This challenges the common notion of using neutral face images
for face recognition. As smile is the most common expression, it can be used
for optimizing performance when tens of thousands or even million faces are to
be recognized.
To test this, Yakoob and Davis, from University of Maryland, conducted an experiment in two
parts. Two databases containing neutral and smiling face images of same person
in each database were used. For first part of experiment, 10 subjects from
first database were used. The discrimination power was calculated for smiling
and neutral faces.
It was found that the discrimination power was more for
smiling expression than for neutral expressions which showed that smiling faces
are better in face recognition.
In second part of the experiment, more face images were
added from second database. In first stage of second part, 0-15 faces were
added. Then in second stage, 0-40 faces were added again. The average
discriminating power was calculated.
In both stages, the average discriminating
power of smiling faces was more than the average discriminating power of
neutral faces which again showed that smiling faces are better for face
recognition.
To detect a face of person in video clips, expressive faces such
as smiles will be more cooperative for access applications as well as
surveillance purposes.
-----------------------------------------------------------------------------------------------------------------------
Reference – Yaser Yakoob and Larry Davis. Computer Vision
Laboratory, University of Maryland, College Park, MD 20472. Research paper name
- Smiling Faces Are Better for Face Recognition. Published In - Proceedings of
the Fifth IEEE International
Conference on Automatic Face and Gesture Recognition (2002).
-----------------------------------------------------------------------------------------------------------------------
(Image courtesy of photostock at FreeDigitalPhotos.net)