Expression Recognition Based on Gabor Filter

Abstract

We propose a comprehensive approach that integrates traditional methods with deep learning techniques to address the challenges of insufficient feature extraction, limited feature discriminability, and dimensionality catastrophe in facial expression recognition. This method begins by artificially extracting image features using Gabor filters to effectively capture relevant information within the images. Subsequently, Principal Component Analysis (PCA) is employed to reduce feature dimensions, controlling feature space while eliminating redundant information introduced by Gabor feature extraction, resulting in a novel feature representation. The reduced-dimensional features are fed into a Convolutional Neural Network (CNN), where multi-layered convolution and pooling operations extract abstract features and enable classification training. Experimental results demonstrate that the proposed method achieves impressive performance on the CK+ dataset, with an accuracy of 98.98%, signifying a substantial improvement over traditional approaches.