https://bugs.kde.org/show_bug.cgi?id=523740
Bug ID: 523740
Summary: age subdivision of people recognition
Classification: Applications
Product: digikam
Version First 9.1.0
Reported In:
Platform: Debian testing
OS: Linux
Status: REPORTED
Severity: normal
Priority: NOR
Component: Faces-Recognition
Assignee: [email protected]
Reporter: [email protected]
Target Milestone: ---
I'm impressed with the new face recognition engine and now I have to dig
through 10k+ faces, and it gives time to think about stuff when waiting for the
scanning/interface updates, etc.
People have phases in life where their faces are rather different (baby, young
child, older child, teenager, young adult, adult, older adult, pensioners)
Would it help the recognition engine to first estimate the age of the face
(this is often quite typical) and then identify a person in the range of a
certain age? The uniqueness of a person's face in relation to other faces of
the same age group is much easier to determine when other ages of the person's
face don't "interfere" within that age group. Of course it would help to keep
track of a person's approximate date of birth.
If this is interesting, it would make sense to create a "sub-person" for
persons, with specific sets of faces to define the face. It may not make sense
to create a fixed set of "ages", because you may not "known" a person through
all their phases in life and people are all different, so their characteristics
change at different ages. But I imagine a kind of organic "splitting" of a
recognition set when the faces diverge too much and split into two more close
but different sets for the same person.
Example: you get a baby (a bazillion photos happen), the baby grows up and
changes into a child, perhaps develops very distinctive features like a mole or
a scar or changes hair color or style, the faces are now so different that they
barely look the same as the baby they were earlier, but locally, they are very
similar. Still it's the same person. If you split the total set of faces into
two or more distinct sets which look similar within the subset, they are easier
to keep specific than when you persist in one set from just born to much older.
--
You are receiving this mail because:
You are watching all bug changes.
