Geometrical Facial Expression Recognition Approach Based on Fusion CNN-SVM

摘要

Facial expression recognition (FER) effectively enhances human-computer interaction, robot interfaces, and emotion-aware smart agent systems.Most existing FER algorithms focus solely on extracting facial features from the pixels of the face, disregarding the relative geometric positions that depend on facial landmark points.However, these approaches need to be revised regarding accuracy and robustness.This paper introduces a novel approach to FER that combines convolutional neural network (CNN) and support vector machine (SVM) techniques to extract hybrid features, thereby enhancing discriminative power.The proposed system employs the β-skeleton undirected graph for improved geometric feature extraction through several pre-processing stages.A 1D-CNN is utilized for training the geometric features while simultaneously training the same image using a 2D-CNN.The resulting feature vectors from both sub-networks are merged for SVM classification.The performance of the proposed system is evaluated on two widely used datasets: the extended Cohn-Kanade (CK+) dataset and the Japanese female facial expression (JAFFE) dataset, which are commonly employed in face recognition research.Experimental results demonstrate the superiority of the proposed system over existing methods, achieving a recognition accuracy of 96.19% on the CK+ dataset and 89.23% on the JAFFE dataset.