Facial Emotion Recognition Using Wavelet Scattering Transform And Deep Learning Techniques

Recently, a lot of interest has been focused on improving systems that dynamically respond to the emotional state of a person. Facial expressions serve as a primary channel for understanding emotions, making facial analysis a critical component of emotion recognition systems. In this paper, we propose a novel methodology that combines wavelet scattering and deep learning. The process begins with facial detection using Multi-Task Cascaded Convolutional Neural Networks (MTCNN). Features are then extracted using the Wavelet Scattering Transform (WST), which provides robust and informative representations. The extracted features are evaluated across three deep learning models: Convolutional Neural Networks (CNNs), Graph Convolutional Networks (GCNs), and Capsule Networks (CapsNets). The experimental results on the Extended Cohn Kanade (CK+) and Japanese Female Facial Expression (JAFFE) databases demonstrate the effectiveness of the proposed architecture.

I. Introduction

To develop effective analytical models for Facial Emotion Recognition (FER), researchers are exploring methods to replicate strategies that optimize algorithm performance. Wavelet scattering networks, in particular, enable the extraction of low-variance features from image data with minimal configuration, enhancing machine learning and deep learning applications.