User:Martsniez/sandbox/blockhash

Documentation

[create Create a translatable help page (preferably in English) for this property to be included here]

Motivation

Blockhash is a standard of a perceptual hash of an image. By use of the hash, it programmatically confirms the identity of the linked image (e.g. when the image substantially changes at its source will have a different perceptual hash, when it is transformed slightly it will have the same hash) on wikidata or other wiki-projects. It can get the data back when a link/connection is lost.

Blockhash is a standard that describes 256 bit perceptual hashes. A perceptual hash, unlike a cryptographical hash, remains the same after the image has been resized/slightly altered. For example the blockhash of the Nights Watch remains: 0103217f21ff317b0000047b87f9f7ef71f11d710ce34c630c610860074effff, regardless of the size or other changes in format.

Elog.io created hashes for 21 million images on commons.wikimedia.org. If we add these to wikidata, the content of these images can be verified in a reliable way. In order to do so we would need a property blockhash that can store the URN to this hashing function. For example when the image is re-used on other platforms without proper attribution.

Discussion

•  Oppose The blockhash is a property of the image file, not the underlying painting. A different image of the painting -- even if only very subtly different -- has a significantly different blockhash. The appropriate place to store the blockhash in accessible form is with the image, not on the wikidata item for the painting. If somebody changes the image for the painting (better lit / different photographer / contrast adjustment / even a different crop) that is referenced from Wikidata, most probably your property won't reflect that, and will then be giving the wrong value. Jheald (talk) 18:16, 30 November 2015 (UTC)

Thanks for the feedback. I am not proposing a property, I am proposing a reference attached to the image URLs in the image reference. That way we can confirm that the reference image remains the same.

@Martsniez: Okay, that does make a bit more sense. In Wikidata terminology, you are proposing a property, but one with the constraint that it should only be used as a qualifier on statements that have values of type "Commons Media".

I am still not sure, though, about the balance of advantages and disadvantages in creating such a property.

Yes, the value could be used to check for recent changes/alterations/vandalism in linked Commons images. But would that have many advantages over a report that simply looked for image re-uploads of files linked from Wikidata ?

Also, it would serve to advertise the existence of the blockhash, and make it easy to pull a list of "Wiikidata" blockhashes that one could compare with images in an offline repository. But is that valuable? I'm not sure.

Against that, the property would need considerable maintaining; and there are likely to be a flow of routine changes from Commons -- eg image cropping, image colour adjustments, image substitution -- that someone would need be manually look through and okay, which might be far greater in frequency than any vandalism. (Though one could perhaps envisage a vandalism detection tool, with people scrolling down a presentation of paired thumbnails of the before and after versions).

But even then, does the blockhash really give much advantage over a tool that eg just presents the whole of the "recent re-uploads" stream on Commons, perhaps optionally filtered to look only at images that are used on a particular Wiki, or in a wikidata statement? You're going to need something like that stream to feed any comparisons anyway. (Also, there are some kinds of vandalism that may not change the blockhash very much -- particular kinds of defacement may only change a few blocks, and fall below the threshold of what blockhash considers to be a significant change).

So I'm not as set against the proposal as I was; but I do still need a bit of convincing that this would actually give us anything useful. Jheald (talk) 12:58, 1 December 2015 (UTC)

•  Oppose Agree with Jheald. This information belongs on Wikimedia Commons, probably only when structured metadata is in place there (I heard this will be worked on from early 2016) - and then only when it is possible to connect this hash to a specific image file (taking into account that images on Commons are also sometimes replaced). The principle is good and useful but it must be applied at the right location. Spinster (talk) 08:19, 1 December 2015 (UTC)