An Empirical Study on the Effect of Face Occupancy on the Generalization Performance of CNN Models

DOI

10.2991/978-94-6463-540-9_61How to use a DOI?

Keywords

Face Occupancy; Facial Expression Recognition; CNN Generalization

Abstract

This empirical study investigated the impact of face occupancy on the generalization performance of Convolutional Neural Networks (CNNs), specifically focusing on three widely-used architectures: ResNet50, VGG16, and MobileNetV2. The face occupancy ratio, defined as the proportion of the image occupied by the face, is hypothesized to affect the model’s ability to generalize across varying conditions. The study employs two benchmark datasets, EFE and FER-2013, to conduct this study’s experiments. The datasets are preprocessed, and face detection is performed using Multi-task Cascaded Convolutional Networks (MTCNN). The face occupancy ratio is calculated for each image and categorized into bins for detailed analysis. This study trains and evaluates the CNN models on these datasets and calculates key performance metrics, including accuracy, loss, and mean absolute error (MAE). This study’s analysis includes the Pearson correlation coefficient to measure the linear relationship between face occupancy and model accuracy. Additionally, the study visualizes model performance using confusion matrices and scatter plots with regression lines, highlighting the trends in model accuracy relative to face occupancy. Results indicate a strong positive correlation between face occupancy and model accuracy across all three models, with ResNet50 showing the highest Pearson correlation coefficient, followed by VGG16 and MobileNetV2. These findings suggest that higher face occupancy ratios contribute to better generalization performance in CNNs, offering valuable insights for improving facial recognition systems.

Copyright

© 2024 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Jialin Tian
PY  - 2024
DA  - 2024/10/16
TI  - An Empirical Study on the Effect of Face Occupancy on the Generalization Performance of CNN Models
BT  - Proceedings of the 2024 2nd International Conference on Image, Algorithms and Artificial Intelligence (ICIAAI 2024)
PB  - Atlantis Press
SP  - 611
EP  - 620
SN  - 2352-538X
UR  - https://doi.org/10.2991/978-94-6463-540-9_61
DO  - 10.2991/978-94-6463-540-9_61
ID  - Tian2024
ER  -