Wednesday, March 06, 2013

Lie Detection

So, I was reviewing an episode from Lie to me and then went to Paul Eckman's site and looked up the resources on lie detection. I wanted to know if there were any softwares which would attempt to detect lies from facial microexpressions on the basis of extensive training using machine learning. The catch is first convincing yourself that you can get a 100% accuracy and then convincing the world with a proof that we can crack the liars. Seems quite the dream, but I would love to work towards it. From the perspective of a first time plan, I believe we can divide the job pretty easily.


Get datasets of all facial expressions.

train system on all expressions.

Run system on suspect while asking questions and see results of different micro-expressions made by the accused.

The challenge: Are the features of micro-expressions which are the basis for emotion detection reliably trackable?

Here are some relevant products I could find :
http://www.noldus.com/human-behavior-research/products/facereader
http://www.affectiva.com/affdex/#pane_overview
http://nviso.ch/technology.html


And here seems to be a relevant paper.

I will go through related materials and get some more content into this post in the weekend.

2 comments:

Unknown said...

goddammit!

bringing science fiction to life. This is supposed to be MY DREAM!

oh well. can't find fault with you, if you're gonna do this. GO GO GO!!!

Gururaj said...

:)
Yep, this will be one must-to-do if not for anything else at least to see accuracy.
But It is still Eliyah that finds out the liars in the Asimov books and not R.Daneel. Empowering R.Daneel is what I want to do