Research On Facial Expression Recognition

Affective computing is a new research area, which tries to enable machine (computer) have the ability of understanding and expressing affection, just like human beings. Affective computing plays an important role in intelligent human computer interface (HCI). Since human affection is expressed mainly by facial expression, researchers begin to pay more and more attention on facial expression analysis. Facial expression analysis has wide application in many areas such as emotion and paralinguistic communication, clinical psychology, psychiatry, neurology, pain assessment, lie detection, intelligent environments, and multi-modal human computer interface (HCI).The general approach to automatic facial expression analysis (AFEA) consists of three steps: face acquisition, facial data extraction and representation, and facial expression recognition. Now, there are many researches in the three aspects, but the problems on facial data extraction and facial expression recognition still haven't been resolved.This thesis presents an effective method for facial expression recognition. First, in the facial feature extraction aspect, Local Binary Pattern (LBP) operator is used to extract face appearance features. It efficiently describes facial expressions. Then, in the facial extraction classification, a two-stage classification method is proposed. At the first (coarse classification) stage, distances from each template to a testing sample are calculated and two nearest expression classes are selected as expression candidates (candidate pair). At the second (fine classification) stage, multi-template pairs are formed for each candidate pair. A simple K-nearest neighbor classifier is used with weighted Chi-square statistic to verify one of the candidate pair as final classification result. This algorithm is tested on the JAFFE database and the average recognition rate is 77.5%. Experiments show that our method performs better than others with the same database.Also this thesis proposes a fully automatic system for facial expression analysis, it recognizes the color image with the same method, tests on personal color image, and obtains the effective recognition rate.