Robust Facial Expression Recognition Using Local Binary Patterns and Gabor Filters

Facial expressions and gestures provide intuitional cues for interpersonal communication. Imparting intelligence to computer for identifying facial expressions is a crucial task. Facial expressions and emotions are governed by identification of facial muscle movement by visual cortex and training a machine to identify these highly in-situ movements is our primary interest. This thesis presents robust facial expression analysis algorithms for static images as well as an efficient extension to sequence of images. We present an efficient preprocessing method which eliminates the effect of illumination on the detected face images thus making them efficient for feature extraction. Robust Local Binary Patterns and Gabor filters are implemented for feature extraction which are known to provide efficient face representation and analysis.LBP facial features are represented in form of weighted histograms which are best classified using Kullback Leibler divergence measure .Artificial Neural Network classifier is also tested for classification of fused Gabor and LBP features. Further expressions are rarely defined by static images as their complete essence lies in a sequence of images. So further exploration is concentrated on analyzing expressions from a sequence of images. To eliminate head pose variations in consecutive frames and register images to keep the spatial information intact which is necessary for LBP feature representation we adopted SIFT flow alignment procedure and further tested the resultant image classification with implemented algorithms. The classification accuracy resulted in 95.24% for static expression images and 86.31% for sequence of images which is indeed appreciable when compared to other standard methods