Bug 392025

Summary:	Wishlist: ability to separate images by similarity (instead of album or format)
Product:	[Applications] digikam	Reporter:	Jens <jens-bugs.kde.org>
Component:	Albums-Sort	Assignee:	Digikam Developers <digikam-bugs-null>
Status:	REPORTED ---
Severity:	wishlist	CC:	caulier.gilles, iwannaberich, mario.frank
Priority:	NOR
Version First Reported In:	5.9.0
Target Milestone:	---
Platform:	Other
OS:	Linux
Latest Commit:		Version Fixed/Implemented In:
Sentry Crash Report:

Description Jens 2018-03-18 20:54:27 UTC

I have a lot of images in albums which are quite similar but chronologically far apart.
I would like to have a view in the main album view that sorts (or groups, or separates) images in one album by their similarity - with an adjustable "minimum similarity" factor which would influence how many groups to create.

Since you can already search images by similarity and find duplicates, there must be some similarity indicator for each image (to each other image). 

This is somewhat related to #384396 where we could use a very similar algorithm to group similar (but not yet assigned) faces in the "Persons" view.

Comment 1 Jens 2018-08-24 21:12:20 UTC

I think a "group by similarity" feature would make a lot of sense for the faces view, but it could also be a useful option for an album or tag view.

The basic feature set is already there, since Digikam can search by similarity, and thus calculate a "similarity distance" between two images. It would only need to do this for every combination of two images in a pair, which should be feasible for sets of <1000 images with current CPUs (and the results can be saved too).

Comment 2 Mario Frank 2018-08-27 11:22:56 UTC

Hi Jens,

currently, only few similarity values are really stored in database. This was not possible before 6.0.0 beta, as the database structure did not allow this without really nasty constructions. Especially the similarities used by the duplicates search are not stored easily extractible. To implement your wish, some (not minor) work has to be done on the HaarIface. Sadly, I did not have much time lately. I'll try to get some time free to prepare a better use of the new similarity database. But this will probably not be part of the final 6.0.0 release.

Regards,
Mario

Comment 3 Jens 2018-09-01 14:55:05 UTC

I would be very happy and this feature would be worth another €50 donation to the Digikam project from my side if it is implemented.

If I can help, e.g. with implementation questions, use cases, example images or SQL knowledge (I know my way around MySQL and Postgres, sadly not around C++ though) just let me know.