Has anyone seen studies of facial emblems, according to Ekman's description (Ekman & Friesen, 1969) of emblems? Preferably with using FACS.
There are some information about facial emblems in Ekman's work (1979), ex. disbelief: AU 1+2+10+15+17+43 and/or rocking the head from side to side, but I've never seen other studies about facial emblems?
Maybe You know?
Dear Vladyslav,
although we don't term them as such, in some of our work, we do study expressions that may fall under Ekman's definition of "facial emblems".
A database that contains such expressions (which we refer to as conversational expressions) can be found here:
http://www.plosone.org/article/info%3Adoi%2F10.1371%2Fjournal.pone.0032321
We also have done several studies that are currently under review in terms of how such expressions may (or may not) differ from processing of "standard", emotional expressions.
I hope this helps a bit. Contact me if you want to know more...
Best
Christian
Hello Christian,
thank you for the article you linked. I never came across that one, and I'm always interested in studies of facial expressions that go beyond the traditional 7-emotion classification.
If you don't mind me asking - has your database been FACS-coded following the study?
It would be interesting to see how the expressions you obtained with a (semi-)naturalistic procedure compare to Ekman's findings.
@Marcello: we would love to have this resource FACS-coded. Unfortunately, the number of videos that are currently in the database (20 actors * 55 expressions (* number of frames per sequence)) means that this is somewhat prohibitive (in terms of cost / human resources).
Any ideas on how to have this FACS-coded with reasonable effort would be very appreciated, of course :)
You could consider an automated coding software. They are becoming more common, precise and reliable.
Unfortunately, most are expensive - but I suggest you to follow Intraface (http://www.humansensing.cs.cmu.edu/intraface/), which is currently in development and free for research purposes. Another free one which was commonly used is CERT, but it became a commercial product and the new version (FACET) is no longer free of charge.
I am currently looking for an automated coding software as well :)
Thanks, Marcello :) - unfortunately, we tried CERT and FACET on our data already, with highly mixed results. Since many of our expressions actually feature head rotations, the software often loses track of the feature points and - worse - misclassifies the corresponding AUs. This problem even occurs for frontal expressions, which should be no problem to track. Additionally, many expressions in the database are actually quite subtle (think, slightly raised eyebrow, or cocked head), which are far below the robust tracking threshold for these algorithms...
In my experience, for recognition of the basic six expressions, the AU detection performance seems sufficient, but anything beyond that is very, very brittle with today's technology.
How strange. I'm not surprised for difficulties with head rotation - but I thought that FACET was reliable for frontal poses. I wonder if your database presents some peculiar feature that lowers the software accuracy?
I hope someone more knowledgeable than me notices this discussion. I'm very interested, both in understanding why automated coding failed on your data and in how the problem could be solved.
- Badges
- Science topic