Analyzing the Effectiveness of Sobel and Laplacian Filters Integration in Facial Emotion Recognition

This paper focuses on improving the accuracy of the facial emotion recognition (FER) system using the FER2013 dataset by applying both single and combined Sobel and Laplacian filters. These filters are known for their edge detection and feature extraction abilities, which are investigated to determine their effect on emotion recognition accuracy. The filtered data is further processed to train the convolutional neural network (CNN) architectures. The performance of CNN models is extensively examined to identify any differences in classification accuracy due to the filter used. Through thorough experimentation and analysis, it is discovered that models trained on combined filter data perform better in class-wise classification than those trained on data filtered using Sobel or Laplacian filters separately. Our experiment results demonstrate that deep learning models trained with the combined filter exhibit superior accuracy and lower loss than those trained with single filters. Our experiment results and analysis show the effectiveness of the mixed filter strategy in improving CNN models’ discriminative skills for FER. The FER results from our approach provide vital insights into improving recognition accuracy by combining modern image processing techniques with deep learning approaches, supporting the creation of more empathic and responsive artificial intelligence systems. The data and code that support the findings of this study are openly available in https://doi.org/10.5281/zenodo.11046710.