2. Convolution
Fourier Convolution
Outline
· Review linear imaging model
· Instrument response function vs Point spread function
· Convolution integrals
• Fourier Convolution
• Reciprocal space and the Modulation transfer function
· Optical transfer function
· Examples of convolutions
· Fourier filtering
· Deconvolution
· Example from imaging lab
• Optimal inverse filters and noise
admission.edhole.com
22.058 - lecture 4, Convolution and Fourier Convolution
3. Instrument Response Function
The Instrument Response Function is a conditional mapping, the form of the map
depends on the point that is being mapped.
{ }
IRF(x,y|x0,y0)=Sd(x-x0 )
This is often given the symbol h(r|r’).
Of course we want the entire output from the whole object function,
¥
òò
-¥
¥
òò
-¥
and so we need to know the IRF at all points.
22.058 - lecture 4, Convolution and Fourier Convolution
d(y-y0 )
E(x,y)=
¥
òò
-¥
I(x,y)S{d(x-x0)d(y-y0)}dxdydx0dy0
E(x,y)=
¥
òò
-¥
I(x,y)IRF(x,y|x0,y0)dxdydx0dy0
admission.edhole.com
4. Space Invariance
Now in addition to every point being mapped independently onto the detector,
imaging that the form of the mapping does not vary over space (is independent of
r0). Such a mapping is called isoplantic. For this case the instrument response
function is not conditional.
IRF(x,y|x0,y0)=PSF(x-x0,y-y0 )
The Point Spread Function (PSF) is a spatially invariant approximation of the IRF.
admission.edhole.com
22.058 - lecture 4, Convolution and Fourier Convolution
5. Space Invariance
Since the Point Spread Function describes the same blurring over the entire sample,
IRF(x,y|x0,y0)ÞPSF(x-x0,y-y0 )
The image may be described as a convolution,
or,
¥
òò
-¥
E(x,y)=
I(x0,y0)PSF(x-x0,y-y0)dx0dy0
Image(x,y)=Object(x,y)ÄPSF(x,y)+noise
admission.edhole.com
22.058 - lecture 4, Convolution and Fourier Convolution
6. Convolution Integrals
Let’s look at some examples of convolution integrals,
¥ òg(x')h(x-x')dx'
f(x)=g(x)Äh(x)=
So there are four steps in calculating a convolution integral:
#1. Fold h(x’) about the line x’=0
#2. Displace h(x’) by x
#3. Multiply h(x-x’) * g(x’)
#4. Integrate
22.058 - lecture 4, Convolution and Fourier Convolution
-¥
admission.edhole.com
7. Convolution Integrals
Consider the following two functions:
#1. Fold h(x’) about the line x’=0
#2. Displace h(x’) by x x
admission.edhole.com
22.058 - lecture 4, Convolution and Fourier Convolution
11. Some Properties of the Convolution
commutative:
associative:
fÄg=gÄf
fÄ(gÄh)=(fÄg)Äh
multiple convolutions can be carried out in any order.
distributive:
fÄ(g+h)=fÄg+fÄh
admission.edhole.com
22.058 - lecture 4, Convolution and Fourier Convolution
12. Convolution Integral
Recall that we defined the convolution integral as,
¥ò
fÄg=f(x)g(x'-x)dx
-¥
One of the most central results of Fourier Theory is the convolution
theorem (also called the Wiener-Khitchine theorem.
where,
Á{fÄg}=F(k)×G(k)
f(x)ÛF(k)
g(x)ÛG(k)
admission.edhole.com
22.058 - lecture 4, Convolution and Fourier Convolution
13. Convolution Theorem
Á{fÄg}=F(k)×G(k)
In other words, convolution in real space is equivalent to
multiplication in reciprocal space.
admission.edhole.com
22.058 - lecture 4, Convolution and Fourier Convolution
14. Convolution Integral Example
We saw previously that the convolution of two top-hat functions
(with the same widths) is a triangle function. Given this, what is
the Fourier transform of the triangle function?
15
10
5
-3 -2 -1 1 2 3
22.058 - lecture 4, Convolution and Fourier Convolution
=
15 ?
10
5
-3 -2 -1 1 2 3
admission.edhole.com
15. Proof of the Convolution Theorem
¥ò
fÄg=f(x)g(x'-x)dx
The inverse FT of f(x) is,
-¥
¥ò
f(x)=1
F(k)eikxdk
-¥
and the FT of the shifted g(x), that is g(x’-x)
22.058 - lecture 4, Convolution and Fourier Convolution
2p
¥ò
g(x'-x)=1
G(k')eik'(x'-x)dk'
-¥
2p
admission.edhole.com
16. Proof of the Convolution Theorem
So we can rewrite the convolution integral,
as,
¥ò
fÄg=f(x)g(x'-x)dx
-¥
¥ò
fÄg=1
¥ò
¥ò
dxF(k)eikxdk
4p2
change the order of integration and extract a delta function,
¥ò
¥ò
dkF(k)dk'G(k')eik'x'1
-¥ 14 4 2 4 4 3
22.058 - lecture 4, Convolution and Fourier Convolution
-¥
-¥
G(k')eik'(x'-x)dk'
-¥
fÄg=1
2p
eix(k-k')dx
2p
-¥
d(k-k')
¥ò
-¥
admission.edhole.com
17. Proof of the Convolution Theorem
or,
¥ò
fÄg=1
¥ò
dkF(k)dk'G(k')eik'x'1
2p
eix(k-k')dx
-¥ 14 4 2 4 4 3
¥ò
¥ò
dkF(k)dk'G(k')eik'x'd
(k-k')
Integration over the delta function selects out the k’=k value.
22.058 - lecture 4, Convolution and Fourier Convolution
2p
-¥
d(k-k')
¥ò
-¥
fÄg=1
2p
-¥
-¥
¥ò
fÄg=1
dkF(k)G(k)eikx'
-¥
2p
admission.edhole.com
18. Proof of the Convolution Theorem
¥ò
fÄg=1
dkF(k)G(k)eikx'
-¥
2p
This is written as an inverse Fourier transformation. A Fourier
transform of both sides yields the desired result.
Á{fÄg}=F(k)×G(k)
admission.edhole.com
22.058 - lecture 4, Convolution and Fourier Convolution
20. Reciprocal Space
real space reciprocal space
admission.edhole.com
22.058 - lecture 4, Convolution and Fourier Convolution
21. Filtering
We can change the information content in the image by
manipulating the information in reciprocal space.
Weighting function in k-space.
admission.edhole.com
22.058 - lecture 4, Convolution and Fourier Convolution
22. Filtering
We can also emphasis the high frequency components.
Weighting function in k-space.
admission.edhole.com
22.058 - lecture 4, Convolution and Fourier Convolution
23. Modulation transfer function
i(x,y) = o(x,y)ÄPSF(x,y) + noise
c c c c
I(kx,ky)=O(kx,ky)×MTF(kx,ky)+Á{noise}
¥
òò
-¥
E(x,y)=
¥
òò
-¥
I(x,y)S{d(x-x0)d(y-y0)}dxdydx0dy0
¥
òò
-¥
E(x,y)=
¥
òò
-¥
I(x,y)IRF(x,y|x0,y0)dxdydx0dy0
admission.edhole.com
22.058 - lecture 4, Convolution and Fourier Convolution
24. Optics with lens
‡ Input bit mapped image
sharp =Import @" sharp . bmp" D;
Shallow @InputForm @sharp DD
Graphics@Raster@<<4>>D,Rule@<<2>>DD
s =sharp @@1, 1DD;
Dimensions @s D
8480,640<
ListDensityPlot @s , 8PlotRange ÆAll , Mesh ÆFalse <D admission.edhole.com
22.058 - lecture 4, Convolution and Fourier Convolution