GCV-Based Regularized Extreme Learning Machine for Facial Expression Recognition

Extreme learning machine (ELM) with a single-layer feed-forward network (SLFN) has acquired overwhelming attention. The structure of ELM has to be optimized through the incorporation of regularization to gain convenient results, and the Tikhonov regularization is frequently used. Regularization benefits in improving the generalized performance than traditional ELM. The estimation of regularization parameter mainly follows heuristic approaches or some empirical analysis through prior experience. When such a choice is not possible, the generalized cross-validation (GCV) method is one of the most popular choices for obtaining optimal regularization parameter. In this work, a new method of facial expression recognition is introduced where histogram of oriented gradients (HOG) feature extraction and GCV-based regularized ELM are applied. Experimental results on facial expression database JAFFE demonstrate promising performance which outperforms the other two classifiers, namely support vector machine (SVM) and k-nearest neighbor (KNN).