But that's ok, because when I finally got it to work (fuck you, cfitsio, for storing data in a stupid manner), it became immediately clear that this was probably going to work. I think this means that this week I've solved two major problems that have been around for years. Problem one seems to be definitely solved, with my test cases immediately showing improvement. I'm not sure how well it'll work for historical data, as we don't have the right images to do this correction for those, but maybe I can construct an estimate.
Problem two is where I came up with what seemed to be the correct algorithm. There's noise, it's confined to certain regions of the FFT power spectrum, so that should be removable by doing some sort of smoothing over the peaks that contain the power for the noise. Easy. Turns out
someone already wrote a paper suggesting that method, which I didn't find until after I had my implementation. Oh well. Still, the initial test suggests that it's probably going to help solve problem two.
Given that the time budget for these two problems was something like a month, this has been a remarkably productive week.
 |
| Really? |
 |
| "Dash dash dash dash FUCK! WET BRICKS!" |
 |
| Well. Thanks, Lois. Now I have to go buy this movie, I guess. |
No comments:
Post a Comment