Convolution - Image Processing

I was just following the example from the notes I had.. [I'm assuming you mean S(x) = e^{-ax} and NOT f(x) = e^{-ax}]S(x) = e^{-ax} is the one-dimensional point spread function (PSF) of an optical system. It represents how much a point of light is blurred by the system. In this case, the PSF is a simple exponential that depends on a single parameter, a. The Fourier transform (FT) of the PSF, F[S(x)], is another name for the optical transfer function (OTF) of the system.The optical transfer function (OTF) is defined as the Fourier transform of the point spread function (
  • #1
Hart
169
0

Homework Statement



[tex]I(x)[/tex] is the intensity of an image after passing through a material which
blurs each point according to a point spread function given by:

[tex]S\left(x'-x\right)=e^{-a\left|x'-x\right|}[/tex]

The Fourier transform of [tex]I(x)[/tex] is given by:

[tex]I(k) = \frac{A}{\left( a^{2}+k^{2} \right) \left( b^{2}+k^{2} \right)}[/tex]

Where A is a constant.

(i) Find the Fourier transform [tex]I_{0}^{~}[/tex] of the unblurred image intensity.

(ii) Hence find the original unblurred image intensity [tex]I_{0}x[/tex]


Homework Equations



[tex]I(x') = \int_{-\infty}^{\infty}I_{0}(x)S\left(x'-x\right)dx = \left(I_{0}*S \right)(x')[/tex]

The Attempt at a Solution



[tex]F[I_{0}] = \frac{1}{\sqrt{2 \pi}}\left( \frac{F}{F} \right)[/tex]

(Then can inverse Fourier transform this to get the undistorted image intensity [tex][I_{0}k][/tex])

Calculation of F:

..after some calculations..

[tex] F = \sqrt{\frac{2}{\pi}}\left(\frac{a}{a^{2}+x^{2}} \right) [/tex]

Calculation of F:

[tex]F = A\int_{-\infty}^{\infty}\frac{e^{-ixk}}{\left(a^{2}+k^{2}\right)+\left(b^{2}+k^{2}\right)} dk[/tex]

[tex]F = A\left[\frac{e^{-ixk}}{(-ix)(a^{2}+k^{2})(b^{2}+k^{2})}\right]\right|^{\infty}_{-\infty}[/tex]

[tex]F = A\left[ \frac{2e^{-ix}}{ix(ab)^{2}}\right][/tex]

[tex]F = \left[ \frac{2Ae^{-ix}}{ix(ab)^{2}}\right][/tex]

Therefore can now combine these expressions to get the answer:

[tex]F[I_{0}] = \left( \frac{A\left(iax(ab)^{2}\right)e^{-ix}}{a^{2}+x^{2}} \right)[/tex]

But this looks rather messy, so I assume I've done something wrong somewhere?
 
Physics news on Phys.org
  • #2
Hart said:
Calculation of F:

[tex]F = A\int_{-\infty}^{\infty}\frac{e^{-ixk}}{\left(a^{2}+k^{2}\right)+\left(b^{2}+k^{2}\right)} dk[/tex]

[tex]F = A\left[\frac{e^{-ixk}}{(-ix)(a^{2}+k^{2})(b^{2}+k^{2})}\right]\right|^{\infty}_{-\infty}[/tex]

[tex]F = A\left[ \frac{2e^{-ix}}{ix(ab)^{2}}\right][/tex]

[tex]F = \left[ \frac{2Ae^{-ix}}{ix(ab)^{2}}\right][/tex]



I'm not sure what you are doing here. You are given [itex]I(k)=F[I(x)][/itex], so why are you trying to (inverse?) Fourier transform it and then call the result [itex]F[/itex]?
 
  • #3
To clarify, for the last part of my attempt I used:

[tex]F[I_{0}] = \frac{1}{\sqrt{2 \pi}}\left( \frac{F}{F} \right)[/tex]

.. confused now where to go with this, not really sure how much of my attempt is correct or vaguely right or indeed just completely useless!?
 
  • #4
Again, you are already given the Fourier transform of [itex]I(x)[/itex]

[tex]I(k)\equiv F[/tex]
 
  • #5
[tex]I(k)\equiv F[/tex]


.. where does this come from? I don't recall it being stated before within this thread.


anyways, so:

[tex]I(k) = F = \left[ \frac{2Ae^{-ix}}{ix(ab)^{2}}\right] [/tex]

??

..which is the Fourier transform of I(x).
 
  • #6
Hart said:
.. where does this come from? I don't recall it being stated before within this thread.

Did you not read my first response?

gabbagabbahey said:
... You are given [itex]I(k)=F[I(x)][/itex]...

It certainly looks to me like I said [itex]I(k)=F[I(x)][/itex]


anyways, so:

[tex]I(k) = F = \left[ \frac{2Ae^{-ix}}{ix(ab)^{2}}\right] [/tex]

??

..which is the Fourier transform of I(x).


I'm not sure how to make this any clearer. You are told what [itex]I(k)[/itex] is in your problem statement

[tex]I(k) = \frac{A}{\left( a^{2}+k^{2} \right) \left( b^{2}+k^{2} \right)}[/tex]

[itex]I(k)[/itex] and [itex]F[I(x)][/itex] are two different ways of writing the exact same thing; both represent the Fourier transform of [itex]I(x)[/itex].

[tex]F=I(k)=\frac{A}{\left( a^{2}+k^{2} \right) \left( b^{2}+k^{2} \right)}[/tex]

Does this not make sense to you?
 
  • #7
...

Hart said:
The Fourier transform of [tex]I(x)[/tex] is given by:

[tex]I(k) = \frac{A}{\left( a^{2}+k^{2} \right) \left( b^{2}+k^{2} \right)}[/tex]

Which is equivelent to stating:

[tex]F[I(x)] = I(k) = \frac{A}{\left( a^{2}+k^{2} \right) \left( b^{2}+k^{2} \right)}[/tex]

Ok.. I do get that now.

So now I need to find Fourier transform [tex]I_{0}[/tex].. Assume I don't just set [tex]k=0[/tex] ?
 
  • #8
Hart said:
So now I need to find Fourier transform [tex]I_{0}[/tex].. Assume I don't just set [tex]k=0[/tex] ?

Right, so now you use the convolution theorem:

Hart said:
[tex]F[I_{0}] = \frac{1}{\sqrt{2 \pi}}\left( \frac{F}{F} \right)[/tex]


You now know what [itex]F[/itex] is, so if you calculate [itex]F[/itex], you can use the above convolution theorem. There was nothing wrong with this part of your attempt, however, both your original [itex]F[/itex] and your [itex]F[/itex] were incorrect.

Calculation of F:

..after some calculations..

[tex] F = \sqrt{\frac{2}{\pi}}\left(\frac{a}{a^{2}+x^{2}} \right) [/tex]


You do realize that [itex]F[/itex] is supposed to represent the Fourier transform of [itex]S(x)[/itex], and hence your result should be a function of [itex]k[/itex], not [itex]x[/itex], right?...Clearly, you've done something very wrong in your calculation.

Since you are trying to find the Fourier transform of [itex]S(x)[/itex], a good place to start would be to find [itex]S(x)[/itex]...So,, if [itex]S(x'-x)=e^{-a|x'-x|}[/itex], what is [itex]S(x)[/itex]?.
 
  • #9
Calculation of F:

[tex]f(x) = e^{-\alpha|x|}[/tex]

[tex]\tilda{f}(k) = \frac{1}{\sqrt{2\pi}}\int_{-\infty}^{\infty}f(x)e^{-ikx}dx = \sqrt{\frac{2}{\pi}}\left(\frac{\alpha^{2}}{\alpha^{2}+k^{2}}\right)[/tex]

.. it was actually meant to be with k not x in the final expression, just noticed it when I looked through my calculations again. Hopefully this is actually the correct method and answer.

Then using the Convolution Formula:

[tex]F[I_{0}] = \frac{1}{\sqrt{2 \pi}}\left( \frac{F}{F} \right) = \frac{1}{\sqrt{2 \pi}}\left( \frac{\frac{A}{\left( a^{2}+k^{2} \right) \left( b^{2}+k^{2} \right)}}{\sqrt{\frac{2}{\pi}}\left(\frac{\alpha^{2}}{\alpha^{2}+k^{2}}\right)}\right)[/tex]

[tex]F[I_{0}] = \frac{1}{\sqrt{2 \pi}}\left({\frac{A\sqrt{\frac{2}{\pi}}\left(\frac{\alpha^{2}}{\alpha^{2}+k^{2}}\right)}{\left( a^{2}+k^{2} \right) \left( b^{2}+k^{2} \right)}\right)[/tex]

.. which just needs to be simplified.

Any good?!?
 
  • #10
Hart said:
Calculation of F:

[tex]f(x) = e^{-\alpha|x|}[/tex]



Shouldn't this be

[tex]S(x) = e^{-a|x|}[/tex]

...where are [itex]\alpha[/itex] and [itex]f(x)[/itex] coming from? [itex]F[/itex] is the Fruier transform of [itex]S(x)[/itex], not [itex]f(x)[/itex].
 

1. What is convolution in image processing?

Convolution in image processing is a mathematical operation that involves combining two functions to produce a third function. In image processing, it is used to apply a filter or kernel to an image to enhance or extract certain features.

2. How does convolution work in image processing?

Convolution works by sliding a kernel over an image and calculating the sum of element-wise multiplication between the kernel and the corresponding pixel values in the image. This results in a new image with enhanced or extracted features.

3. What are the applications of convolution in image processing?

Convolution is widely used in image processing for tasks such as image filtering, edge detection, noise reduction, image sharpening, and feature extraction. It is also used in deep learning for tasks such as object detection and image classification.

4. What is the difference between 1D, 2D, and 3D convolution?

1D convolution is used for processing 1-dimensional signals, such as audio signals. 2D convolution is used for processing 2-dimensional signals, such as images. 3D convolution is used for processing 3-dimensional signals, such as video data.

5. What are the limitations of convolution in image processing?

One limitation of convolution in image processing is that it assumes that the kernel is stationary throughout the image, which may not always be the case. Another limitation is that it can be computationally expensive, especially for larger kernels and high-resolution images.

Similar threads

  • Advanced Physics Homework Help
Replies
3
Views
494
  • Advanced Physics Homework Help
Replies
2
Views
907
  • Advanced Physics Homework Help
Replies
4
Views
995
  • Advanced Physics Homework Help
Replies
8
Views
799
  • Advanced Physics Homework Help
Replies
4
Views
913
  • Advanced Physics Homework Help
Replies
3
Views
886
  • Advanced Physics Homework Help
Replies
19
Views
930
  • Advanced Physics Homework Help
Replies
8
Views
1K
  • Advanced Physics Homework Help
Replies
2
Views
819
  • Advanced Physics Homework Help
Replies
10
Views
568
Back
Top