Documentation about hashing: hash size, resize algorithm and hash type #1662

Observations and follow-up questions, or generally "why do these settings matter to us?":

• Hash size: if full image is used for second pass for potential matches, presumably setting this too low would produce more false negatives.

  • Tested: on my 12 year old MBP, set of 275 images
    • @ 8: Found ⁨138⁩ similar image files in ⁨49⁩ groups (all false positives) / time ~11s
    • @ 64: Found 0 similar image files / time ~12s
  • False positives would indicate there is no 2nd pass
  • Real world practicality? using a single pass, any time saved on computation would seem to be more than outweighed by false positive matches to sort through. Even on my ancient machine, the difference in processing time seems negligible. Only downside of using largest hash size then would be cache size? Based on these results, I'm switching to 32 from the default 16

• Resize algorithm: images scaled way down will appear different using different algorithms = different matches

  • Beyond my level of knowledge but the gist I usually get is that Lanczos is the newest/best. In my tests CatmullRom gave good (visually similar) results; some others were pretty poor
  • Real world practicality? Idk if this has much practical effect that you don't achieve by playing with "max difference"

• Hash type: This is why I'm here - I don't know how the various hashes are calculated or why you would choose one over another, but I would like to know

• Geometric invariance: Thank goodness for this addition! Unless your images are very homogeneous idk why you wouldn't use this always. Absolutely should be on by default.

Takeaways: What hash size produces great results, and are there big enough downsides to warrant including smaller options? Is there a practical situation where a user would want to change the resize algorithm from the default? What effect does hash type have?

In most situations I want more settings/control over the software I use - I tell the computer what to do, not the other way around! Yet when there is really a best setting or one that's meaningful to few users, it may be better UX to remove the choice. Would it be appropriate to remove some of these options (or at least hide them in an "advanced settings" for the few people who need them)?

I've done my best without documentation, please correct me on any misunderstandings.