Application of Poisson distribution to image processing I'm trying to write a program to detect water bubbles in heated oil. I've applied a canny edge detection filter to the image and the results look like the following:
No bubbles: http://imageshack.us/photo/my-images/20/nobubbles.png/ Bubbles: http://imageshack.us/photo/my-images/35/swirlsg.png/
I'm trying to find a way to determine when the bubbles are present.
Would it be appropriate to check if the white pixels are Poisson distributed?
 A: If this was something that I had to get done, and I once had to get done something similar, I would go a totally different direction.  I would take a "characteristic bubble" and use it as a "image-let" then use frequency domain methods to determine locations of candidate bubbles in the field.  It avoids all this "edge" complexity and can give you the bubble location accurate to substantially less than a pixel.
Here is a thesis on the subject.  (link)  It is not a bad start and gives a useful 1d "toy problem" with which you can refine your methods before going to higher dimensions.  It also shows application to bubble tracking in 2d but that is toward the end and rests heavily on the 1d staging.  Please check out the bibliography including Adrian and Reed.  You can also consider the tool "PIV Sleuth" here.  Particle Image Velocimetry (PIV) might be a good subject to explore here.
A: If anyone is interested, the solution that ended up working for me was to do a gaussian smoothing filter of the image before doing the canny edge detection. This eliminated much of the noise so that the canny edge detection filter only detected bubble edges. This is much simpiler than having to create eigen bubbles then use FFT. The simpiler technique worked very well for me.
