Re: Super-resolution image processing

[Steve Wolter]

Hi Sebastien,

yes, I confirm this, and my answer aimed for this exact question. (Even though the formulation sounds wierd - a pixel is by definition a point and doesn't have a size, so it can't have a top left corner.) Your discrepancy is probably not exactly 0.5 pixels, but depends on the subpixelation: for 10 subpixels, the offset should be 0.45.

Best regards, Steve

2014-06-26 16:17 GMT+02:00 Sebastien Tosi <sebasti...@irbbarcelona.org>:

Hi Steve,

My question was in fact more accurate: It looks like RapidSTORM (0,0) is not the top left corner of the top left pixel (convention commonly used by different other super-resolution packages) but the center of the top left pixel. I consistently found a discrepancy of 0.5 pixel in X and Y between RapidSTORM outputs and the outputs of these other packages. Do you confirm this?

Best,

Sébastien

On Wed, Jun 25, 2014 at 10:22 PM, Steve Wolter <steve.w...@gmail.com> wrote:

Hi Sebastien,

yes, the top left pixel has the coordinate (0,0) in rapidSTORM, and the super-resolved image's top-left pixel also has the coordinate (0,0).

Best regards, Steve

2014-06-25 17:15 GMT+02:00 Sebastien Tosi <sebasti...@irbbarcelona.org>:

Hi Steve,

Just a quick question about RapidSTORM. I have been comparing the localization results to the ones obtained with some other software and it seems there is a consistent shift of half a pixel in both X and Y directions. Are you by any chance using the center of the top left pixel as reference?

Regards,

S.

On Sat, Feb 11, 2012 at 4:12 PM, Steve Wolter <steve....@uni-wuerzburg.de> wrote:

Hello Sebastien,

Sebastien Tosi schrieb:
> I have to say that the
> software is extremely unstable on my machine (Windows XP 64, JAVA 1.6.21)

If you can specify the nature of the unstability, I can do something
about it. Our bug tracker is at:

https://idefix.biozentrum.uni-wuerzburg.de/bugzilla3/

Please feel free to report any bugs you find. Note that I don't think
there is any progress indicator for fluorophore PSF generation and that
PSF generation takes quite some time (especially for 3D), so the program
might appear to hang with one CPU core fully loaded for some time at
job start.

> but I
> eventually managed to generate a dataset for the randomly located fluorophores
> case (with default parameteres). With organized fluorophores (lines or squares)
> I could not get any meaningful dataset (crash or seemingly only noisy
> output)... I guess the default parameters are erroneous or my machine
> configuration is conflicting with something (what is your setup?).

I've used the following parameters apart from the defaults:

Input type                                              Randomly generated values
Pattern to distribute simulated fluorophores with       Fluorophores on lines
Line angle in XY plane in degree                        60 (got corrected to 61.2)
Distance between fluorophores on line                   100

Worked right away, but the job took a little time to start. Does it for you, too?

The square lattice distribution seems to fail, especially given the
short starting time. Sorry, too many features, too few tests; I'll
investigate at some time, the progress of this is tracked in bug #186:
https://idefix.biozentrum.uni-wuerzburg.de/bugzilla3/show_bug.cgi?id=186

> I understand that the z position of an emitting fluorophore is not affecting
> much the shape of the PSF in the image plane in TIRF illumination.

I think you already know this, but formulating it the other way round:
The Z position does not vary much in TIRF because the evanescent field
is so shallow.

> I would now
> like to characterize the variations of intensity at the maxima (pdf of this
> distribution). To my understanding they are connected to the integration time,
> the emission frequency and the on/off statistics. Do you think there is a
> "workable" regime where these variations can be made "reasonably small"? I
> guess it can always happen that a photon starts to emit at the really end of
> the exposure leading to a faint response but there might be some tricks to play
> to reduce the variations with some frame overlay or something similar. I am
> interested in reducing these variations to test some localization strategies.
> What's your opinion?

Live with the variations. Real data show a broad distribution of
intensities, even broader than our simulated data. I think this is due
to triplet fluctuations, which we cannot resolve and see as mean
intensity differences. Any localization strategy worth its salt has
to live with the variations. This is especially true for TIRF, since
the evanescent field yields a variety of excitation intensities.

Best regards, Steve

--
 Steve Wolter ( Würzburg Univ.) | Web page: http://swolter.sdf1.org
                                | vCard:    http://swolter.sdf1.org/swolter.vcf
 A witty saying proves nothing. | Schedule: http://swolter.sdf1.org/sched.cgi
    -- Voltaire (1694-1778)     | E-mail:   swo...@sdf.lonestar.org

--
Sébastien Tosi
Advanced Digital Microscopy
Institute for Research in Biomedicine - IRB Barcelona
Baldiri Reixac, 10
Tel:  +349340-20499
Fax: +34934031116
E-08028 Barcelona - Spain
sebasti...@irbbarcelona.org

--
Sébastien Tosi
Advanced Digital Microscopy
Institute for Research in Biomedicine - IRB Barcelona
Baldiri Reixac, 10
Tel:  +349340-20499
Fax: +34934031116
E-08028 Barcelona - Spain
sebasti...@irbbarcelona.org