[clug] 'Monument' photo/video storage: how to build one at home?

[Hal Ashburner]

On 16 June 2016 at 07:32, steve jenkin <sjenkin at canb.auug.org.au> wrote:

>  - One of the indexes they create is ‘faces' (+ date/time, camera).
>     - Is there good Opensource to identify faces in photos?
>       Does anyone use any of this software?


DigiKam is the only Free software I know of that does face
recognition. I haven't used it.
I once messed around with OpenCV and Haar Cascades to do face identification
http://www.docs.opencv.org/master/d7/d8b/tutorial_py_face_detection.html#gsc.tab=0
I used python bindings and it was relatively straightforward to get
something basic working. Eg drawing a bounding box on the part of the
photo that should be a face.

For recognition the usual answer is to do Principal Component
Analysis. (I think the apple software iPhoto does this) and it
basically sucks. You could improve the guess rate dramatically by
combining it with a machine learning model using features like time
(photos with similar time have a similar set of people in them) and by
frequency (you have a hell of a lot of photos of your family, they're
more likely to be the match than someone who you photographed and
identified 3 times in the past 5 years).

It'd be a fun project but was too much work for me to push on with it.
I should give digiKam a go sometime though.