [homepages.udayton.edu]
- DBLP

### Other Repositories/Bibliography

Venue: | IEEE Transactions on Image Processing |

Citations: | 169 - 2 self |

### BibTeX

@ARTICLE{Hardie97jointmap,

author = {Russell C. Hardie and Kenneth J. Barnard and Ernest E. Armstrong},

title = {Joint MAP registration and high-resolution image estimation using a sequence of undersampled images},

journal = {IEEE Transactions on Image Processing},

year = {1997},

volume = {6},

pages = {1621--1633}

}

### Years of Citing Articles

### OpenURL

### Abstract

Abstract — In many imaging systems, the detector array is not sufficiently dense to adequately sample the scene with the desired field of view. This is particularly true for many infrared focal plane arrays. Thus, the resulting images may be severely aliased. This paper examines a technique for estimating a high-resolution image, with reduced aliasing, from a sequence of undersampled frames. Several approaches to this problem have been investigated previously. However, in this paper a maximum a posteriori (MAP) framework for jointly estimating image registration parameters and the high-resolution image is presented. Several previous approaches have relied on knowing the registration parameters a priori or have utilized registration techniques not specifically designed to treat severely aliased images. In the proposed method, the registration parameters are iteratively updated along with the high-resolution image in a cyclic coordinate-descent optimization procedure. Experimental results are provided to illustrate the performance of the proposed MAP algorithm using both visible and infrared images. Quantitative error analysis is provided and several images are shown for subjective evaluation. Index Terms—Aliased, high resolution, image registration, image sequence, MAP estimation. I.

### Citations

3990 |
Stochastic relaxation, Gibbs distributions and the Bayesian restoration of images
- Geman, Geman
- 1984
(Show Context)
Citation Context ...ly reflect the characteristics of the random process from which is viewed as a realization. While the selection of an image prior is application dependent, we believe that a simple Gibbs distribution =-=[23]-=-, [24] can provide a useful solution for a wide variety of imaging systems. As will be seen, this image prior model used with a Gaussian noise model has the desirable property of yielding a convex MAP... |

1225 |
Spatial interaction and the statistical analysis of lattice systems
- Besag
- 1974
(Show Context)
Citation Context ...l cliques with nonzero clique potential functions look like the set of shaded pixels in Fig. 2 including the center sample labeled . Such cliques can be derived from a third-order neighborhood system =-=[25]-=-, [26]. The choice of a prior statistical model for the registration parameters will be highly application specific. Furthermore, in many cases, the registration parameters may be significantly overde... |

1133 |
Linear and Nonlinear Programming
- Luenberger
- 1984
(Show Context)
Citation Context ...in this paper we seek to minimize a MAP cost function with respect to the high-resolution image and the registration parameters simultaneously using a cyclic coordinate-descent optimization procedure =-=[18]-=-. In this iterative technique, the registration parameter estimates are updated using the current best estimate of the high-resolution image. In this way, information from all the observed frames is u... |

499 |
Introduction to Fourier Optics
- Goodman
- 1996
(Show Context)
Citation Context ...of the detectors is 40 m 40 m, yielding a fill factor of 80%. Assuming a midband wavelength of 4 m, the theoretical diffraction-limited optical transfer function (OTF) cuts off at 8.33 cyc/mrad [27], =-=[28]-=- and the effective sampling rate of the FPA is 2 cyc/mrad. Thus, the sampling rate must be increased by a factor of 8.33 to guarantee there will be no aliasing effects when imaging an arbitrary scene.... |

282 |
and Shmuel Peleg: Improving Resolution by Image Registration
- Irani
- 1990
(Show Context)
Citation Context ...e correlation is suggested in [9], [10] as a means of estimating the required motion parameters. A related multiframe technique which considers global translational shift and rotation is presented in =-=[11]-=-. In [12], this technique is extended to treat the more general case of a perspective projection of a plane. All of these methods rely on a two stage estimation procedure, whereby the registration is ... |

249 |
Modeling and segmentation of noisy and textured images using Gibbs random fields
- Derin, Elliott
- 1987
(Show Context)
Citation Context ...ues with nonzero clique potential functions look like the set of shaded pixels in Fig. 2 including the center sample labeled . Such cliques can be derived from a third-order neighborhood system [25], =-=[26]-=-. The choice of a prior statistical model for the registration parameters will be highly application specific. Furthermore, in many cases, the registration parameters may be significantly overdetermin... |

248 | A generalized Gaussian image model for edge-preserving MAP estimation
- Bouman, Sauer
- 1993
(Show Context)
Citation Context ...lect the characteristics of the random process from which is viewed as a realization. While the selection of an image prior is application dependent, we believe that a simple Gibbs distribution [23], =-=[24]-=- can provide a useful solution for a wide variety of imaging systems. As will be seen, this image prior model used with a Gaussian noise model has the desirable property of yielding a convex MAP cost ... |

220 | Extraction of High-Resolution Frames from Video Sequences
- Stevenson
- 1996
(Show Context)
Citation Context ...ndently of the high-resolution image reconstruction. The problem has also been approached from a statistical estimation framework. Specifically, a maximum a posteriori (MAP) estimator is developed in =-=[13]-=- and [14], which is an extension of a single frame image expansion algorithm proposed in [15]. The MAP estimator in [13] and [14] uses an edge preserving Huber–Markov random field for the image prior.... |

201 |
Multi-Frame Image Restoration and Registration
- Huang, Tsai
- 1984
(Show Context)
Citation Context .... This has motivated the development of the approach presented here. This basic problem of high resolution image recovery using multiple frames was first addressed in the literature by Tsai and Huang =-=[4]-=-. Their observation model is based on the shift property of the Fourier transform. Each aliased observation provides a set of equations in the frequency domain. Provided that enough frames are availab... |

121 | Virtual bellows: Constructing high quality stills from video
- Mann, Picard
- 1994
(Show Context)
Citation Context ...tion is suggested in [9], [10] as a means of estimating the required motion parameters. A related multiframe technique which considers global translational shift and rotation is presented in [11]. In =-=[12]-=-, this technique is extended to treat the more general case of a perspective projection of a plane. All of these methods rely on a two stage estimation procedure, whereby the registration is done inde... |

113 |
Recursive Reconstruction f High Resolution Image from Noisy Undersampled Multiframes
- Kim, Bose, et al.
- 1990
(Show Context)
Citation Context ...ause of prohibitively high computational complexity. Furthermore, it requires a set minimum number of frames to operate, which may not be available. The image recovery algorithm in [4] is extended in =-=[5]-=- for the case where noise is considered by using 1sa recursive least squares solution for the set of frequency domain linear equations. This is extended again for the case where blurring in considered... |

101 | Super-resolved surface reconstruction from multiple images
- Cheeseman, Kanefsky, et al.
- 1996
(Show Context)
Citation Context ... exploit all the observed frames when estimating the motion parameters for each image. Another related multiframe MAP technique, which uses an approach similar to that described here, can be found in =-=[16]-=-. A statistical estimation approach to multiframe resolution enhancement was first applied to staring infrared imager data in [17]. In particular, a maximum likelihood (ML) technique using the expecta... |

76 |
Linear Systems, Fourier Transforms, and Optics
- Gaskill
- 1978
(Show Context)
Citation Context ... area of the detectors is 40 m 40 m, yielding a fill factor of 80%. Assuming a midband wavelength of 4 m, the theoretical diffraction-limited optical transfer function (OTF) cuts off at 8.33 cyc/mrad =-=[27]-=-, [28] and the effective sampling rate of the FPA is 2 cyc/mrad. Thus, the sampling rate must be increased by a factor of 8.33 to guarantee there will be no aliasing effects when imaging an arbitrary ... |

71 |
High-resolution image recovery from image-plane arrays, using convex projections
- Stark, Oskoui
- 1989
(Show Context)
Citation Context ... suboptimal scheme for estimating the global frame shifts is described in [7]. Another approach to the high-resolution image reconstruction problem uses a projection onto convex sets (POCS) algorithm =-=[8]-=-. The POCS approach has been extended to treat motion blur and noise in [1], [9], [10]. Block matching or phase correlation is suggested in [9], [10] as a means of estimating the required motion param... |

48 |
Recursive high-resolution reconstruction of blurred multiframe images
- Kim, Su
- 1993
(Show Context)
Citation Context ...olution for the set of frequency domain linear equations. This is extended again for the cases1622 IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 6, NO. 12, DECEMBER 1997 where blurring in considered in =-=[6]-=-. However, the estimation of the global frame shifts is not addressed in [5] or [6]. A technique that uses the least squares solution, similar to that in [5], along with a fast suboptimal scheme for e... |

42 |
High-resolution image reconstruction from a low-resolution image sequence in the presence of time-varing motion blur
- Patti, Sezan, et al.
- 1994
(Show Context)
Citation Context ...nother approach to the high-resolution image reconstruction problem uses a projection onto convex sets (POCS) algorithm [8]. The POCS approach has been extended to treat motion blur and noise in [1], =-=[9]-=-, [10]. Block matching or phase correlation is suggested in [9], [10] as a means of estimating the required motion parameters. A related multiframe technique which considers global translational shift... |

27 |
Probability and Stochastic Processes for Engineers
- Helstrom
- 1991
(Show Context)
Citation Context ...is photon noise is the dominant noise source in the system. When there exists a large number of photon generated charge carriers, the Poisson statistics can be approximated by a Gaussian distribution =-=[21]-=-. Furthermore, if the imaging system is not photon-noise limited, then the superposition of the contributing events from all noise sources may result in a probability density function (pdf) that appro... |

15 |
Infrared Detectors and Systems
- Dereniak, Boreman
- 1996
(Show Context)
Citation Context ...les with variance . We believe that a Gaussian noise model is useful for a variety of imaging systems and scenarios. In most practical imaging applications, detected photons follow Poisson statistics =-=[19]-=-. The measurable manifestation of photon-noise depends on the particular detector type. For photon detectors that detect light through the conversion of photons into charge carriers, those carriers al... |

14 |
High-resolution infrared image reconstruction using multiple low resolution aliased frames
- Kaltenbacher, Hardie
- 1996
(Show Context)
Citation Context ...ts is not addressed in [5] or [6]. A technique that uses the least squares solution, similar to that in [5], along with a fast suboptimal scheme for estimating the global frame shifts is described in =-=[7]-=-. Another approach to the high-resolution image reconstruction problem uses a projection onto convex sets (POCS) algorithm [8]. The POCS approach has been extended to treat motion blur and noise in [1... |

4 |
Aliasing and Blurring in Microscanned Imagery
- Watson, Muse, et al.
- 1992
(Show Context)
Citation Context ...l generally provide the necessary motion between the focal plane array and the scene at each acquisition time with minimal occlusion effects. This process is referred to as uncontrolled microscanning =-=[2, 3]-=-. The key to exploiting these multiple frames is accurate knowledge of the sub-pixel registration parameters for each frame. If the images are severely undersampled, we have found that traditional mot... |

3 |
Reduction of aliasing in staring infrared imagers utilizing subpixel techniques
- Gillette, Stadtmiller, et al.
- 1995
(Show Context)
Citation Context ...l generally provide the necessary motion between the focal plane array and the scene at each acquisition time with minimal occlusion effects. This process is referred to as uncontrolled microscanning =-=[2, 3]-=-. The key to exploiting these multiple frames is accurate knowledge of the sub-pixel registration parameters for each frame. If the images are severely undersampled, we have found that traditional mot... |

3 |
Restoration of aliased video sequences via a maximum-likelihood approach
- Cain, Hardie, et al.
- 1996
(Show Context)
Citation Context ...hich uses an approach similar to that described here, can be found in [16]. A statistical estimation approach to multiframe resolution enhancement was first applied to staring infrared imager data in =-=[17]-=-. In particular, a maximum likelihood (ML) technique using the expectation maximization (EM) algorithm is developed which seeks to estimate translational motion and a high-resolution image [17] in the... |

2 |
Bayesian approach to image expansion for improved definition
- “A
- 1994
(Show Context)
Citation Context ...m a statistical estimation framework. Specifically, a maximum a posteriori (MAP) estimator is developed in [13] and [14], which is an extension of a single frame image expansion algorithm proposed in =-=[15]-=-. The MAP estimator in [13] and [14] uses an edge preserving Huber–Markov random field for the image prior. The motion is estimated by a block matching algorithm applied to the individual frames. Thes... |

1 |
Fundamentals of Electronic Imaging Systems, 2nd ed
- Schreiber
- 1991
(Show Context)
Citation Context ...hoton-noise depends on the particular detector type. For photon detectors that detect light through the conversion of photons into charge carriers, those carriers also follow Poisson statistics [19], =-=[20]-=-. A system is photon-noise limited when this photon noise is the dominant noise source in the system. When there exists a large number of photon generated charge carriers, the Poisson statistics can b... |