Rough Image Hash

[admin]

BTW, I'm currently working on something I internally call "Rough Image Hash". It should make identifying similar images near-automatic or at least very easy. And it will be scale-invariant, brightness-invariant, saturation-invariant, hue-tolerant, and, of course, blindingly fast. So you will be able to find "Image Dupes", and you'll get "Perceptual Sorting" in the file list.

[klownboy]

Sounds interesting. So how would it handle simple changes to aspect ratio or rotation? Typically after getting home from a trip, I'll spend some time manipulating photos a bit. Nothing dramatic - just some minor rotation if I had a horizon tilted (which also results in some cropping) or cropping photos from a camera 4:3 to 16:9. In some instances I may lighten or darken slightly but very rarely.

[Marco]

BTW, I'm currently working on something I internally call "Rough Image Hash". It should make identifying similar images near-automatic or at least very easy. And it will be scale-invariant, brightness-invariant, saturation-invariant, hue-tolerant, and, of course, blindingly fast. So you will be able to find "Image Dupes", and you'll get "Perceptual Sorting" in the file list.

Wow, AI wetness!!!
Jokes aside, that's an uber-welcome feature. Sometimes I really need to find image dupes and have to resort to a third party, now defunct, utility. Glad to see that coming!

[admin]

Sounds interesting. So how would it handle simple changes to aspect ratio or rotation? Typically after getting home from a trip, I'll spend some time manipulating photos a bit. Nothing dramatic - just some minor rotation if I had a horizon tilted (which also results in some cropping) or cropping photos from a camera 4:3 to 16:9. In some instances I may lighten or darken slightly but very rarely.

We'll see. It won't be learning, but I'll be learning...

[admin]

Ok, I'm learning... Looks like it's a *little* harder than I thought initially and I'm not sure yet if that *little* is ironic. But I'm still optimistic...

[jupe]

What formats does this actually support Don, or what criteria is used to decide?, I couldn't ascertain because those I have codecs for like WebP don't work for me, whereas HEIF does though, so is it based on codec support or a fixed list of extensions?. Also, it's just a suggestion, I think it would be good to open Tolerance up to quicksearch if it's not too difficult, maybe something like #image.hash: 4 == 18cde751667618ce

[admin]

"WebP don't work for me"... no hash is generated?

WebP does work for me. There are no limitations AFAIK. All images that can be shown should be hashable.

[jupe]

Hmm not what I experience.

2023-07-24_020707.png (19.51 KiB) Viewed 482 times

[admin]

All those file types have thumbnails and hover box?

[jupe]

Yes.
I think for some filetypes the issue is x64 related, even though for WebP I have x86 plugin installed too.

[admin]

Ah yes, that might be. 64-bit thumbnails are currently not supported by the image hash.

Nevertheless, I worked on it a bit and the next beta might give other results.

[jupe]

It seems to be generating more hashes but all the items get duplicated with ImageHash suffix filenames of non existent files, and it doesn't complete, so its hard to tell.

2023-07-24_032801.png (85.24 KiB) Viewed 460 times

[admin]

Fuck, forgot to remove a line of debug code. Wait...

[admin]

Okay, download v24.60.0118 again.

[jupe]

Yep WebP are working now, not sure the factors at play regarding the others that still don't though.