## The dual-bootstrap iterative closest point algorithm with application to retinal image registration (2003)

### Cached

### Download Links

Venue: | IEEE Trans. Med. Img |

Citations: | 57 - 18 self |

### BibTeX

@ARTICLE{Stewart03thedual-bootstrap,

author = {Charles V. Stewart and Chia-ling Tsai and Badrinath Roysam},

title = {The dual-bootstrap iterative closest point algorithm with application to retinal image registration},

journal = {IEEE Trans. Med. Img},

year = {2003},

volume = {22},

pages = {2003}

}

### Years of Citing Articles

### OpenURL

### Abstract

Abstract—Motivated by the problem of retinal image registration, this paper introduces and analyzes a new registration algorithm called Dual-Bootstrap Iterative Closest Point (Dual-Bootstrap ICP). The approach is to start from one or more initial, low-order estimates that are only accurate in small image regions, called bootstrap regions. In each bootstrap region, the algorithm iteratively: 1) refines the transformation estimate using constraints only from within the bootstrap region; 2) expands the bootstrap region; and 3) tests to see if a higher order transformation model can be used, stopping when the region expands to cover the overlap between images. Steps 1): and 3), the bootstrap steps, are governed by the covariance matrix of the estimated transformation. Estimation refinement [Step 2)] uses a novel robust version of the ICP algorithm. In registering retinal image pairs, Dual-Bootstrap ICP is initialized by automatically matching individual vascular landmarks, and it aligns images based on detected blood vessel centerlines. The resulting quadratic transformations are accurate to less than a pixel. On tests involving approximately 6000 image pairs, it successfully registered 99.5 % of the pairs containing at least one common landmark, and 100 % of the pairs containing at least one common landmark and at least 35 % image overlap. Index Terms—Iterative closest point, medical imaging, registration, retinal imaging, robust estimation.

### Citations

2466 |
RANdom SAmpling Consensus: a paradigm for model fitting with application to image analysis and automated cartography
- Fischler, Bolles
- 1981
(Show Context)
Citation Context ...d transformation estimation problems [33, 76, 77, 79], including applications of ICP [13, 46]. The difference that the Dual-Bootstrap ICP algorithm offers is important. Minimal subset random sampling =-=[26, 58]-=- requires a sufficient set of correspondences to generate a full model. When using the Dual-Bootstrap ICP algorithm, a much weaker initial model may be used, with fewer initial correspondences. This i... |

2230 |
A method for registration of 3-d shapes
- Besl, McKay
- 1992
(Show Context)
Citation Context ...oduction 1.1 The Iterative Closest Point Algorithm The iterative closest point (ICP) registration algorithm was invented almost simultaneously in the early 1990’s by several different research group=-=s [3, 12, 14, 49, 78], an-=-d has been used in many different applications since then [22, 43, 49, 55]. ICP is a point-based registration algorithm, where the “points” may be raw measurements such as (x, y, z) values from ra... |

1876 | Numerical recipes: the art of scientific computing - Press, Teukolsky, et al. - 2007 |

1834 |
Robust Statistics
- Huber
- 1981
(Show Context)
Citation Context ...ses bias in the transformation estimate. A different issue is that as written, thesRPI-CS-TR 02-9 9 objective function shown in ( 1) is minimized by σ → ∞. This can be avoided by adding a log(σ)=-= term [37] or-=- by (robustly) estimating and then fixing σ using a separate process during minimization [70]. • To avoid the trivial minimum corresponding to C = {}, restrictions are often placed on C. A common o... |

1731 | A Combined Corner and Edge Detector
- Harris, Stephens
- 1988
(Show Context)
Citation Context ...TR 02-9 7 from image I2. These points could be as simple as (x, y, z) T coordinates taken from range images, or they could be descriptions of edge elements, interest points [64, 68], corner locations =-=[32], or-=- other image structures. The problem is to find the transformation parameters, θ, and associated set of correspondences, C ⊂ P × Q, that minimize an appropriate error distance. A general objective... |

1479 | Good features to track
- Shi, Tomasi
- 1994
(Show Context)
Citation Context ...er quadratic model.sRPI-CS-TR 02-9 7 from image I2. These points could be as simple as (x, y, z) T coordinates taken from range images, or they could be descriptions of edge elements, interest points =-=[64, 68], co-=-rner locations [32], or other image structures. The problem is to find the transformation parameters, θ, and associated set of correspondences, C ⊂ P × Q, that minimize an appropriate error distan... |

1094 |
Robust Regression and Outlier Detection
- Rousseeuw
- 2003
(Show Context)
Citation Context ...ithms. This will provide a formal context for the main contributions of the Dual-Bootstrap ICP algorithm. One difference between this formulation and standard versions is the use of robust estimation =-=[31, 59, 70]-=-. The formulation starts with two sets of point vectors, P = {p i} from image I1 and Q = {qj}sRPI-CS-TR 02-9 6 (a) (b) (c) (d) (e) Figure 3: Illustrating the Dual Bootstrap ICP algorithm in retinal im... |

759 | Alignment by maximization of mutual information
- Viola, Wells
- 1997
(Show Context)
Citation Context ... medical imaging domain exist [31, 41, 47]. Many medical image registration techniques address the problem of accurate alignment of intra- and intermodality images given reasonable starting estimates =-=[31, 45, 82]-=-. Other research in medical image registration focuses on the deformations necessary to align images, taking cues from physics-based models [47, 49]. In many applications such as retinal image registr... |

705 | A survey of image registration techniques
- Brown
- 1992
(Show Context)
Citation Context ...egions, the two images are considered to be accurately aligned and registration succeeds. 2 Background 2.1 Approaches to Registration Registration is a fundamental problem in automatic image analysis =-=[10]-=-. A number of useful surveys of registration within the medical imaging domain exist [31, 41, 47]. Many medical image registration techniques address the problem of accurate alignment of intra- and in... |

585 |
Robust statistics: The approach based on influence functions
- Hampel, Ronchetti, et al.
- 1986
(Show Context)
Citation Context ...ithms. This will provide a formal context for the main contributions of the Dual-Bootstrap ICP algorithm. One difference between this formulation and standard versions is the use of robust estimation =-=[31, 59, 70]-=-. The formulation starts with two sets of point vectors, P = {p i} from image I1 and Q = {qj}sRPI-CS-TR 02-9 6 (a) (b) (c) (d) (e) Figure 3: Illustrating the Dual Bootstrap ICP algorithm in retinal im... |

579 | Hierarchical model-based motion estimation
- Bergen, Anandan, et al.
- 1992
(Show Context)
Citation Context ...e) set. Enriched features and DualBootstrap ICP can be used in combination. • Multiresolution techniques have been used for many years in registration [24, 62, 63] and a variety of other application=-=s [2]-=-. Multiresolution works from coarse descriptions of thesRPI-CS-TR 02-9 30 entire image / data set and (perhaps) simpler models, using the results at coarse resolutions as starting points for finer res... |

572 |
G.: Object modelling by registration of multiple range images
- CHEN, MEDIONI
- 1992
(Show Context)
Citation Context ...oduction 1.1 The Iterative Closest Point Algorithm The iterative closest point (ICP) registration algorithm was invented almost simultaneously in the early 1990’s by several different research group=-=s [3, 12, 14, 49, 78], an-=-d has been used in many different applications since then [22, 43, 49, 55]. ICP is a point-based registration algorithm, where the “points” may be raw measurements such as (x, y, z) values from ra... |

487 | Multimodality image registration by maximization of mutual information
- Maes, Vandermeulen, et al.
- 1997
(Show Context)
Citation Context ... medical imaging domain exist [31, 41, 47]. Many medical image registration techniques address the problem of accurate alignment of intra- and intermodality images given reasonable starting estimates =-=[31, 45, 82]-=-. Other research in medical image registration focuses on the deformations necessary to align images, taking cues from physics-based models [47, 49]. In many applications such as retinal image registr... |

486 | Iterative point matching for registration of free-form curves and surfaces
- Zhang
- 1994
(Show Context)
Citation Context ...oduction 1.1 The Iterative Closest Point Algorithm The iterative closest point (ICP) registration algorithm was invented almost simultaneously in the early 1990’s by several different research group=-=s [3, 12, 14, 49, 78], an-=-d has been used in many different applications since then [22, 43, 49, 55]. ICP is a point-based registration algorithm, where the “points” may be raw measurements such as (x, y, z) values from ra... |

467 |
Multiple View Geometry
- Hartley, Zisserman
- 2000
(Show Context)
Citation Context ...olution methods by applying it at the coarsest resolution. • Robust techniques based on minimal subset random-sampling have been used for difficult registration and transformation estimation problem=-=s [33, 76, 77, 79]-=-, including applications of ICP [13, 46]. The difference that the Dual-Bootstrap ICP algorithm offers is important. Minimal subset random sampling [26, 58] requires a sufficient set of correspondences... |

452 | Efficient variant of the ICP algorithm
- Rusinkiewicz, Levoy
- 2001
(Show Context)
Citation Context ...eometric attributes [35, 38], multiresolution methods [24], and initial matching of distinctive features [16, 19, 67, 74]. The literature on ICP has concentrated on initialization, efficient matching =-=[1, 60]-=-, and applications, while leaving the algorithmic structure unchanged. The motivating observation of this paper is different, however: there are situations where initial estimates alone are not enough... |

452 | Deformable models in medical image analysis: a survey
- McInerney, Terzopoulos
- 1996
(Show Context)
Citation Context ...lity images given reasonable starting estimates [31, 45, 82]. Other research in medical image registration focuses on the deformations necessary to align images, taking cues from physics-based models =-=[47, 49]-=-. In many applications such as retinal image registration the most important issues 3sare initialization, convergence, and robustness to missing and misaligned structures; handling substantial deforma... |

409 | A survey of medical image registration
- Maintz, Viergever
- 1998
(Show Context)
Citation Context ... 2 Background 2.1 Approaches to Registration Registration is a fundamental problem in automatic image analysis [10]. A number of useful surveys of registration within the medical imaging domain exist =-=[31, 41, 47]-=-. Many medical image registration techniques address the problem of accurate alignment of intra- and intermodality images given reasonable starting estimates [31, 45, 82]. Other research in medical im... |

391 | Three-dimensional object recognition from single two-dimensional images
- Lowe
- 1987
(Show Context)
Citation Context ...alization can be addressed in a variety of ways, including image-wide measurements [36, 40], multiresolution [5, 27, 66, 67], indexing and initial matching of distinctive features or sets of features =-=[18, 21, 44, 69, 75]-=-, and minimal-subset (of possible correspondences) randomsampling techniques [28, 34, 64, 78, 79, 88]. The major distinction in minimization methods is between intensity-based and feature-based approa... |

363 |
Least median of squares regression
- Rousseeuw
- 1984
(Show Context)
Citation Context ...obust estimation and use what amounts to a truncated quadratic. Finally, further experiments showed that the use of MUSE scale estimator over a more common estimator such as median absolute deviation =-=[57]-=- improved the effectiveness of the overall algorithm (93.3% and 88.3%). Clearly, these experiments show that all components of the Dual-Bootstrap algorithm are important, with importance increasing su... |

323 |
Distance transformations in digital images
- Borgefors
- 1986
(Show Context)
Citation Context ...tandard ICP technique is to find the closest feature point to p ′ = Mmt( ˆ θmt; p), where p is one of the feature points in Pt selected from Rt. A variety of techniques, such as digital distance m=-=aps [6, 21]-=- in 2D, octree splines [72] and z-buffering [1], may be used to accelerate this process [3, 60]. All of the matches are placed in the correspondence set Ct for this iteration. Given Ct, the new estima... |

322 | Determining the epipolar geometry and its uncertainty: A review
- Zhang
- 1998
(Show Context)
Citation Context ...olution methods by applying it at the coarsest resolution. • Robust techniques based on minimal subset random-sampling have been used for difficult registration and transformation estimation problem=-=s [33, 76, 77, 79]-=-, including applications of ICP [13, 46]. The difference that the Dual-Bootstrap ICP algorithm offers is important. Minimal subset random sampling [26, 58] requires a sufficient set of correspondences... |

241 | Mlesac: a new robust estimator with application to estimating image geometry
- Torr, Zisserman
- 2000
(Show Context)
Citation Context ...olution methods by applying it at the coarsest resolution. • Robust techniques based on minimal subset random-sampling have been used for difficult registration and transformation estimation problem=-=s [33, 76, 77, 79]-=-, including applications of ICP [13, 46]. The difference that the Dual-Bootstrap ICP algorithm offers is important. Minimal subset random sampling [26, 58] requires a sufficient set of correspondences... |

230 |
Geometric Invariance in Computer Vision
- Mundy, Zisserman
- 1992
(Show Context)
Citation Context ...itial estimation [Step 2a)] steps is as follows. Matches between two landmarks, one in each image, or between pairs of landmarks in each image are generated by computing and comparing invariants [8], =-=[55]-=-. Invariants for a single landmark are blood vessel width ratios and blood vessel orientations (Fig. 6), giving a five-component invariant signature vector. 1 The invariant signature of a set of two l... |

220 | The development and comparison of robust methods for estimating the fundamental matrix
- Torr, Murray
- 1997
(Show Context)
Citation Context |

171 | Euclidean distance mapping
- Danielsson
- 1980
(Show Context)
Citation Context ...tandard ICP technique is to find the closest feature point to p ′ = Mmt( ˆ θmt; p), where p is one of the feature points in Pt selected from Rt. A variety of techniques, such as digital distance m=-=aps [6, 21]-=- in 2D, octree splines [72] and z-buffering [1], may be used to accelerate this process [3, 60]. All of the matches are placed in the correspondence set Ct for this iteration. Given Ct, the new estima... |

157 | A new algorithm for non-rigid point matching
- Chui, Rangarajan
- 2000
(Show Context)
Citation Context ...itial correspondences. This implies that the DualBootstrap ICP algorithm can succeed where minimal subset random sampling fails. • The final comparison is to the robust point matching (RPM) algorith=-=m [17]-=-. This uses a soft assignment of point correspondences, smoothing-spline transformations, and deterministic annealing to align point sets. Thus, it globally considers all matches and gradually refines... |

152 |
Robust regression using iteratively reweighted least-squares,” Commun
- Holland, Welsch
- 1977
(Show Context)
Citation Context ...Cauchy loss function and the quadratic loss function, which equates to least-squares estimation. The weight function is used in the iteratively reweighted least-squares implementation of M-estimators =-=[36]. T-=-he Beaton-Tukey is chosen because it most aggressively rejects outliers, providing no weight to matches with normalized distances greater than about 4 standard deviations. • σ is the error scale, w... |

130 | Matching 3-D Anatomical Surfaces with Non-Rigid Deformations using OctreeSplines
- Szeliski, Lavallée
- 1998
(Show Context)
Citation Context ...nd the closest feature point to p ′ = Mmt( ˆ θmt; p), where p is one of the feature points in Pt selected from Rt. A variety of techniques, such as digital distance maps [6, 21] in 2D, octree spli=-=nes [72]-=- and z-buffering [1], may be used to accelerate this process [3, 60]. All of the matches are placed in the correspondence set Ct for this iteration. Given Ct, the new estimate of the transformation pa... |

129 | Robust parameter estimation in computer vision
- Stewart
- 1999
(Show Context)
Citation Context ...ithms. This will provide a formal context for the main contributions of the Dual-Bootstrap ICP algorithm. One difference between this formulation and standard versions is the use of robust estimation =-=[31, 59, 70]-=-. The formulation starts with two sets of point vectors, P = {p i} from image I1 and Q = {qj}sRPI-CS-TR 02-9 6 (a) (b) (c) (d) (e) Figure 3: Illustrating the Dual Bootstrap ICP algorithm in retinal im... |

112 | Robust video mosaicing through topology inference and local to global alignment
- Sawhney, Hsu, et al.
- 1998
(Show Context)
Citation Context ...ion for a given, fixed registration point (feature) set. Enriched features and DualBootstrap ICP can be used in combination. • Multiresolution techniques have been used for many years in registratio=-=n [24, 62, 63]-=- and a variety of other applications [2]. Multiresolution works from coarse descriptions of thesRPI-CS-TR 02-9 30 entire image / data set and (perhaps) simpler models, using the results at coarse reso... |

109 | True multi-image alignment and its application to mosaicing and lens distortion correction
- Sawhney, Kumar
- 1997
(Show Context)
Citation Context ...ion for a given, fixed registration point (feature) set. Enriched features and DualBootstrap ICP can be used in combination. • Multiresolution techniques have been used for many years in registratio=-=n [24, 62, 63]-=- and a variety of other applications [2]. Multiresolution works from coarse descriptions of thesRPI-CS-TR 02-9 30 entire image / data set and (perhaps) simpler models, using the results at coarse reso... |

108 | An automatic registration method for frameless stereotaxy, imageguided surgery, and enhanced reality visualization
- Grimson, Ettinger, et al.
- 1996
(Show Context)
Citation Context ..., 43, 49, 55]. ICP is a point-based registration algorithm, where the “points” may be raw measurements such as (x, y, z) values from range images, intensity points in three-dimensional medical ima=-=ges [24, 29]-=-, and edge elements, corners and interest points [64] that locally summarize the geometric structure of the data. ICP should be used when correspondences between the point sets are not known and when ... |

107 | Registering multiview range data to create 3D computer objects
- Blais, Levine
- 1995
(Show Context)
Citation Context ...s of techniques that have been used to improve ICP: • Several published papers have used genetic algorithms and simulated annealing for a broad search of the landscape of possible parameter estimate=-=s [5, 13, 15, 45]-=-. Closest point correspondences are used to evaluate each parameter estimate, with eventual application of ICP for final refinement. The Dual-Bootstrap ICP makes such coarse search techniques unnecess... |

101 |
Rigid, affine and locally affine registration of free-form surfaces
- Feldmar, Ayache
- 1996
(Show Context)
Citation Context ...n of a point coordinate vector from ℜ n to ℜ n . The mapping may be more general and it may be extended to apply to tangent vectors, normal vectors, and other geometric or even photometric propert=-=ies [23, 39]. �-=-�� d(M(θ; pi), qj) is a distance metric in the coordinate system of I2 between the mapped vector and the corresponding vector qj. 2 The distance metric depends on the types of point vectors. For dete... |

98 | Locating blood vessels in retinal images by piecewise threshold probing of a matched filter response
- Hoover, Kouznetsova, et al.
(Show Context)
Citation Context ...o images to be aligned, multi-image registration plays no role. 6s2.4 Retinal Vascular Feature Extraction Many techniques have been proposed in the research literature for vascular feature extraction =-=[12, 38, 60, 72, 74, 86]. In-=- the implementation of Dual-Bootstrap ICP, we employ an algorithm that extracts elongated structures using two-sided boundary following [1, 12, 29, 81], which we have termed “tracing”. This algori... |

97 | noise, singularities, and scale in height ridge traversal for tubular objects centerline extraction
- Aylward, Initialization
- 2002
(Show Context)
Citation Context ...done using a feature-based method here, but other methods are certainly possible. Of particular interest is the idea of aligning vascular features of one image with the intensity structure of another =-=[2, 3]. -=-We can think of this as a partial feature-based approach. Dual-Bootstrap ICP uses the ICP algorithm, which was invented almost simultaneously in the early 1990’s by several different research groups... |

90 | ICP registration using invariant features
- Sharp, Lee, et al.
- 2002
(Show Context)
Citation Context ...s [30]. Clearly, this assumes much more about the quality of the data than the Dual-Bootstrap ICP algorithm. • By enriching the feature set, matching can be improved, both prior to ICP and during IC=-=P [44, 65, 74]-=-. Both lead to a broader domain of convergence. This approach is complementary to the Dual Bootstrap ICP algorithm, which is designed to maximize the effectiveness of registration for a given, fixed r... |

88 | Medical image analysis: progress over two decades and the challenges ahead
- Duncan, Ayache
(Show Context)
Citation Context ...CP) registration algorithm was invented almost simultaneously in the early 1990’s by several different research groups [3, 12, 14, 49, 78], and has been used in many different applications since the=-=n [22, 43, 49, 55]. IC-=-P is a point-based registration algorithm, where the “points” may be raw measurements such as (x, y, z) values from range images, intensity points in three-dimensional medical images [24, 29], and... |

87 | Registration and Integration of Textured 3-D Data
- Johnson, Kang
- 1997
(Show Context)
Citation Context ...n of a point coordinate vector from ℜ n to ℜ n . The mapping may be more general and it may be extended to apply to tangent vectors, normal vectors, and other geometric or even photometric propert=-=ies [23, 39]. �-=-�� d(M(θ; pi), qj) is a distance metric in the coordinate system of I2 between the mapped vector and the corresponding vector qj. 2 The distance metric depends on the types of point vectors. For dete... |

86 | Comparing and evaluating interest points
- Schmid, Mohr, et al.
- 1998
(Show Context)
Citation Context ...m, where the “points” may be raw measurements such as (x, y, z) values from range images, intensity points in three-dimensional medical images [24, 29], and edge elements, corners and interest poi=-=nts [64]-=- that locally summarize the geometric structure of the data. ICP should be used when correspondences between the point sets are not known and when matching based on the properties of individual points... |

80 | Rapid Automated Tracing and Feature Extraction from Retinal Fundus Images Using Direct Exploratory Algorithms
- Can, Shen, et al.
- 1999
(Show Context)
Citation Context ...nexudative age-related macular degeneration) together with image features used in ICP registration. Panels (a) and (b) show the images, with landmarks extracted by our retinal image tracing algorithm =-=[8, 27]-=-. The landmarks are branching and cross-over points of the retinal vasculature. These are used in initializing ICP. Panels (c) and (d) show the centerline points obtained by the tracing algorithm and ... |

80 |
New Feature Points Based on Geometric Invariants for 3D Image Registration
- Thirion
- 1996
(Show Context)
Citation Context ...ic constraints [54, 55], image-wide measures on the data sets such as statistical moments and geometric attributes [35, 38], multiresolution methods [24], and initial matching of distinctive features =-=[16, 19, 67, 74]-=-. The literature on ICP has concentrated on initialization, efficient matching [1, 60], and applications, while leaving the algorithmic structure unchanged. The motivating observation of this paper is... |

79 |
A Robust Method for Registration and Segmentation of Multiple Range Images
- Masuda, Yokoya
- 1995
(Show Context)
Citation Context ...esolution. • Robust techniques based on minimal subset random-sampling have been used for difficult registration and transformation estimation problems [33, 76, 77, 79], including applications of IC=-=P [13, 46]-=-. The difference that the Dual-Bootstrap ICP algorithm offers is important. Minimal subset random sampling [26, 58] requires a sufficient set of correspondences to generate a full model. When using th... |

79 |
A survey of hierarchical non-linear medical image registration
- Lester, Arridge
- 1999
(Show Context)
Citation Context ... 2 Background 2.1 Approaches to Registration Registration is a fundamental problem in automatic image analysis [10]. A number of useful surveys of registration within the medical imaging domain exist =-=[31, 41, 47]-=-. Many medical image registration techniques address the problem of accurate alignment of intra- and intermodality images given reasonable starting estimates [31, 45, 82]. Other research in medical im... |

76 |
A comparison of similarity measures for use in 2-D-3-D medical image registration
- Penney, Weese, et al.
- 1998
(Show Context)
Citation Context ...mization methods is between intensity-based and feature-based approaches. Intensity-based approaches generally optimize an objective function based on comparison of intensities or intensity gradients =-=[59]-=-, or based on measures such as mutual information [31, 45, 82]. Feature-based techniques align images based on correspondences between automatically detected features [15]. In the retina application, ... |

72 |
3D Free-Form Surface Registration and Object Recognition
- Chua, Jarvis
- 1996
(Show Context)
Citation Context ...ic constraints [54, 55], image-wide measures on the data sets such as statistical moments and geometric attributes [35, 38], multiresolution methods [24], and initial matching of distinctive features =-=[16, 19, 67, 74]-=-. The literature on ICP has concentrated on initialization, efficient matching [1, 60], and applications, while leaving the algorithmic structure unchanged. The motivating observation of this paper is... |

72 |
Alternatives to the median absolute deviation
- Rousseeuw, C
- 1993
(Show Context)
Citation Context ...d transformation estimation problems [33, 76, 77, 79], including applications of ICP [13, 46]. The difference that the Dual-Bootstrap ICP algorithm offers is important. Minimal subset random sampling =-=[26, 58]-=- requires a sufficient set of correspondences to generate a full model. When using the Dual-Bootstrap ICP algorithm, a much weaker initial model may be used, with fewer initial correspondences. This i... |

71 | RANSAC-based DARCES: A New Approach for Fast Automatic Registration of Partially Overlapping Range Images
- Chen, Hung, et al.
- 1999
(Show Context)
Citation Context ...s of techniques that have been used to improve ICP: • Several published papers have used genetic algorithms and simulated annealing for a broad search of the landscape of possible parameter estimate=-=s [5, 13, 15, 45]-=-. Closest point correspondences are used to evaluate each parameter estimate, with eventual application of ICP for final refinement. The Dual-Bootstrap ICP makes such coarse search techniques unnecess... |

66 |
Viergever M. A survey of medical image registration. Medical Image Analysis
- Maintz
- 1998
(Show Context)
Citation Context ...ND A. Approaches to Registration Registration is a fundamental problem in automatic image analysis [10]. A number of useful surveys of registration within the medical imaging domain exist [31], [41], =-=[47]-=-. Many medical image registration techniques address the problem of accurate alignment of intramodality and intermodality images given reasonable starting estimates [31], [45], [82]. Other research in... |