switch of interpolation

classic Classic list List threaded Threaded
11 messages Options
fabrice senger fabrice senger
Reply | Threaded
Open this post in threaded view
|

switch of interpolation

Hi i'm using PIVlab to asses intra-cellular dynamics.
I would like to switch of interpolation as those vectors mainly appear outside the cells (good point, indeed) .
When I perform post-processing I can switch this option off but when I go further to the option "Derive parameters" especially if I go for smoothing data the program seems to apply interpolation by default...Did I miss something ?

Thank you very much,

Fabrice
William William
Reply | Threaded
Open this post in threaded view
|

Re: switch of interpolation

fabrice senger wrote
When I perform post-processing I can switch this option off but when I go further to the option "Derive parameters" especially if I go for smoothing data the program seems to apply interpolation by default...Did I miss something ?
Dear Fabrice,
if you enable smoothing, then the missing data will be interpolated, yes. So if you want to disable any interpolation, you need to disable smoothing too.

Please consider donating for the PIVlab project:
http://paypal.me/pools/c/81Y8wBVi4y

The developer needs to update to a recent Matlab release in order to keep the project running.

Vitto Vitto
Reply | Threaded
Open this post in threaded view
|

Re: switch of interpolation

Hi,

this is probably more of a MATLAB based question than anything else. How do you actually disable smoothing/interpolating ? I tried commenting out all the lines in pivFFTmulti.m file from 154 t 514, but the result of u,v does not seem to differ. Or if I only chose single pass, no smoothing/interpolation happens ?

Regards,
V
William William
Reply | Threaded
Open this post in threaded view
|

Re: switch of interpolation

Hi,
there is no smoothing of the results turned on by default. If you use DFT
multipass, then BETWEEN the passes results will be validated and smoothed. But
that has no effect on the result of the last pass. It only makes the prediction
between the passes more reliable. The result of the last pass will not be
smoothed or validated or interpolated if you don't enable that manually in the
post-processing.

Please consider donating for the PIVlab project:
http://paypal.me/pools/c/81Y8wBVi4y

The developer needs to update to a recent Matlab release in order to keep the project running.

Vitto Vitto
Reply | Threaded
Open this post in threaded view
|

Re: switch of interpolation

Thanks for the reply!

That's intriguing... I seem to get more accurate results with single pass FFT method compared to DCC. That's really puzzling!
William William
Reply | Threaded
Open this post in threaded view
|

Re: switch of interpolation

Indeed... Could you post example image files?

-- Dr. William Thielicke
http://William.Thielicke.org

Am 19. November 2014 11:37:00 MEZ, schrieb "Vitto [via PIVlab - Time-Resolved Digital Particle Image Velocimetry Tool for MATLAB]" <[hidden email]>:

>
>
>Thanks for the reply!
>
>That's intriguing... I seem to get more accurate results with single
>pass
>FFT method compared to DCC. That's really puzzling!
>
>
>
>_______________________________________________
>If you reply to this email, your message will be added to the
>discussion below:
>http://pivlab-time-resolved-digital-particle-image-velocimetry-tool-f.33690.n7.nabble.com/switch-of-interpolation-tp24p37.html
>To start a new topic under PIVlab - Time-Resolved Digital Particle
>Image Velocimetry Tool for MATLAB, email
>[hidden email]
>To unsubscribe from PIVlab - Time-Resolved Digital Particle Image
>Velocimetry Tool for MATLAB, visit
>

Please consider donating for the PIVlab project:
http://paypal.me/pools/c/81Y8wBVi4y

The developer needs to update to a recent Matlab release in order to keep the project running.

Vitto Vitto
Reply | Threaded
Open this post in threaded view
|

Re: switch of interpolation

http://s16.postimg.org/uar8qdkg5/image.png

http://s28.postimg.org/bleddlybh/image.png

A simple case of 1 pixel shift to the right. Now it either has something to do with it being 16 bit (unlikely, I guess) or that DCC uses Different sized search area (which intuitively should yield better results) and FFT doesn't as far as I am aware (maybe FFT's zero padding actually does help in this case?). Plot the u,v scatter plot from statistics using either subpixel fitting (difference is small, but I would expect DCC to always yield more accurate or at least as accurate results...). Maybe the correlation matrices of DCC simply don't do very well for such fitting (I would suggest trying centroid (yes, centroid, even if it's normally least accurate))...
William William
Reply | Threaded
Open this post in threaded view
|

Re: switch of interpolation

Hi,
These images are not suitable for a PIV analysis. In this case, the problem is not the correlation algorithm, but the subpixel peak estimator. The "particle size" is 1 pixel, therefore the peak in the correlation matrix is not suitable to be approximated by a Gaussian curve. The particles need to have an image diameter of 3-5 pixels for a good analysis. They should also have a Gaussian intensity distribution.
Hope this info helps, you can also find more detail in my thesis.

-- Dr. William Thielicke
http://William.Thielicke.org

Please consider donating for the PIVlab project:
http://paypal.me/pools/c/81Y8wBVi4y

The developer needs to update to a recent Matlab release in order to keep the project running.

Vitto Vitto
Reply | Threaded
Open this post in threaded view
|

Re: switch of interpolation

Ah yes, I forgot about that! I believe Prasaad did mention this (or at least Adrian and Westerweel did in their book!).

Thanks!
Jenni Jenni
Reply | Threaded
Open this post in threaded view
|

Re: switch of interpolation

This post was updated on .
In reply to this post by William
Hi William,

I'm having a similar problem to Fabrice, plus another additional problem, wondering if I'm going wrong somewhere?

I'm also trying to avoid interpolation in my results.
I am using Windows 7 and MATlab version 2013b, everything else seems to work fine, except for the two problems outlined below.

I was previously using FFT analysation with two passes and ensuring all smoothing and interpolation options were turned off during post processing.  I was very happy with the results.  When, however, I decided to save the session and close the program, the interpolated vectors would appear next time I opened it up.  They would also appear if I tried to calculate any derivatives.

After reading your previous explanation I realised that second pass might be causing this, so I tried the other 2 analysis methods, which yielded the following results:

FFT - single pass: same as above, interpolated vectors appear when loading a saved PIV and when calculating derivatives.

DCC - no interpolated vectors (hooray!), but a different, more strange problem:
I use plot --> Derive parameters/modify data --> calculate mean vectors
To create a new frame of mean vectors
In the same menu I then use Display parameter --> vorticity --> apply to current frame
To calculate vorticity for the mean vector frame (the same happens with other derivatives too)

This worked fine when I was using the FFT analysis methods, but when using DCC the PIVlab window turns white, and the following error appears int he MATlab window:

Error using image
Image CData can not be complex

Error in PIVlab_GUI>sliderdisp (line 321)
                image(rescale_maps(currentimage), 'parent',gca, 'cdatamapping', 'scaled');

Error in PIVlab_GUI>apply_deriv_Callback (line 2926)
sliderdisp

Error in gui_mainfcn (line 96)
        feval(varargin{:});

Error in PIVlab_GUI (line 56)
    gui_mainfcn(gui_State, varargin{:});

Error in @(hObject,eventdata)PIVlab_GUI('apply_deriv_Callback',hObject,eventdata,guidata(hObject))

 
Error while evaluating uicontrol Callback

Additionally when trying to create histograms of this frame only positive values of u are shown (despite there being both positive and negative u values in all of the datasets I've been looking at).  Trying to create a v histogram results in a grey histogram window instead of a plot...very strange.

Apologies for the length of the explanation!
Many thanks for your previous help,

Jenni
William William
Reply | Threaded
Open this post in threaded view
|

Re: switch of interpolation

Hi Jenni,
I will try to answer your comments asap, currently I don't have enough time... Could you maybe attach some example data? That makes it easier for me to find the problems.

Thanks and all the best,
William

Please consider donating for the PIVlab project:
http://paypal.me/pools/c/81Y8wBVi4y

The developer needs to update to a recent Matlab release in order to keep the project running.