## Fast slant stack: A notion of Radon transform for data in a Cartesian grid which is rapidly computible, algebraically exact, geometrically faithful and invertible (2001)

Venue: | SIAM J. Sci. Comput |

Citations: | 49 - 11 self |

### BibTeX

@ARTICLE{Averbuch01fastslant,

author = {A. Averbuch and R. R. Coifman and D. L. Donoho and M. Israeli and J. Waldén},

title = {Fast slant stack: A notion of Radon transform for data in a Cartesian grid which is rapidly computible, algebraically exact, geometrically faithful and invertible},

journal = {SIAM J. Sci. Comput},

year = {2001}

}

### Years of Citing Articles

### OpenURL

### Abstract

Abstract. We define a notion of Radon Transform for data in an n by n grid. It is based on summation along lines of absolute slope less than 1 (as a function either of x or of y), with values at non-Cartesian locations defined using trigonometric interpolation on a zero-padded grid. The definition is geometrically faithful: the lines exhibit no ‘wraparound effects’. For a special set of lines equispaced in slope (rather than angle), we describe an exact algorithm which uses O(N log N) flops, where N = n2 is the number of pixels. This relies on a discrete projection-slice theorem relating this Radon transform and what we call the Pseudopolar Fourier transform. The Pseudopolar FT evaluates the 2-D Fourier transform on a non-Cartesian pointset, which we call the pseudopolar grid. Fast Pseudopolar FT – the process of rapid exact evaluation of the 2-D Fourier transform at these non-Cartesian grid points – is possible using chirp-Z transforms. This Radon transform is one-to-one and hence invertible on its range; it is rapidly invertible to any degree of desired accuracy using a preconditioned conjugate gradient solver. Empirically, the numerical conditioning is superb; the singular value spread of the preconditioned Radon transform turns out numerically to be less than 10%, and three iterations of the conjugate gradient solver typically suffice for 6 digit accuracy. We also describe a 3-D version of the transform.

### Citations

2033 |
Matrix Computations
- Golub, Loan
- 1989
(Show Context)
Citation Context ...– solution of a Hermitian system by conjugate gradients. Solving Hermitian systems by conjugate gradients is, of course, a central part of modern scientific computing and is extremely well understood =-=[16, 17]-=-. Defining the condition number κ of the Gram operator G = (adj ˜ P) ◦ ˜ P to be the ratio of largest to smallest eigenvalues of G, then the error of the k-th iteration’s approximate solution Ik is bo... |

335 |
Iterative Methods for Solving Linear Systems
- Greenbaum
- 1997
(Show Context)
Citation Context ...– solution of a Hermitian system by conjugate gradients. Solving Hermitian systems by conjugate gradients is, of course, a central part of modern scientific computing and is extremely well understood =-=[16, 17]-=-. Defining the condition number κ of the Gram operator G = (adj ˜ P) ◦ ˜ P to be the ratio of largest to smallest eigenvalues of G, then the error of the k-th iteration’s approximate solution Ik is bo... |

208 |
The Radon Transform and Some of its Applications
- Deans
- 1983
(Show Context)
Citation Context ...on of this article 1s1 Introduction The Radon Transform is a fundamental tool in a wide range of disciplines, including radar imaging, geophysical imaging, nondestructive testing, and medical imaging =-=[10]-=-. In these areas, the Radon transform is typically discussed in continuum terms, as a mapping from functions f(x, y) with (x, y) ∈ R 2 to a function Rf(t, θ) with t ∈ R and θ ∈ [0,π), defined by � Rf(... |

111 |
Image Reconstruction from Projections
- Herman
- 1980
(Show Context)
Citation Context ... a broken line, with a break at the transition angle θ = π/4. Another field in which this continuum transform has been (independently) developed is medical tomography, where Sf is called the Linogram =-=[12, 13]-=-, in reference to the fact that points map under S into broken lines, whereas in the usual Radon transform, points map into sinusoids; because of this in medical tomography, the usual Radon transform ... |

46 |
A linear filtering approach to the computation of the discrete Fourier transform
- Bluestein
- 1970
(Show Context)
Citation Context ...denotes the usual inverse FFT (up to normalization). Lemma 3 Let X be a vector with 2n entries. The 2n entries of FαX can be evaluated in O(n log(n)) flops. This lemma has been known since the 1960’s =-=[26, 5]-=-; the Chirp-Z transform is the most well-known way to obtain it. An extensive discussion and a convenient algorithm for solving this problem using the fractional Fourier transform terminology can be f... |

45 | The fractional Fourier transform and applications
- Bailey, Swarztrauber
- 1991
(Show Context)
Citation Context ...w medical scanners and the search for extraterrestrials. Although in writing this paper we were initially unaware of earlier work in this direction we became aware of work of Bailey and Swartztrauber =-=[2, 3]-=-, Pasciak [25], Edholm and Herman [12, 13], and Lawton [19]. Each of these cited works makes a major algorithmic contribution, which contains within it key ingredients which could be used to establish... |

41 |
Convolution-based interpolation for fast, high quality rotation of images
- Unser, Thevenaz, et al.
- 1995
(Show Context)
Citation Context ...ously with roles of v and u reversed, and with the names ‘column’ and ‘row’ reversed. See Figure 3. For more discussion of shearing using trigonometric interpolation, see Unser, Thevenaz, Yaroslavsky =-=[28]-=-. We have the following formal equivalence; see Figure 4 for a graphical illustration. Lemma 1 For each θ, summing along lines as in the Radon Transform produces the same result as shearing the array ... |

35 |
Digital reconstruction of multidimensional signals from their projections
- Merseraeau, Oppenheim
- 1974
(Show Context)
Citation Context ...n, |k| indexes (pseudo-) radius and ℓ indexes angle. The level sets of (pseudo-) radius are squares; so this grid may be called a Concentric Squares Grid; it was introduced by Mersereau and Oppenheim =-=[22]-=-. Obtaining the values of the Fourier transform at this pointset will be so important for us that we formalize the operation. Definition. The Pseudopolar Fourier Transform P is the linear transformati... |

33 |
Signal Recovery and the Large Sieve
- Donoho, Logan
- 1992
(Show Context)
Citation Context ... =1of the pseudopolar domain is at most √ 2 times as large as �Î�. Similarly for panel s = 2. Hence || ˜ PI|| ≤ √ 8�Î�, ∀I. We now prove the lemma. We use a large sieve inequality of Donoho and Logan =-=[11]-=-. This says that if S(τ)= �m−1 k=0 ck exp{i2π m kτ} is a trigonometric polynomial of degree m and period 1, and if µ is nonnegative measure on [0,1), then for each δ>0 � 1 0 |S| 2 (τ)dµ(τ) ≤ (n − 1+δ ... |

33 | The Gridding method for image reconstruction by Fourier transformation - Schomberg, Timmer - 1995 |

26 |
Discrete Radon transform
- Beylkin
- 1987
(Show Context)
Citation Context ... to have analogs of Rf for arrays of digital data I = (I(u, v):−n/2 ≤ u, v < n/2). This has been the object of attention of many authors over the last twenty years; a very large literature has ensued =-=[4, 6, 7, 20, 21, 18]-=-. Despite many attempts at defining a “digital Radon transform”, we believe that there is at present no definition which is both intellectually and practically satisfying. 1.1 Desiderata To support ou... |

21 |
A fast discrete approximation algorithm for the Radon transform
- Brady
- 1998
(Show Context)
Citation Context ... to have analogs of Rf for arrays of digital data I = (I(u, v):−n/2 ≤ u, v < n/2). This has been the object of attention of many authors over the last twenty years; a very large literature has ensued =-=[4, 6, 7, 20, 21, 18]-=-. Despite many attempts at defining a “digital Radon transform”, we believe that there is at present no definition which is both intellectually and practically satisfying. 1.1 Desiderata To support ou... |

18 |
Image representation via a finite Radon transform
- Matúˇs, Flusser
- 1993
(Show Context)
Citation Context ... to have analogs of Rf for arrays of digital data I = (I(u, v):−n/2 ≤ u, v < n/2). This has been the object of attention of many authors over the last twenty years; a very large literature has ensued =-=[4, 6, 7, 20, 21, 18]-=-. Despite many attempts at defining a “digital Radon transform”, we believe that there is at present no definition which is both intellectually and practically satisfying. 1.1 Desiderata To support ou... |

18 |
The chirp z-transform algorithm and its applications,” Bell Sys
- Rabiner, Schafer, et al.
- 1969
(Show Context)
Citation Context ...denotes the usual inverse FFT (up to normalization). Lemma 3 Let X be a vector with 2n entries. The 2n entries of FαX can be evaluated in O(n log(n)) flops. This lemma has been known since the 1960’s =-=[26, 5]-=-; the Chirp-Z transform is the most well-known way to obtain it. An extensive discussion and a convenient algorithm for solving this problem using the fractional Fourier transform terminology can be f... |

17 |
A fast digital Radon transform – an efficient means for evaluating the Hough transform
- GÖTZ, J
- 1996
(Show Context)
Citation Context ...ransform should turn out to be invertible (which may be very difficult to determine) the transform is typically not invertible by any straightforward algorithm. • Multiscale Approaches. Other authors =-=[6, 7, 8, 15]-=- have attempted to exploit two-scale relations, which say that if one knows the Radon transform over four dyadic subsquares of a dyadic square these can be combined to obtain the Radon transform over ... |

14 |
Fast calculation of multiple line integrals
- Brandt, Dym
- 1999
(Show Context)
Citation Context |

13 |
A new polar Fourier-transform for computeraided tomography and spotlight synthetic aperture radar
- Lawton
- 1988
(Show Context)
Citation Context ...ough in writing this paper we were initially unaware of earlier work in this direction we became aware of work of Bailey and Swartztrauber [2, 3], Pasciak [25], Edholm and Herman [12, 13], and Lawton =-=[19]-=-. Each of these cited works makes a major algorithmic contribution, which contains within it key ingredients which could be used to establish Theorems 2 and 4 above. We believe that prior to this pape... |

12 | A fast and accurate multilevel inversion of the Radon transform
- Brandt, Mann, et al.
- 1999
(Show Context)
Citation Context ...ransform should turn out to be invertible (which may be very difficult to determine) the transform is typically not invertible by any straightforward algorithm. • Multiscale Approaches. Other authors =-=[6, 7, 8, 15]-=- have attempted to exploit two-scale relations, which say that if one knows the Radon transform over four dyadic subsquares of a dyadic square these can be combined to obtain the Radon transform over ... |

9 | Direct-Fourier reconstruction in tomography and synthetic aperture radar, Int - Choi, Jr, et al. - 1998 |

8 |
Image reconstruction from linograms: Implementation and evaluation
- Edholm, Herman, et al.
- 1988
(Show Context)
Citation Context ... a broken line, with a break at the transition angle θ = π/4. Another field in which this continuum transform has been (independently) developed is medical tomography, where Sf is called the Linogram =-=[12, 13]-=-, in reference to the fact that points map under S into broken lines, whereas in the usual Radon transform, points map into sinusoids; because of this in medical tomography, the usual Radon transform ... |

8 |
A Note on the Fourier Algorithm for Image Reconstruction
- Pasciak
- 1973
(Show Context)
Citation Context ...rs and the search for extraterrestrials. Although in writing this paper we were initially unaware of earlier work in this direction we became aware of work of Bailey and Swartztrauber [2, 3], Pasciak =-=[25]-=-, Edholm and Herman [12, 13], and Lawton [19]. Each of these cited works makes a major algorithmic contribution, which contains within it key ingredients which could be used to establish Theorems 2 an... |

5 |
The discrete periodic Radon transform
- Hsung, Lun, et al.
- 1996
(Show Context)
Citation Context |

5 | Fourier reconstruction in tomography - Natterer - 1985 |

3 |
Schnelle Fourier-tranformation bei nichtäquidistanten Gittern und tomographische anwendungen
- Fourmont
- 1999
(Show Context)
Citation Context ...mplete discussion would take space we do not have, so we content ourselves with three examples, which also help to illustrate the meaning of our desiderata above. 2s• Fourier Approaches. Some authors =-=[14, 20]-=- have attempted to exploit the projection-slice theorem, which says that the 1-dimensional constant-θ slice of the Radon transform ((Rf)(t, θ):−∞ <t<∞) and the 1-dimensional radial slice of the Fourie... |

2 |
Linogram Reconstruction for Magnetic Resonance Imaging
- Axel, Herman, et al.
- 1990
(Show Context)
Citation Context ...rms of the samples in the Cartesian grid: �T �Cart = � n� k=−n |T(k)| 2 � 1/2 and second, the density-weighted norm of the samples in the pseudopolar grid. We have the following: Lemma 5 For each α ∈ =-=[0,1]-=-, ⎛ �T �Frac = √ α · ⎝ n/2 � k=−n/2 |T(αk)| 2 �T �Frac ≤ √ 2 ·�T �Cart. It follows immediately from this that the density-weighted norm of the samples lying in panel s =1of the pseudopolar domain is a... |

2 |
The Discrete Radon Transform: Part I – Theory
- Kelley, Madisetti
- 1993
(Show Context)
Citation Context |

1 |
Efficient detection of a continuous-wave signal with a linear frequency drift
- Swarztrauber, Bailey
(Show Context)
Citation Context ...olation. They then show that, using the fractional Fourier transform, one can calculate rapidly the 24sarray of sums over all lines with slopes β · k/n and offsets γt, for constants β and γ. See also =-=[3]-=-. Obviously this comes close to containing Theorem 2. However, Bailey and Swartztrauber do not explicitly define or defend a notion “Radon transform”; they seem merely interested in getting sums over ... |

1 | A fast sinc gridding algorithm for Fourier inversion in computer tomography - O’Sullivan - 1985 |