I'm using geequie to find 99% similar (internet) images (well, duplicates).
Is there some free software that could try and guess which image is 'original' and which is recoded ?
not in a 'photoshopped' sense, but with images that are almost identical.
resolution is the same, size slightly different.
smaller image should be 'fake' but its not always the case.

thanks.

I'm not sure that would be possible unless original images were to have a standard which certified authenticity.

but there are tools that detect fake .flac files (ones converted from mp3).
I know it requires some AI, but could it simply select sharper image?
( if no filtering was done, it should mostly work? )


edit:
I don't need anything fancy, just some better way than comparing size.
persumably, such tool would take into account filename, timestamp, exif.

Why do you need to know - what's the actual problem you're trying to solve by identifying the original?

I have some images downloaded from net, like ,um, landscapes... and there are a lot of duplicates. I'd like to keep only only the one that is of 'best quality'. Because internet has a lot of re-coded images. Like if you convert jpg->png->jpg or something.

You can use identify (part of ImageMagick) or mediainfo to look at the images' resolution.

For the visual similary: Geeqie uses a custom routine for this.

https://github.com/BestImageViewer/g.../src/similar.c

On the command line, you can use compare to get statistics on how visually similar two images are:

http://www.imagemagick.org/Usage/compare/#statistics

thanks, but I need it to tell me which image is objectively better.
I just remembered there are some tests that compares video codecs like HEVC vs AV1.
I need something like that for images.

ImageMagick's "compare" program can quantify how similar are two images. The command is:

I use this to find photos of the same scene.
Ed

1 members found this post helpful.

To be objective you need to define "better".

If I take a 6016x4016 image, resize to 1024x684 and apply 10% JPEG compression, it might come out at 400KB.
Or I can send the original through an optimizer and come out with a 1024x684 file which is 200KB and is equal or better image quality to the previous file.
Or I can save the original full size but with 99% JPEG compression and get an 800KB file which is inferior to both of them.

So you can't trust number of pixels, nor can you trust number of bytes.

Maybe there a way to compare JPEG compression level (I think some software adds it to metadata), but this is limited use without knowing if the image has been resized/sharpened before or after, and whether a sharpened image adds or detracts to the image quality depends on the subject.


In summary, I don't know of any objective measurements that will give you a valid judgement without a human's visual input.

I suspect you're stuck with identifying similar images then using your own eyes to pick the best one - but if you do find an automated method I'm interested in knowing about it.

1 members found this post helpful.

There are many tool available online to detect duplicate images. Jigsaw from google. But not sure the accurate percentage.