AdaBoost based Random forest model for Emotion classification of Facial images

Area of Science:

• Computer Science
• Artificial Intelligence
• Image Processing

Background:

• Social media image sharing is prevalent, with users posting facial images depicting various emotions.
• Accurate facial expression mining from these images is crucial for understanding user sentiment.
• Existing emotion analysis models often lack sufficient accuracy and reliability for social media data.

Purpose of the Study:

• To enhance the performance of emotion analysis in social media image posts.
• To develop a novel model for accurate and reliable facial expression mining.
• To improve the efficiency of emotion classification from visual data.

Main Methods:

• Feature extraction using the 2D Ortho-normal Stockwell Transformation (DOST) method.
• Feature selection implemented via bi-variate t-test.
• Classification using an AdaBoost based Random Forest classifier for Emotion Classification (ARFEC).

Main Results:

• The ARFEC model demonstrated high accuracy rates across multiple datasets: 89.5% on Flickr8k, 92.5% on CK+, and 89.5% on FER2013.
• Comparative analysis showed ARFEC outperformed Support Vector Machine and K-Nearest Neighbors in overall accuracy.
• The model effectively classifies six distinct emotional states: Sad, Fear, Awful, Happy, Surprised, and Satisfied.

Conclusions:

• The proposed ARFEC model offers a significant improvement in facial emotion analysis accuracy and reliability.
• Transformed features and the ARFEC approach are effective for mining emotions from social media images.
• This research contributes a robust method for understanding human emotions expressed visually online.