An Efficient Multi-modal Facial Gesture-based Ensemble Classification and Reaction to Sound Framework for Large Video Sequences

Abstract

Machine learning-based feature extraction and classification models play a vital role in evaluating and detecting patterns in multivariate facial expressions. Most conventional feature extraction and multi-modal pattern detection models are independent of filters for multi-class classification problems. In traditional multi-modal facial feature extraction models, it is difficult to detect the dependent correlated feature sets and use ensemble classification processes. This study used advanced feature filtering, feature extraction measures, and ensemble multi-class expression prediction to optimize the efficiency of feature classification. A filter-based multi-feature ranking-based voting framework was implemented on different multiple-based classifiers. Experimental results were evaluated on different multi-modal facial features for the automatic emotions listener using a speech synthesis library. The evaluation results showed that the proposed model had better feature classification, feature selection, prediction, and runtime than traditional approaches on heterogeneous facial databases.