The relationship between the Pascal Triangle and the Radon …mboutin/News_files/... ·...
Transcript of The relationship between the Pascal Triangle and the Radon …mboutin/News_files/... ·...
The relationship between the Pascal Triangle and the Radon Transform of an image �
Copyright 2012, all rights reserved.� 1
Copyright 2012, all rights reserved.� 2
3 Copyright 2012, all rights reserved.
1. The Pascal Triangle of a gray scale image 1.1 Inspiration
Shanshan Huang and I invented the Pascal Triangle in order to gain a geometric interpretation of the content of a gray scale image. We were mostly interested in the geometric features of simple objects, such as
• Dots, • Straight line segments, • Loops, • Crosses • Etc.
Our inspiration was the concept of principal curvatures in differential geometry, which can be used to classify surfaces. However, we did not want to rely on derivatives, as these are ill-suited for application to discrete images. Rather, we worked with moments, a discrete analogue, which we could define directly in terms of the data given in the image.
The Pascal triangle of an image is a pyramidal arrangement of moments.
4 Mathematics in Science and Society Lecture, UIUC, Nov. 13, 2012 Copyright 2012, all rights reserved. 4
1. The Pascal Triangle of a gray scale image 1.2 Definition
Start with a gray scale image with
pixel locations (xk,yk)
pixel intensity wk
for k = 1, …, N
(x1,y1)=(1,1)
Example: N=9 pixels
5 Copyright 2012, all rights reserved.
1. The Pascal Triangle of a gray scale image 1.2 Definition
Start with a gray scale image with
pixel locations (xk,yk)
pixel intensity wk
for k = 1, …, N
(x2,y2)=(1,2)
Example: N=9 pixels
6 Copyright 2012, all rights reserved.
1. The Pascal Triangle of a gray scale image 1.2 Definition
Start with a gray scale image with
pixel locations (xk,yk)
pixel intensity wk
for k = 1, …, N
(x3,y3)=(1,3) Example: N=9 pixels
7 Copyright 2012, all rights reserved.
1. The Pascal Triangle of a gray scale image 1.2 Definition
Start with a gray scale image with
pixel locations (xk,yk)
pixel intensity wk
for k = 1, …, N
w1=0
Example: N=9 pixels
8 Copyright 2012, all rights reserved.
1. The Pascal Triangle of a gray scale image 1.2 Definition
Start with a gray scale image with
pixel locations (xk,yk)
pixel intensity wk
for k = 1, …, N
w1=255
Example: N=9 pixels
9 Copyright 2012, all rights reserved.
1. The Pascal Triangle of a gray scale image 1.2 Definition
Start with a gray scale image with
pixel locations (xk,yk)
pixel intensity wk
for k = 1, …, N
w1=53
Example: N=9 pixels
10 Copyright 2012, all rights reserved.
1. The Pascal Triangle of a gray scale image 1.2 Definition
Write pixel location using complex numbers
(xk,yk) → zk= xk+ i yk
for k = 1, …, N
11 Copyright 2012, all rights reserved.
1. The Pascal Triangle of a gray scale image 1.2 Definition
Write pixel location using complex numbers
(xk,yk) → zk= xk+ i yk
for k = 1, …, N
Compute complex moments
!
µmn = wkzkm
k=1
N
" zk( )n
12 Copyright 2012, all rights reserved.
1. The Pascal Triangle of a gray scale image 1.2 Definition
We form the Pascal triangle of the image by arranging the moments this way:
Moments of order two
13 Copyright 2012, all rights reserved.
1. The Pascal Triangle of a gray scale image 1.2 Definition
We form the Pascal triangle of the image by arranging the moments this way:
Moments of order three
14 Copyright 2012, all rights reserved.
1. The Pascal Triangle of a gray scale image 1.2 Definition
15 Mathematics in Science and Society Lecture, UIUC, Nov. 13, 2012 Copyright 2012, all rights reserved. 15
1. The Pascal Triangle of a gray scale image 1.3 Relation to Radon transform
16 Mathematics in Science and Society Lecture, UIUC, Nov. 13, 2012
The Radon transform fθ( r ) is the projection of the image intensities onto a straight line through the origin with direction vector eiθ.
z1 z2 z3
Copyright 2012, all rights reserved. 16
1. The Pascal Triangle of a gray scale image 1.3 Relation to Radon transform
17 Mathematics in Science and Society Lecture, UIUC, Nov. 13, 2012
The Radon transform fθ( r ) is the projection of the image intensities onto a straight line through the origin with direction vector eiθ.
θ
Copyright 2012, all rights reserved. 17
1. The Pascal Triangle of a gray scale image 1.3 Relation to Radon transform
18 Mathematics in Science and Society Lecture, UIUC, Nov. 13, 2012
The Radon transform fθ( r ) is the projection of the image intensities onto a straight line through the origin with direction vector eiθ.
θ
!
f" (r) = wkks.t.r=xkcox" +yk sin"
#
Copyright 2012, all rights reserved. 18
1. The Pascal Triangle of a gray scale image 1.3 Relation to Radon transform
19 Mathematics in Science and Society Lecture, UIUC, Nov. 13, 2012
Denote by mn(θ) the nth order moment of the Radon transform fθ( r ).
!
mn (") = rnrs.t .f" (r )#0
$ f" (r) = (xk cos" + yk sin")n
k=1
N
$ wk
Lemma: For any n=0,1, 2, …
!
mn (") =12n
nl#
$ % &
' ( µl,n) le
i(n)2l )"
l=0
n
*
Copyright 2012, all rights reserved. 19
1. The Pascal Triangle of a gray scale image 1.3 Relation to Radon transform
Row r of the Pascal triangle corresponds to the Fourier series coefficients of the rth order moment of the Radon transform:
Coefficients of Fourier series of Third order moment of Radon transform of the image
20 Copyright 2012, all rights reserved.
1. The Pascal Triangle of a gray scale image 1.3 Relation to Radon transform
Coefficients of the Fourier series of the Second order moment of Radon transform of image
21 Copyright 2012, all rights reserved.
Row r of the Pascal triangle corresponds to the Fourier series coefficients of the rth order moment of the Radon transform:
1. The Pascal Triangle of a gray scale image 1.3 Relation to Radon transform
22 Copyright 2012, all rights reserved.
I = { (zk,wk) }k=1,…,N
T2N-2 (I) { Mn(θ) }n=0,…,2N-2
fθ (r)
1. The Pascal Triangle of a gray scale image 1.3 Relation to Radon transform
For more details:
M. Boutin, S. Huang, “The Pascal Triangle of a discrete image: definition, properties, and application to shape analysis,” available at http://arxiv.org/abs/1209.4850
A.W. Haddad, S. Huang, M. Boutin and E. J. Delp, ``Detection of symmetric shapes on a mobile device with applications to automatic sign interpretation,'' Proc. of SPIE-IS&T Electronic Imaging 8304, p. 83040G, 2012.
23 Copyright 2012, all rights reserved.
1. The Pascal Triangle of a gray scale image 1.5 Conclusion
We are grateful for the support provided by the following funding agencies:
NSF Department of Homeland Security Indiana 21st century fund DARPA