|Generating Data||Main project page|
Previous entry Next entry
dammit i screwed up the backslash up above. I guess it doesn't matter but i'll have to fix it when i figure out how to move pages in the openverse. OK that was easy. It's fixed. Phew or Whew every one can calm down now.
Adding Shot Noise is the next step in completing Cheezum's three steps. I've been doing some reading on it and found a textbook with a buncha pages on it called Probability, Random Variables, and Stochastic Process it is written by Papoulis in 1984. It doesn't seem like a bad read but math heavy and i am looking more for an example right now. But the math doesn't seem that bad but he doesn't explain all the variables. I am guessing they were explained earlier since i am only looking at the shot noise chapter.
Anyway it seems like it comes from a variability in photon counting. Does this matter on such high numbers like I have? I don't know. My gut says no but Cheezum puts it in. And i thought it would matter earlier. So I'm gonna look into Shot tonight and look at what Cheezum did tomorrow. Hopefully i can implement it tomorrow. And that'll complete my Cheezum run. With that i can either put all my photons in place (something i have a .vi for already) or I can just use a rectangle and skip the placing photon thing. I am not sure. If the microtubule was 100% covered I'd say the rectangle was a good enough idea, but without it, it seems placing the dots is more realistic.
Interestingly, Cheezum takes a Poisson random number for each pixel in his frame. I did only for each dot, but this makes sense. He takes N to be the mean and thus radical N to be the standard deviation as is the rule for Poisson Processes. I am just fuzzy on what exactly N is. He says it is the number of photons taken in by the detector. But then says that it is the level intensity for the pixel. I think that is what i was doing. So the game plan is to randomly put down dyes along a rectangle. These dyes are calculated from convoluting the PSF and the dot. Then each pixel in the frame is given a random adjustment based on the Poisson process. Now i have all that except the Poisson done. So that doesn't sounds so bad.
Note to self: when we get the DIC working there is a paper that says how to generate DIC images "Towards automatic cell identification in DIC microscopy" by Young 1998. Might want to check that out one day. But this Cheezum seems like the guy to go to to learn how to generate these images i've seen this paper cited a couple of times.