Debanding the World | Hacker News Reader

Debanding the World

blog.mapbox.com

162 pointsakalin6 comments

This is a cool example of the intersection of two fields: computational geometry/GIS, and image filtering algorithms.

In this case, information from GIS land (e.g. the orientation of the satellite’s aperture relative to true north on the ground below) was fed into the image filter’s autocalibration, because “natural” grid-like structures that should be preserved (cities, farm plots) tend to be aligned to true north, unlike the camera’s distortion, which is aligned to the aperture. (In the article, this is the part where they draw a diagonal mask stripe across the FFT image.)

Anyone working on a cross-subdisciplinary problem like this? Have any interesting “systems of algorithms” to share?

Maps involve a wide range of these. Navigation involves a lot of computational geometry mixed with graph theory. Telematics deals with signal processing and machine learning. Cartography mixes GIS with rendering algorithms. The need for high performance and accuracy means there is a lot of motivation to continue pushing deeper into integration of the fields.

working on FFT-transformed images is quite the dark magic if you don't understand what you're looking at; this is especially interesting when you look at what e.g. H.264 does to images in FFT: https://sidbala.com/h-264-is-magic/ (look at the 2% mask and the resulting image - it provokes a 'this shouldn't be possible' reaction).

Honestly, I'm more impressed with the 11% image compared to the 100% image than I am with the 2% image compared to any of the others. But, then again, I have no idea what I'm doing.

(2014)

I remember an image processing application back in the late 90s that would do FFT/reverse FFT that I used a couple of times to “paint out” certain regular artifacts. Same basic idea as this, but not nearly as successful. I can’t remember the app name though.

Reply on news.ycombinator.com