wildtreetech / sentinel2-bot

🛰📷🌏 Tweeting pictures taken by the Sentinel 2 satellite.
https://twitter.com/sentinel2bot
MIT License
36 stars 11 forks source link

Visually interesting images #5

Open betatim opened 7 years ago

betatim commented 7 years ago

Good yet simple algorithm for picking visually interesting pictures.

(This issue is about finding a super simple algorithm to make this decision, if you want to use something complicated like machine-learning open a new issue)

rth commented 7 years ago

Well, there is a paper by Elazary & Itty (2008) where they suggest that visually interesting objects correlated with visual saliency (no idea what that is). There is a few python libraries here and here, so that could be tried.

Otherwise, if the GPS position of the image is known, it might be possible to query OpenStreet Maps to see if there is any interesting landmarks there, or at least avoid processing deep sea images though I suppose this can become quickly complicated..

betatim commented 7 years ago

Another avenue might be to calculate the entropy of the image (e.g. http://scikit-image.org/docs/dev/auto_examples/plot_entropy.html).

Some images of the sea are actually not as boring as you'd think: https://twitter.com/Sentinel2Bot/status/785777365079195648 though probably the Sentinel guys don't like the banding effect we can see.

(n.b. The real "ocean miles away from anywhere" stuff is excluded already by virtue of the satellite not taking pictures there)