## High Performance Automatic Image Registration For Remote Sensing

Citations: | 10 - 0 self |

### BibTeX

@MISC{Chalermwat_highperformance,

author = {Prachya Chalermwat},

title = {High Performance Automatic Image Registration For Remote Sensing},

year = {}

}

### OpenURL

### Abstract

HIGH-PERFORMANCE AUTOMATIC IMAGE REGISTRATION FOR REMOTE SENSING Prachya Chalermwat, Ph.D. George Mason University, 1999 Thesis Director: Dr. Tarek El-Ghazawi Image registration is one of the crucial steps in the analysis of remotely sensed data. A new acquired image must be transformed, using image registration techniques, to match the orientation and scale of previous related images. Image registration requires intensive computational effort not only because of its computational complexity, but also due to the continuous increase in image resolution and spectral bands. Thus, high-performance computing techniques for image registration are critically needed. Very few works have addressed image registration on contemporary highperformance computing systems. Furthermore, issues of load balancing, scalability, and formal analysis of algorithmic efficiency were seldom considered. iv This dissertation introduces high-performance automatic image registration (HAIR) algorithms. High performance is achieved by: 1) reduction in search data, 2) reduction in search space, and 3) parallel processing. Reduction in search data is achieved by performing registration using only subimages. A new metric called registrability is used to select those subimages such that accuracy is maintained. In addition, a histogram comparison is used to discard anomalous subimages, such as those with clouds. Further data reduction is obtained using an iterative refinement search (IRA), which exploits the wavelet multi-resolution representation. This technique starts searching images with lower resolution first, then refining the results using higher resolution images to use the least possible data points in the overall registration task. Reduction of search space is achieved through two methods. Firs...

### Citations

3066 | Adaptation in natural and artificial systems’. The - Holland - 1975 |

780 | A survey of image registration techniques
- Brown
- 1992
(Show Context)
Citation Context ...of superimposing two images and transforming one of them to find the best transform to make them match [Campbel87] [Jensen96].sBroad range of techniques for image registration has been categorized in =-=[Brown92]-=-.sThe complexity of the search and similarity measurement in image registration could be computationally intensive. Computational time becomes even more critical with the current increase in data.sAs ... |

343 |
Introductory Digital Image Processing: A Remote Sensing Perspective. Segunda Edición
- Jensen
- 1996
(Show Context)
Citation Context ...ed input image, and a reference image.sIn general, image registration is the process of superimposing two images and transforming one of them to find the best transform to make them match [Campbel87] =-=[Jensen96]-=-.sBroad range of techniques for image registration has been categorized in [Brown92].sThe complexity of the search and similarity measurement in image registration could be computationally intensive. ... |

285 |
Improving resolution by image registration [J
- Irani, Peleg
- 1991
(Show Context)
Citation Context ...tial extra computational and memory space requirement. The previously mentioned applications have led to the development of many different algorithms for subpixel image registration [Huang82][Irani89]=-=[Irani91]-=- [Keren88][Kim93] [Kuglin75][Shekarf96]. Four algorithms for achieving subpixel accuracy are intensity interpolation, correlation interpolation, differential method, and phase correlation [Tian86].sIn... |

211 |
Some computer organizations and their effectiveness
- Flynn
- 1972
(Show Context)
Citation Context ...y or in parallel as: 1) Single Instruction Single Data (SISD), 2) Single Instruction Multiple Data (SIMD), 3) Multiple Instruction Single Data (MISD), and 4) Multiple Instruction Multiple Data (MIMD) =-=[Flynn72]-=-.sModern parallel computers follow the MIMD model.sMIMD computers that have hundreds or thousands of processors are called Massively Parallel Processors (MPPs).sMPPs typically have nonstandard dedicat... |

211 |
D.C.: The phase correlation image alignment method
- Kuglin, Hines
- 1975
(Show Context)
Citation Context ...d memory space requirement. The previously mentioned applications have led to the development of many different algorithms for subpixel image registration [Huang82][Irani89][Irani91] [Keren88][Kim93] =-=[Kuglin75]-=-[Shekarf96]. Four algorithms for achieving subpixel accuracy are intensity interpolation, correlation interpolation, differential method, and phase correlation [Tian86].sIntensity Interpolation create... |

193 |
Algorithms for Image Processing and Computer Vision
- Parker
- 1996
(Show Context)
Citation Context ...t the new dimension of the reference image is equal to (10N x 10N). The registration accuracy depends on how accurately the new reference image approximates (by resampling methods) the original image =-=[Parker97]-=-. The intensity interpolation yields the best accuracy among others.sThe correlation interpolation calculates the discrete correlation function between two images, fits an interpolation surface to sam... |

190 |
Scientific Computing : An introductory Survey
- Heath
- 1996
(Show Context)
Citation Context ...sing least squares estimation.sGauss developed the technique of least squares for solving problems in astronomy, in particular, to determine the orbits of celestial 8bodies such as planets and comets =-=[Heath97]-=-.sLeast squares solution minimizes the sum of squares of vertical distances between the data points and the curve over the used polynomials.sHowever, there are some practical difficulties that must be... |

182 |
Parallel Computing: Theory and Practice
- Quinn
- 1994
(Show Context)
Citation Context ...essing techniques like data and control parallelism are frequently used.sData parallelism is the use of multiple functional units to applysthe same operation simultaneously to different data elements =-=[Quinn94]-=-.sIn contrast, control parallelism is achieved by applying different operations to different data elements 11 simultaneously.sWhile parallel processing reduces the total execution time, it introduces ... |

125 |
Introduction to Remote Sensing
- Campbell
- 1996
(Show Context)
Citation Context ...control points are a set of selected pixels (or regions) that contains important features like intersection of roads or coastlines.sThey are features that can be located accurately with ease on a map =-=[Campbell87]-=-.sThese control points can be used to identify the transforms between the input image and the reference image.sAn assumption is made that the reference image is geometrically correct [Richards86].sTo ... |

94 |
Multiresolution Signal Decomposition
- Akansu, Haddad
- 1992
(Show Context)
Citation Context ...ize of the search data can be reduced by initially searching at lowest resolution images (smallest data size) and then proceeding to higher resolution images where the search results are only refined =-=[Akansu95]-=- [Corvi95] [LeMoigne94] [LeMoigne96] [LeMoigne97].sWavelet-based multi-resolution preserves most of the important features of the original data even at a low resolution.sIt also eliminates weak featur... |

85 | Harnessing the Power of Parallelism in a Pile-of-PCs - Ridge, Becker, et al. - 1997 |

68 |
Image sequence enhancement using sub-pixel displacement
- Keren, Peleg, et al.
- 1988
(Show Context)
Citation Context ... computational and memory space requirement. The previously mentioned applications have led to the development of many different algorithms for subpixel image registration [Huang82][Irani89][Irani91] =-=[Keren88]-=-[Kim93] [Kuglin75][Shekarf96]. Four algorithms for achieving subpixel accuracy are intensity interpolation, correlation interpolation, differential method, and phase correlation [Tian86].sIntensity In... |

67 |
Algorithms for subpixel registration
- Tian, Huhns
- 1986
(Show Context)
Citation Context ...surements in remote sensing, image sequence analysis, and nondestructive evaluation require registration results with an error less than one pixel, also called super-resolution 9and subpixel accuracy =-=[Tian86]-=-.sHowever, subpixel registration greatly introduces additional computation power and memory space requirement. Original Subpixel Accuracy = 0.5 Subpixel Accuracy = 0.25 Subpixel Accuracy = 0.125 Figur... |

51 | The impact of misregistration on change detection
- Townshend
- 1992
(Show Context)
Citation Context ...Accuracy in registration is also very important. Misregistration by only 1 pixel can introduce over 50% error in some change detection applications using Normalized Difference Vegetation Indexs(NDVI) =-=[Townshend92]-=-. Due to the computational intensive requirement of the registration process and the large amount of image data to be registered, parallel image registration is needed to sustain such requirements.sPa... |

45 |
Super resolution from image sequences
- Irani, Peleg
- 1990
(Show Context)
Citation Context ...s substantial extra computational and memory space requirement. The previously mentioned applications have led to the development of many different algorithms for subpixel image registration [Huang82]=-=[Irani89]-=-[Irani91] [Keren88][Kim93] [Kuglin75][Shekarf96]. Four algorithms for achieving subpixel accuracy are intensity interpolation, correlation interpolation, differential method, and phase correlation [Ti... |

40 |
Correlation techniques of image registration
- Pratt
- 1974
(Show Context)
Citation Context ... normalized cross-correlation, statistical correlation, match filters, phase-correlation, sum of absolute differences, root mean square, and masked correlation. Studies in [Eppler96], [Tilton97], and =-=[Pratt74]-=- have shown that normalized cross-correlation is, although expensive, ranked as one of the best similarity measures. The normalized cross-correlation function can be written as ∑∑∑∑ ∑∑ − = − = − = − =... |

32 |
Genetic Algorithms, The
- Buckles, Petry
- 1992
(Show Context)
Citation Context ... Chromosome Representation A chromosome of length n is a vector of the form <x1,x2,…,xn>, where each xi is an allele or gene.sThe type of a gene can be binary digit, integer, or floating point number =-=[Buckles92]-=-.sBinary genes are widely used in GA applications.sIn this dissertation, we use a 32-bit chromosome represented by a binary string of ‘0’s and ‘1’s. Selection Scheme Two popular selection schemes are:... |

29 | Subpixel image registration by estimating the polyphase decomposition of cross power spectrum
- Shekarforoush, Rerthod, et al.
- 1996
(Show Context)
Citation Context ...pace requirement. The previously mentioned applications have led to the development of many different algorithms for subpixel image registration [Huang82][Irani89][Irani91] [Keren88][Kim93] [Kuglin75]=-=[Shekarf96]-=-. Four algorithms for achieving subpixel accuracy are intensity interpolation, correlation interpolation, differential method, and phase correlation [Tian86].sIntensity Interpolation creates a much de... |

25 | Image registration by genetic search - Fitzpatrick, Grefenstette, et al. - 1984 |

24 | Introduction to genetic algorithms - Chipperfield - 1998 |

23 |
Subpixel accuracy image registration by spectrum cancellation
- Kim, Su
- 1993
(Show Context)
Citation Context ...ional and memory space requirement. The previously mentioned applications have led to the development of many different algorithms for subpixel image registration [Huang82][Irani89][Irani91] [Keren88]=-=[Kim93]-=- [Kuglin75][Shekarf96]. Four algorithms for achieving subpixel accuracy are intensity interpolation, correlation interpolation, differential method, and phase correlation [Tian86].sIntensity Interpola... |

23 | Maintaining Diversity in Genetic Search
- Mauldin
- 1984
(Show Context)
Citation Context ...ture. This work tries to solve the premature convergence problem of simple GAs due to the presence of multiple local optima in searching a high 24 dimensional parameter space [Dasgupta92][Goldburg89] =-=[Mauldin84]-=-. The sGA overcomes the problem by using the genetic redundancy and hierarchical gnomic structures in its chromosome.sIn sGA, each chromosome is represented by a set of binary strings.sIt differs from... |

23 | Achieving a Balanced LowCost Architecture for Mass Storage Managenment through Multiple Fast
- Sterling, Becker, et al.
- 1996
(Show Context)
Citation Context ...ulf cluster. Since the development of NASA’s Beowulf project, parallel processing based on parallel computers using commodity off-the-shelf (COTS) components has gained significant attention recently =-=[Sterling96]-=-[Berry96][Ridge97].sThe Beowulf clusters are distinguished from non-dedicated Networks of Workstations (NOWs) by several characteristics. First, the nodes in the cluster are dedicated in order to have... |

22 | Sub-pixel bayesian estimation of albedo and height - Shekarforoush, Berthod, et al. - 1996 |

16 | Reconstruction of high resolution 3D visual information - Berthod, Shekarforoush, et al. - 1994 |

13 |
Geometric Approach to Subpixel Registration Accuracy” CVGIP
- Bernstein
- 1987
(Show Context)
Citation Context ...ine effective GCPs [Campbell87].sBernstein has shown that registration error decreases as the number of GCPs is increased, however, the quality of GCPs accuracy may decrease as their number increases =-=[Bernstein87]-=-.sOn the other hand, spatial correlation between the two images can be used for automatic registration by finding the maximum correlation between the input image and the transformed versions of the re... |

13 |
Multi-resolution Image Registration
- Corvi, Nicchiotti
(Show Context)
Citation Context ...search data can be reduced by initially searching at lowest resolution images (smallest data size) and then proceeding to higher resolution images where the search results are only refined [Akansu95] =-=[Corvi95]-=- [LeMoigne94] [LeMoigne96] [LeMoigne97].sWavelet-based multi-resolution preserves most of the important features of the original data even at a low resolution.sIt also eliminates weak features in high... |

12 |
Scalable Parallel Computing: Technology,Architecture,Programming
- Hwang, Xu
- 1998
(Show Context)
Citation Context ...cessors are called Massively Parallel Processors (MPPs).sMPPs typically have nonstandard dedicated interconnection network (ICN) which allows the processors to collaborate and facilitate input/output =-=[Hwang97]-=-. Examples of current MPPs are the IBM SP2, Cray T3E, and SGI Origin 2000.sMPPs use commodity microprocessors (or processing nodes) and physically distributed memory, where each memory module may be a... |

12 |
Parallel registration of multi-sensor remotely sensed imagery using wavelet coefficients
- Moigne
- 1994
(Show Context)
Citation Context ...a can be reduced by initially searching at lowest resolution images (smallest data size) and then proceeding to higher resolution images where the search results are only refined [Akansu95] [Corvi95] =-=[LeMoigne94]-=- [LeMoigne96] [LeMoigne97].sWavelet-based multi-resolution preserves most of the important features of the original data even at a low resolution.sIt also eliminates weak features in higher resolution... |

11 | A hardware architecture for a parallel genetic algorithm for image registration
- Turton, Arslan, et al.
- 1994
(Show Context)
Citation Context ...ing of search scope, some researchers have attempted to apply Genetic Algorithms (GAs) to help search over the complex search scope in image registration [Dasgupta92] [Fitzpatrick84] [Ou96] [Ozkan89] =-=[Turton94]-=-. Genetic Algorithms are based on an adaptive stochastic random search [Holland75]. GAs mimic the natural evolutionary process by representing the potential solutions with a set of chromosomes in the ... |

9 |
Genetic Algorithms in Optimisation and Adaptation
- Husbands
- 1990
(Show Context)
Citation Context ... chromosome to guarantee that the probability of searching a particular subspace of the problem space is never zero [Chipperfield96].sThis prevents the algorithm from becoming trapped in local optima =-=[Husband90]-=- [Goldburg89].sThe whole population is evaluated again in the next generation.sThe process continues until a termination criteria is reached.sThe termination criteria may be finding an acceptable appr... |

8 |
Video rate image correlation processor
- Pearson, Hines, et al.
- 1977
(Show Context)
Citation Context ...th respect to their sizes [Huang81] [Tian86]. The phase correlation technique has also been used for implementing a video-rate image correlation processors as well as an image registration [Kuglin75] =-=[Pearson77]-=- [Shekarf95]. The subpixel accuracy is achieved by examining the relative insensitivity to scene content, illumination differences, and narrow-band noise. This technique is based on the fact that most... |

7 |
2-phase GA-based image registration on parallel clusters
- Chalermwat, El-Ghazawi, et al.
- 2001
(Show Context)
Citation Context ...ull resolution, the subimage selected by using the automatic RIG-based subimage selection can be applied as a mean to search-data reduction.sOne special case of the HAIR, namely the 2-phase algorithm =-=[Chalermwat99]-=-, has already been considered and experimentally tested in the previous section. 112 Begini End Wavelet decompose input and reference images GA-based search at lowest resolution image Search intermedi... |

7 |
Digital Image Registration using Structured Genetic Algorithms
- Dasgupta, McGregor
- 1992
(Show Context)
Citation Context ...techniques employ linear search over the sampling of search scope, some researchers have attempted to apply Genetic Algorithms (GAs) to help search over the complex search scope in image registration =-=[Dasgupta92]-=- [Fitzpatrick84] [Ou96] [Ozkan89] [Turton94]. Genetic Algorithms are based on an adaptive stochastic random search [Holland75]. GAs mimic the natural evolutionary process by representing the potential... |

7 | Automated multisensor registration: requirements and techniques - Rignot, Kowk, et al. - 1991 |

5 |
Bounds on (deterministic) correlation functions with applications to registration
- Dvorchenko
- 1983
(Show Context)
Citation Context ...lation interpolation calculates the discrete correlation function between two images, fits an interpolation surface to samples of this function, and accurately searches for the maximum of the surface =-=[Dvorchenko83]-=-. The accuracy of this method depends on how well a correlation function around the peak approximates a parabola.sA differential method estimates 2-dimensional translations to process a sequence of vi... |

5 |
Kenney,”Scope and Applications of Translation Invariant Wavelets to Image Registration
- Fonseca
- 1997
(Show Context)
Citation Context ...in image registration. Fonseca and Manjunath presented a multi-resolution registration that relies on the grey level information content of the images and their local wavelet-transform modulas maxima =-=[Fonseca97]-=-. The proposed algorithm consists of five major steps: 1) smoothing the image, 2) decomposing the image using wavelet, 3) extracting the feature points, 4) matching the feature points, and 5) refining... |

5 | A Waveletbased Point Feature Extractor for Multi Sensor Image Registration - Li, Zhou - 1996 |

5 | Techniques for Multi-resolution Image Registration in the Presence of Occlusions - McGuire, Stone - 1997 |

4 | Tarek El-Ghazawi. "Parallel Input/Output Characteristics of NASA Science Applications - Berry - 1996 |

4 |
Wavelet decomposition on highperformance computing systems
- El-Ghazawi, LeMoigne
- 1996
(Show Context)
Citation Context ...WFTs can provide localization information, they do not provide flexible division of the time-frequency plane that can track slow changing phenomena while providing more details for higher frequencies =-=[El-Ghazawi96]-=-.sThe wavelet representation was introduced to correct the drawback of the former two methods using a multi-resolution scheme.sThe Wavelet transforms contribute to the desired sampling by filtering th... |

4 |
Image Registration for a Transputer-Based Distributed System
- Ozkan, Fitzpatrick, et al.
(Show Context)
Citation Context ... the sampling of search scope, some researchers have attempted to apply Genetic Algorithms (GAs) to help search over the complex search scope in image registration [Dasgupta92] [Fitzpatrick84] [Ou96] =-=[Ozkan89]-=- [Turton94]. Genetic Algorithms are based on an adaptive stochastic random search [Holland75]. GAs mimic the natural evolutionary process by representing the potential solutions with a set of chromoso... |

4 |
Real-time image registration based on genetic algorithms
- Ou, Chen, et al.
- 1996
(Show Context)
Citation Context ...ch over the sampling of search scope, some researchers have attempted to apply Genetic Algorithms (GAs) to help search over the complex search scope in image registration [Dasgupta92] [Fitzpatrick84] =-=[Ou96]-=- [Ozkan89] [Turton94]. Genetic Algorithms are based on an adaptive stochastic random search [Holland75]. GAs mimic the natural evolutionary process by representing the potential solutions with a set o... |

4 |
Parallel Genetic Algorithms”, Parallel and Distributed Computing Handbook by
- Chipperfield, Fleming
- 1996
(Show Context)
Citation Context ...ext generation.sThe “mutation” operator is then applied to perturb the string of the chromosome to guarantee that the probability of searching a particular subspace of the problem space is never zero =-=[Chipperfield96]-=-.sThis prevents the algorithm from becoming trapped in local optima [Husband90] [Goldburg89].sThe whole population is evaluated again in the next generation.sThe process continues until a termination ... |

3 |
Image registration by parts
- Chalermwat, El-Ghazawi, et al.
- 1997
(Show Context)
Citation Context ... As noted earlier, substantial reduction in computations can be obtained by simply partitioning original images into subimages (figure 3.1), and using some or all of them to perform image correlation =-=[Chalermwat97]-=-.sThe RIG-based measures can be used for rejecting hard to register subimages such as smooth cloud covered areas and featureless sea surfaces.sUsing such subimages could lead to misregistration while ... |

3 |
Wavelet Transforms: Theory and Applications
- Dianat, Rao
- 1998
(Show Context)
Citation Context ...elets, which is particularly useful for data reconstruction [Daubechies92]. 53 The wavelet transform is defined as the time-integrated product of a signal f(t) with a set of analyzing basis functions =-=[Dianat98]-=-.sThe basis functions are defined as −Ψ=Ψ a bt atba 2 1 , )( , (4.1) where ψab is the mother wavelet and the parameters a and b are the scale and shift parameters respectively.sThe continuous... |

3 | Wavelet-based image Registration on parallel computers - El-Ghazawi, Chalermwat, et al. - 1521 |

3 |
Genetic Algorithms in Search: optimization and machine learning
- Goldburg
- 1989
(Show Context)
Citation Context ...layered structure. This work tries to solve the premature convergence problem of simple GAs due to the presence of multiple local optima in searching a high 24 dimensional parameter space [Dasgupta92]=-=[Goldburg89]-=- [Mauldin84]. The sGA overcomes the problem by using the genetic redundancy and hierarchical gnomic structures in its chromosome.sIn sGA, each chromosome is represented by a set of binary strings.sIt ... |

2 |
Automatic Selction of Control Points for the Registration of Digital Images
- Davis, Kenue
(Show Context)
Citation Context ...perly. Most practitioners manually select the chips before performing the registration [Rignot91].sThe subimages can be selected based on the features obtained from binary edge detection described in =-=[Davis78]-=-. Although multi-resolution techniques can dramatically reduce computations, substantial parts of the computations are in the high-resolution level close to the full resolution and experimental result... |