Face expression recognition based on NGO-BILSTM model

Area of Science:

• Computer Vision
• Artificial Intelligence
• Deep Learning

Background:

• Facial expression recognition is a key area in AI.
• Deep learning models, like BILSTM networks, show promise but require hyperparameter optimization.
• Hyperparameter tuning presents a significant challenge for BILSTM network performance.

Purpose of the Study:

• To introduce a new optimization algorithm for BILSTM hyperparameters.
• To improve facial expression recognition accuracy using an optimized BILSTM network.
• To evaluate the proposed optimization method on benchmark datasets.

Main Methods:

• A Northern Goshawk optimization (NGO) algorithm was developed.
• The NGO algorithm was used to optimize BILSTM network hyperparameters.
• Performance was evaluated on FER2013, FERplus, and RAF-DB datasets, considering diverse factors.

Main Results:

• The NGO-optimized BILSTM network achieved higher accuracy than VGG16 on FER2013 and FERPlus.
• On the RAF-DB dataset, accuracy reached 89.72%, outperforming recent algorithms by up to 9.63%.
• The method demonstrated superior performance compared to DLP-CNN, gACNN, pACNN, and LDL-ALSG.

Conclusions:

• The NGO algorithm effectively optimizes BILSTM hyperparameters for facial expression recognition.
• This approach significantly enhances the performance of facial expression recognition models.
• A novel and effective method for BILSTM hyperparameter optimization in this domain is presented.