Abstract:
Facial expression is one of the most powerful masses of non-verbal communication
through which we can easily enter the world of one's instant emotions or intuitions.
As most of the time, it elicits naturally, so it brings out a lot of applications in the
eld of machine intelligence, behavioral science, clinical practices, biometric security,
gaming, human computer interactions, psychological research, data-driven animation
etc. Proper expression recognition can lead to an intelligent machine for taking commands
more e ectively, can show the insight psychological condition of a patient to a
psychiatrist or researcher, can show the next suggested path for any computer game.
But automatic facial expression recognition is a challenging task due to the di erent
factors such as variations in illumination, pose, facial expression, alignment, di erent
ages, occlusions etc. In this thesis paper, we propose a novel feature representation by
a new feature descriptor, named Patterns of Oriented Motion Flow (POMF) from the
optical
ow information, to recognize the proper facial expression from a facial video.
The POMF computes di erent directional motion information and encodes those directional
ow information with enhanced local texture micro pattern. As it captures the
spatial temporal changes of facial movements through optical
ow and enables to observe
both local and global structures, it shows its robustness for the facial expression.
Finally, the POMF histogram is used to train the expression model by Hidden Markov
Model (HMM). To train through the HMM, the objective sequences are produced by
the generation of codebook using K-means clustering technique. The performance of the
proposed method has been evaluated over the RGB camera based and Depth camera
based video. We also compare the proposed method with the other promising appearance
based methods. Experimental results demonstrate that the proposed POMF descriptor
is more robust in extracting facial information and provides higher classi cation rate
compared to other existing promising methods.
