Facial expression recognition based on active region of interest using deep learning and parallelism

Area of Science:

• Computer Vision
• Machine Learning
• Human-Computer Interaction

Background:

• Automatic facial expression tracking is crucial for various applications like VR and security.
• Traditional methods can be computationally expensive.
• Identifying key facial areas (Active Regions of Interest - AROIs) offers a more efficient approach.

Purpose of the Study:

• To develop a cost-efficient and accurate facial expression recognition framework.
• To investigate the effectiveness of using specific AROIs for expression classification.
• To enhance recognition speed through parallel processing and a novel synthesis method.

Main Methods:

• Face normalization using pose estimation.
• Segmentation of face images into regions and identification of four AROIs (nose-tip, eyes, lips).
• Convolutional Neural Network (CNN) with ten-fold cross-validation for classification.
• Parallel processing of AROIs and a decision-tree-level synthesis framework.

Main Results:

• Achieved high accuracy rates, up to 98.26% on CK+ and 94.423% on JAFFE datasets.
• Overall accuracy of 95.27% with a 2.8% improvement using the synthesis method.
• Processing time was accelerated by three times due to parallel implementation.

Conclusions:

• The proposed framework effectively recognizes facial expressions with high accuracy and efficiency.
• The AROIs approach combined with parallel processing and synthesis significantly improves performance.
• This method demonstrates robustness and potential for real-world applications.