Abstract:
Human facial expression recognition is always a challenging8problem in the computer vision systems. This is a very important channel for communication of human beings. As most of time people uses facial expressions to express their feelings. Now a days automatic facial expressions needed to be applied in many real applications, moreover recent8trends toward8cloud computing and outsourcing8had led to the requirement8for facial8expression8to be8performed remotely. There has been a lot of work done for achieving this goal, people uses different setups, techniques and strategies for accomplishing their goals. In this work we used a set of features that have been identified to be most potentially useful for recognizing facial expressions. So we propose a technique in which we use a combination of 3-different types of features i.e. Scale invariant features transform (SIFT), GABOR wavelets and discrete cosine transform (DCT). Some pre-processing steps have been applied before extracting these features. We used four classifiers as Logistic Regression, K-Nearest Neighbor, Neural networks and Support Vector Machine. By comparing the recognition rates of each classifier we found that Support Vector Machine (SVM) with radial8basis8kernel function will generate better results among all the other classifiers. We evaluate our results on the JAFFE database under the same experimental setup followed in literature. Our proposed methodology of combining three different types of feature sets is effective, as it is found that our recognition results are better than any other previously published results on the JAFFE database.
