ImageJ

Sub-pixel accuracy for SIFT/MOPS Feature extraction possible?

_‹ Previous Topic Next Topic _›

Classic

List

Threaded

5 messages Options

Christoph Siegenthaler

Sub-pixel accuracy for SIFT/MOPS Feature extraction possible?

Hi imagej list

I want to measure the displacement between two large areas of a set of
two pictures. I'm using FIJIs Feature Extraction with MOPS, called
from matlab, which works great. The only problem is that, using
'Measure', I can only get integer values for the landmark's X and Y
coordinates, even if i change the accuracy in 'Set Measurements'.
- Does FIJIs standard SIFT/MOPS Feature Extraction allow for sub pixel
accuracy?
- If it does, how can I measure the landmarks with a higher accuracy?
- If it doesn't: Is there a replacement implementation that does
support it? Or any alternatives that allow the detection of small
displacements?

I'm fairly new to ImageJ/FIJI and any help is appreciated!

Best regards,
Christoph Siegenthaler

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html

Rainer M. Engel

Re: Sub-pixel accuracy for SIFT/MOPS Feature extraction possible?

Hi Christoph,

I asked for the same thing here and got a prompt answer from Stephan
Saalfeld.

15.11.2012
> the implementation stems from when ImageJ had no floating point
> precision in PointRoi. I'll adjust the plugin to use floating points
> some time soon...

There is so much development going on around ImageJ and its contributors
that some refinement work goes round and round too.
It is noticed and it'll come I'd say.

Best regards,
Rainer

Am 10.12.2012 08:46, schrieb Christoph Siegenthaler:

> Hi imagej list
>
> I want to measure the displacement between two large areas of a set of
> two pictures. I'm using FIJIs Feature Extraction with MOPS, called
> from matlab, which works great. The only problem is that, using
> 'Measure', I can only get integer values for the landmark's X and Y
> coordinates, even if i change the accuracy in 'Set Measurements'.
> - Does FIJIs standard SIFT/MOPS Feature Extraction allow for sub pixel
> accuracy?
> - If it does, how can I measure the landmarks with a higher accuracy?
> - If it doesn't: Is there a replacement implementation that does
> support it? Or any alternatives that allow the detection of small
> displacements?
>
> I'm fairly new to ImageJ/FIJI and any help is appreciated!
>
> Best regards,
> Christoph Siegenthaler

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html

Stephan Saalfeld

Re: Sub-pixel accuracy for SIFT/MOPS Feature extraction possible?

In reply to this post by Christoph Siegenthaler

Hi Christopher,

since you're now the second person asking, I felt motivated enough to do
it. Update Fiji and the locations are sub-pixel.

The methods implement sub-pixel accuracy since they exist, as do all
feature based alignment tools in Fiji. Just ImageJ, before 1.46a last
summer didn't enable non-integer Point Rois. Since it does now, I only
had to adjust the coordinates export.

Best,
Stephan

On Mon, 2012-12-10 at 08:46 +0100, Christoph Siegenthaler wrote:

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html

Christoph Siegenthaler

Re: Sub-pixel accuracy for SIFT/MOPS Feature extraction possible?

Hi

Awesome, thanks a lot Stephan! I just realised that I'm using a fairly old version of Fiji. I'll update mine and the one in the Arch Linux User repository soon. Thank you again for the fast reply and your work!

Best regards,
Christoph Siegenthaler

-------- Original-Nachricht --------
> Datum: Mon, 10 Dec 2012 12:45:45 +0100
> Von: Stephan Saalfeld <[hidden email]>
> An: [hidden email]
> Betreff: Re: Sub-pixel accuracy for SIFT/MOPS Feature extraction possible?

> Hi Christopher,
>
> since you're now the second person asking, I felt motivated enough to do
> it. Update Fiji and the locations are sub-pixel.
>
> The methods implement sub-pixel accuracy since they exist, as do all
> feature based alignment tools in Fiji. Just ImageJ, before 1.46a last
> summer didn't enable non-integer Point Rois. Since it does now, I only
> had to adjust the coordinates export.
>
> Best,
> Stephan
>
>
>
>
> On Mon, 2012-12-10 at 08:46 +0100, Christoph Siegenthaler wrote:
> > Hi imagej list
> >
> > I want to measure the displacement between two large areas of a set of
> > two pictures. I'm using FIJIs Feature Extraction with MOPS, called
> > from matlab, which works great. The only problem is that, using
> > 'Measure', I can only get integer values for the landmark's X and Y
> > coordinates, even if i change the accuracy in 'Set Measurements'.
> > - Does FIJIs standard SIFT/MOPS Feature Extraction allow for sub pixel
> > accuracy?
> > - If it does, how can I measure the landmarks with a higher accuracy?
> > - If it doesn't: Is there a replacement implementation that does
> > support it? Or any alternatives that allow the detection of small
> > displacements?
> >
> > I'm fairly new to ImageJ/FIJI and any help is appreciated!
> >
> > Best regards,
> > Christoph Siegenthaler
> >
> >
> >
> > --
> > ImageJ mailing list: http://imagej.nih.gov/ij/list.html
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html

Rainer M. Engel

Re: Sub-pixel accuracy for SIFT/MOPS Feature extraction possible?

In reply to this post by Stephan Saalfeld

Hi Stephan,

thank you very much for this update, great news.

Best regards,
Rainer

Am 10.12.2012 12:45, schrieb Stephan Saalfeld:

> Hi Christopher,
>
> since you're now the second person asking, I felt motivated enough to do
> it. Update Fiji and the locations are sub-pixel.
>
> The methods implement sub-pixel accuracy since they exist, as do all
> feature based alignment tools in Fiji. Just ImageJ, before 1.46a last
> summer didn't enable non-integer Point Rois. Since it does now, I only
> had to adjust the coordinates export.
>
> Best,
> Stephan
>
>
>
>
> On Mon, 2012-12-10 at 08:46 +0100, Christoph Siegenthaler wrote:
>> Hi imagej list
>>
>> I want to measure the displacement between two large areas of a set of
>> two pictures. I'm using FIJIs Feature Extraction with MOPS, called
>> from matlab, which works great. The only problem is that, using
>> 'Measure', I can only get integer values for the landmark's X and Y
>> coordinates, even if i change the accuracy in 'Set Measurements'.
>> - Does FIJIs standard SIFT/MOPS Feature Extraction allow for sub pixel
>> accuracy?
>> - If it does, how can I measure the landmarks with a higher accuracy?
>> - If it doesn't: Is there a replacement implementation that does
>> support it? Or any alternatives that allow the detection of small
>> displacements?
>>
>> I'm fairly new to ImageJ/FIJI and any help is appreciated!
>>
>> Best regards,
>> Christoph Siegenthaler
>>
>>
>>
>> --
>> ImageJ mailing list: http://imagej.nih.gov/ij/list.html
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html
>

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html