نهمین کنگره نقشه برداری مغز

نمایش مقاله

Title:

Towards Interpretable Affective Computing: A Real-Time Convolutional Network for Joint Emotion and Gender Classification

Subject:

Advanced Techniques in Signal Acquisition and Image Processing

Presenter Author:
Name:	Elias Ebrahimzadeh

Author:
Name 1:	Elias Ebrahimzadeh - Corresponding - Presenter
Affiliation 1:	School of Cognitive Sciences, Institute for Research in Fundamental Sciences, Tehran, Iran, School of Electrical and Computer Engineering, College of Engineering, University of Tehran, Tehran, Iran.

Name 2:	Mostafa Asgarinejad
Affiliation 2:	Department of Cognitive Psychology, Institute for Cognitive Science Studies, Tehran, Iran

Name 3:	Melika Akbarimehr
Affiliation 3:	Neuroscience Research Center, Faculty of Medicine, Qom University of Medical Sciences, Qom, Iran.

Name 4:	Lila Rajabion
Affiliation 4:	School of Graduate Studies, SUNY Empire State university , Brooklyn, NY, USA.

Name 5:	Hamid Soltanian-Zadeh
Affiliation 5:	School of Cognitive Sciences, Institute for Research in Fundamental Sciences, Tehran, Iran, School of Electrical and Computer Engineering, College of Engineering, University of Tehran, Tehran, Iran.

Name 6:
Affiliation 6:

Name 7:
Affiliation 7:

Name 8:
Affiliation 8:

Name 9:
Affiliation 9:

Name 10:
Affiliation 10:

Abstract

Original Research

introduction :

The ability to accurately recognize facial expressions and gender in real-time is a foundational challenge with significant implications for social robotics, human-computer interaction (HCI), and affective computing. While deep learning has dramatically advanced offline performance on these tasks, a critical gap persists in deploying models that are simultaneously accurate, computationally efficient, and interpretable for real-time analysis. This paper addresses this gap by proposing a unified, real-time convolutional neural network (CNN) framework capable of simultaneous facial emotion and gender recognition in a single, efficient forward pass.

Methods:

We designed a streamlined yet powerful CNN architecture to process raw facial images and perform dual classification. The network hierarchically extracts discriminative spatial features through a series of convolutional and pooling layers, culminating in dedicated branches for classifying six basic emotion categories (happiness, sadness, anger, fear, surprise, disgust) and two gender classes (male, female). To ensure robustness and prevent overfitting, the architecture incorporates modern design principles, including batch normalization and advanced regularization techniques. Crucially, the model is integrated with a guided backpropagation visualization module, allowing for real-time analysis of the learned feature maps and weight dynamics across layers, thereby enhancing model transparency and interpretability.

Results:

The proposed system was rigorously evaluated on established public benchmarks. It achieved state-of-the-art performance on the challenging FER-2013 dataset, attaining an emotion recognition accuracy of 85%. The integrated gender recognition branch also demonstrated high precision. The visualization module successfully identified the key facial regions—such as eyes, mouth, and brow—that the model prioritizes for its predictions, validating its learning process. Furthermore, the framework was validated through deployment in a complete, integrated vision system, demonstrating its capability to perform unified face detection, gender classification, and emotion recognition in real-time with minimal latency, thus confirming its practical viability.

Conclusion:

This work demonstrates that the integration of a modern CNN design, strategic regularization, and embedded visual explainability is essential for transitioning from high offline accuracy to effective real-time deployment. The presented framework successfully bridges this gap, offering a scalable architecture and an open-source implementation. By providing both high performance and critical interpretability, this system serves as a robust foundation to support further innovation and application in real-time affective vision systems across diverse fields including interactive media, customer analytics, and assistive technologies.

Keywords:

Facial expressions, Gender Classification, Convolutional neural networks, Facial Emotion Recognition, Deep Learning.