Read on Twitter
computer vision nerds: i’m trying to extract swell directions and amplitudes from satellite images. i know in theory i should be able to use a sliding FFT for this but i would love any more tips 
https://www.dropbox.com/s/vhb41rlikdsx1bf/high-resolution-duffs.jpg?dl=0
https://www.dropbox.com/s/vhb41rlikdsx1bf/high-resolution-duffs.jpg?dl=0
the basic idea is to look for peaks in the FT, but i have a feeling it’s not going to be anywhere near this clean 
https://www.cs.unm.edu/~brayer/vision/fourier.html
https://www.cs.unm.edu/~brayer/vision/fourier.html
when you feed these images through a DFT, you do get a few splotches of energy that seem to correspond to swells. i highlight two lower frequency swells (which create small waves in the same direction) as well as one splotch that seems like wind swell.
if you mask the relevant region with a frequency-bandpass and orientation-bandpass(?) filter then you can extract the two dominant swells. i'm pretty sure these are from the southwest and southeast, but there is a 180° ambiguity with this analysis.
you can extract three swells and assign them to three color channels for a reconstruction in a single image. or you can extract 16 evenly spaced orientations and assign them to a hue rainbow, giving an orientation-colored image (don't know if there is a name for this..)
it's a nice effect, and makes some patterns a little easier to read. but it doesn't get me much closer to getting an abstraction like this. DFT clusters give the linear swells, but i think i will need to compare the local DFT to the global DFT to find the curving swells.
