Image Processing for Cell Detection in Atmosphere Storms

[fasterthanjoao]

OK, so this isn't anything I've ever done before and would like some input on feasibility, and whatever else you can throw at me.

I'm examining images (of atmosphere with a storm structure building up) but unfortunately the data that is available (i.e. the data I've got to use) has quite severe limitations on the resolution.

Now, what I'm interested in is smaller cells within the bigger structure, I would like to single out those smaller sections and record their movement so I'm using various image processing algorithms to see if I can find anything appropriate to help me do this?

I'm also wondering about the resolution - the reason the images aren't of good quality is because before the data is passed to me it is JPEG compressed and comes out pixelated (which might have an impact on my observations) so I'm also wondering about any possible smoothing processes that could be of interest.

Thanks for any input, I didn't mention languages or anything since I'm still dabbling with different ones to see which suits, but I still think the underlying theme is valid throughout

 

[chroot]

You should never do any kind of scientific analysis on images compressed with lossy algorithms, like JPEG. They're essentially useless in quantitative analysis.

Futhermore, no image processing algorithms are going to actually increase resolution. There are many algorithms which can infer data from surrounding pixels, and can thus increase the number of pixels in the image -- but they cannot actually increase the amount of information in the image.

Smoothing filters do not increase the information content in an image -- like all filters, they only decrease it.

You might find some utility with edge-detection algorithms or other forms of sharpening filters, but remember that they cannot "discover" any information that was not already present in the original image. You might also be interested in Fourier transforms, as the resulting power spectra may be more useful to you than the images themselves. Just remember that the JPEG artifacts will introduce a significant amount of non-uniform noise into such spectra.

Have you looked at any university-level textbooks on image processing?

- Warren

 

[robphy]

Yeah, [as chroot mentions] with lossy formats like JPEG you shouldn't expect to extract good quantitative information, especially at the pixel-resolution of the image. Maybe you could extract approximate information on large features (like tracking the eye of a hurricane). You might want to think about how to assign an uncertainty in your measurements.

If you fully understand how JPEG compresses the image (say, from http://en.wikipedia.org/wiki/JPEG ), you could do a slightly better job of trying to get back some of the original image. ...but I suspect that it might not be worth it.

Of course, you could try to get better data... from the source [or secondary source].

In terms of possibly useful software... you might look into ImageJ
http://rsb.info.nih.gov/ij/
http://en.wikipedia.org/wiki/ImageJ

 

[Miss_Astro]

Hello, are any of you guys still around? I'm a little stuck on a similar problem. There seems to be a good source here:

http://www.cs.cf.ac.uk/Dave/AI2/node194.html


But my problem is understanding the summation. What I m trying to do is calculate ellipticites of images using their second order moments of ineria and their orientation angle.


My images are in matrix form with values depicting the greyscale of the image. Do you know what I would actually do?

 

[TylerH]

Create a new thread, this one is years old.