Sub-pixel registration of frames within a stack

Bayesian analysis of blinking and bleaching

Sub-pixel registration of frames within a stack

Postby thescoop » Fri Aug 15, 2014 10:21 am

Dear Susan et al.,

Firstly, thanks for creating not just a great technique but also an associated discussion forum. It's a marvelous idea!

My aim is to process a large amount of tiff stacks over the coming weeks, all images are of fixed-cell images taken on a TIRFM.
I've run one stacked ROI on ImageJ with results that appears to imply that I have rather too-much xy-stage drift! So, I immediately need to seek out a good (sub-pixel resolution?) registration algorithm. I have a large amount of images that don't have constantly visible fiducial markers through the stack and I have to reply as best I can on blinking fluorophores as the markers of choice. Either I've overlooked your registration/alignment technique or you simply haven't alluded to your method of choice, so could you recommend an algorithm?

Thank you so much!
-Simon Cooper
thescoop
 
Posts: 1
Joined: Fri Aug 15, 2014 9:53 am

Re: Sub-pixel registration of frames within a stack

Postby edrosten » Fri Sep 05, 2014 2:17 pm

Hi,

Obviously the best solution is to see if there's anything you can fix with your microscope system to avoid drift. Given the sort of timescale that 3B runs over, that's quite a lot of drift you're seeing. However, that's not always possible.

In the published paper, our data was sufficiently low drift that we didn't need to do any correction. However at some points we did have data with drift and we found that this code:

http://www.mathworks.co.uk/matlabcentra ... orrelation

seemed to do a very good job. If the data is pretty dense, you use the first frame as the target and use the code to register all other frames to the first frame. The registration is a little bit noisy, so as a final stage, we extracted the x, y offsets, fitted a smoothing polynomial and then used the smoothed x, y values to do the image shifting. Note that the matlab code above uses Fourier based image shifting, so it can do very accurate, small subpixel shifts without introducing substantial re-sampling errors.

Regards,

-Ed
edrosten
 
Posts: 42
Joined: Wed Jul 18, 2012 2:54 pm


Return to 3B

Who is online

Users browsing this forum: No registered users and 1 guest

cron