Local Texture Feature based Emotion Recognition usingSemi-Supervised Learning Technique

P. Shanthi, S.Nickolas

Recently, various feature descriptors attain better performance in Facial Expression Recognition (FER) system. However, deriving an effective facial feature representation remains a critical task. This paper presents a local appearance feature fusion for automatic emotion recognition using Local Neighborhood Difference XNOR Pattern (LNDXP) and Local Binary Pattern (LBP). The LBP(LBP) has been known to be a powerful texture feature for facial expression recognition. However, only few approaches utilize the relationship among neighborhood pixels itself. First, LNDXP is obtained based on two closest vertical and/or horizontal neighborhood pixel relationships. Next, Local Binary Pattern (LBP) is also calculated and finally, single feature descriptor obtained by the sequential fusion of LNDXP and LBP. The proposed work is also extended to efficiently handle a large amount of unlabeled data using two-stage semi-supervised classification algorithm. At the first stage, five different strong classifiers such as, J48, Random Forest (RF), Multi-Layer Perceptron (MLP), Radial Basis Function Network (RBFN) and Multiclass Support Vector Machine (M-SVM) are trained with small percentage labeled data and based on the trained model, rest of the unlabeled data is assigned with pseudo labels. At the second level, K-Nearest Neighbor (K-NN) is used as the weak classifier and to evaluate the classification accuracy, nearest-neighbor value (K) is changed to find the best condition that improves the average recognition rate.The experimental resultsindicate that the proposed scheme has robustness with 98.2% average accuracy on CK+ dataset.

Volume 11 | 08-Special Issue

Pages: 1686-1689