Editing the FFT of a jpeg

[Robby Connor]

I am trying to load an jpeg into ImageJ and run an FFT on it to edit out a
known pattern I have.  The problem I am having is after I edit the
transformed image and inverse FFT nothing has changed.  I can tell my edit
has no effect because I can black out the entire frequency-domain picture
and run an inverse FFT on it, which will result in my exact same original
picture.  

The weirdest thing about this is that it actually worked for about 2 days on
one of my computers, then stopped working (it never worked on the other
computer).  When it was working, here were my steps:
1)convert to 8-bit black and white under Image>>Type.
2)change my editing background paint color to black.
3)run the Process>>FFT>>FFT on the picture.
4)select circular areas on the transformed picture relating to the
frequencies of the grid pattern of the original, then press backspace.  This
deleted the area and left a black color due to step 2).
5)run the Process>>FFT>>Inverse FFT.

This had been working, then spontaneously stopped working.  I had been
trying to check if any settings were different from the computer whose
ImageJ Inverse FFT worked to the one that didn't, so I could've accidentally
changed something critical.  

MAIN QUESTION:
How do I get this feature to work again?

SECONDARY QUESTION:
Is there any other program (or an add-on to ImageJ) that anybody knows about
that would do this process for me?

EXTRA INFO:
I run XP on both computers.

[Joel Sheffield]

I generally set the foreground to black --you have to be sure that your
black is truly black (value 0).  I then select the area, as you do, and fill
the area (ctl-f).  I have found that if I have been working in an RGB world,
sometimes the black isn't true black, and then there are problems.

Joel


On Fri, Sep 18, 2009 at 6:02 PM, Robby Connor
<[hidden email]>wrote:

> I am trying to load an jpeg into ImageJ and run an FFT on it to edit out a
> known pattern I have.  The problem I am having is after I edit the
> transformed image and inverse FFT nothing has changed.  I can tell my edit
> has no effect because I can black out the entire frequency-domain picture
> and run an inverse FFT on it, which will result in my exact same original
> picture.
>
> The weirdest thing about this is that it actually worked for about 2 days
> on
> one of my computers, then stopped working (it never worked on the other
> computer).  When it was working, here were my steps:
> 1)convert to 8-bit black and white under Image>>Type.
> 2)change my editing background paint color to black.
> 3)run the Process>>FFT>>FFT on the picture.
> 4)select circular areas on the transformed picture relating to the
> frequencies of the grid pattern of the original, then press backspace.
>  This
> deleted the area and left a black color due to step 2).
> 5)run the Process>>FFT>>Inverse FFT.
>
> This had been working, then spontaneously stopped working.  I had been
> trying to check if any settings were different from the computer whose
> ImageJ Inverse FFT worked to the one that didn't, so I could've
> accidentally
> changed something critical.
>
> MAIN QUESTION:
> How do I get this feature to work again?
>
> SECONDARY QUESTION:
> Is there any other program (or an add-on to ImageJ) that anybody knows
> about
> that would do this process for me?
>
> EXTRA INFO:
> I run XP on both computers.
>

--


Joel B. Sheffield, Ph.D
Department of Biology
Temple University
Philadelphia, PA 19122
Voice: 215 204 8839
e-mail: [hidden email]
URL:  http://astro.temple.edu/~jbs

[Gabriel Landini]

On Friday 18 September 2009, Robby Connor wrote:
> I am trying to load an jpeg into ImageJ and run an FFT on it to edit out a
> known pattern I have.

I cannot reproduce it. Works fine here on 1.43g.

Jpegs (obviously the lossy variety, which are the vast majority) are a *very*
bad choice to do imaging. Avoid them...

> 4)select circular areas on the transformed picture relating to the
> frequencies of the grid pattern of the original, then press backspace.

As posted earlier make sure that black is really black.
Also make sure that you are not using an inverted LUT (where black is actually
255).

Cheers,
G.

[Robby Connor]

So I tried filling with cntrl+f instead of deleting and managed to instantly
get it to work on both of my computers.  Then, on one of my computers I
tried to screw around a bit and changed the type of the picture from 8-bit
to 32-bit, and it stopped working.  It didn't work when I switched it back
to 8-bit either.  Black is 0, not 255, so that doesn't seem to be the
problem.  I'm not sure if checking the color LUT is the only test for
finding if ImageJ is filling with true black.  I don't know why ImageJ is
being so finicky, but I am writing a procedure for other people to follow
using this method, so I need to know as much as I can about this process.
Does anyone have suggestions of what the problem is and how to fix it?  Thanks.

[Gabriel Landini]

On Monday 21 September 2009 19:26:46 Robby Connor wrote:
> So I tried filling with cntrl+f instead of deleting and managed to
>  instantly get it to work on both of my computers.  Then, on one of my
>  computers I tried to screw around a bit and changed the type of the
>  picture from 8-bit to 32-bit, and it stopped working.  It didn't work when
>  I switched it back to 8-bit either.

Cannot reproduce. Works fine here in 8, 16, 32bit, in rgb and in 8bit
greyscale with the spectrum lut too.

Can you write a macro that works in one computer but fails in another?

>  I don't know
>  why ImageJ is being so finicky, but I am writing a procedure for other
>  people to follow using this method, so I need to know as much as I can
>  about this process.

Well, it isn't a problem that I can reproduce, so I guess that it might be
some problem in your installation.
What does "Help>About ImageJ" report? And in the machine that works fine?
What type of computer are you running this? What CPU?
Have you added anything to the StartupMacros.txt file?

Cheers
G.

[Joel Sheffield]

From:           Joel Sheffield <[hidden email]>
To:             imageJ Interest Group <[hidden email]>
Subject:         Re: Editing the FFT of a jpeg
Send reply to:   [hidden email]
Date sent:       Mon, 21 Sep 2009 15:36:49 -0400

From:           Joel Sheffield <[hidden email]>
To:             Robby Connor <[hidden email]>
Subject:         Re: Editing the FFT of a jpeg
Send reply to:   [hidden email]
Date sent:       Mon, 21 Sep 2009 15:21:04 -0400

You can tell if the system is reading true black by passing our
coursor over the area and then looking at the menu bar.  You will
see
the xy coordinates of the cursor, and the value of the pixel at that
location.  I wonder if when you switch to 32 bit, your black is not
truly 0.000000.

Joel


> So I tried filling with cntrl+f instead of deleting and managed to
> instantly
> get it to work on both of my computers.  Then, on one of my
> computers I
> tried to screw around a bit and changed the type of the picture
from

> 8-bit
> to 32-bit, and it stopped working.  It didn't work when I switched
> it back
> to 8-bit either.  Black is 0, not 255, so that doesn't seem to be
> the
> problem.  I'm not sure if checking the color LUT is the only test
> for
> finding if ImageJ is filling with true black.  I don't know why
> ImageJ is
> being so finicky, but I am writing a procedure for other people to
> follow
> using this method, so I need to know as much as I can about this
> process.
> Does anyone have suggestions of what the problem is and how to fix
> it?  Thanks.

------- End of forwarded message -------

Attachments:
  C:\DOCUME~1\Joel\LOCALS~1\Temp\WPM$387A.PM$
------- End of forwarded message -------
--
Joel B. Sheffield, Ph.D.
Biology Department, Temple University
1900 North 12th Street
Philadelphia, PA 19122
[hidden email]  
[phone], fax [phone]
http://astro.temple.edu/~jbs

[Joel Sheffield]

Sorry if this came through in an odd format before. mailer problems --Joel

---------- Forwarded message ----------
From: Joel Sheffield <[hidden email]>
Date: Mon, Sep 21, 2009 at 3:20 PM
Subject: Re: Editing the FFT of a jpeg
To: Robby Connor <[hidden email]>


You can tell if the system is reading true black by passing our
coursor over the area and then looking at the menu bar.  You will see
the xy coordinates of the cursor, and the value of the pixel at that
location.  I wonder if when you switch to 32 bit, your black is not
truly 0.000000.

Joel

> So I tried filling with cntrl+f instead of deleting and managed to
> instantly
> get it to work on both of my computers.  Then, on one of my
> computers I
> tried to screw around a bit and changed the type of the picture from
> 8-bit
> to 32-bit, and it stopped working.  It didn't work when I switched
> it back
> to 8-bit either.  Black is 0, not 255, so that doesn't seem to be
> the
> problem.  I'm not sure if checking the color LUT is the only test
> for
> finding if ImageJ is filling with true black.  I don't know why
> ImageJ is
> being so finicky, but I am writing a procedure for other people to
> follow
> using this method, so I need to know as much as I can about this
> process.
> Does anyone have suggestions of what the problem is and how to fix
> it?  Thanks.

--
Joel B. Sheffield, Ph.D.
Biology Department, Temple University
1900 North 12th Street
Philadelphia, PA 19122
[hidden email]
[phone], fax [phone]
http://astro.temple.edu/~jbs <http://astro.temple.edu/%7Ejbs>

[Robby Connor]

I got it working again.  This could very well have been the root cause the
entire time: I didn't select the correct value for black in the Color
Picker.  I didn't even notice there was a value until Joel brought it up.
So now that I am filling with 0,0,0 instead of 4,4,4 I have hopefully
resolved this.  I will post again if I manage to screw ImageJ up another
time.  Thanks for everybody's responses.