Thai Finger-Spelling Recognition Using a Cascaded Classifier Based on Histogram of Orientation Gradient Features

Area of Science:

• Computer Vision
• Machine Learning
• Human-Computer Interaction

Background:

• Hand posture recognition is crucial for applications like human-computer interaction (HCI), gaming, and sign language systems.
• Performance and robustness are key requirements for effective hand gesture recognition systems.
• Thai finger-spelling involves specific hand postures for representing alphabet letters.

Purpose of the Study:

• To develop an automatic classification system for recognizing 21 static hand postures in Thai finger-spelling.
• To utilize Histogram of Orientation Gradient (HOG) features and Adaptive Boost (AdaBoost) for robust hand posture classification.
• To optimize the training process by adjusting parameters like false positive rate (FPR), true positive rate (TPR), and number of training stages (N).

Main Methods:

• Employing Histogram of Orientation Gradient (HOG) features, focusing on regional image information rather than individual pixels.
• Implementing the Adaptive Boost (AdaBoost) learning technique to build a strong classifier from multiple weak classifiers.
• Collecting a dataset of 21 static hand posture images from 10 subjects for training and testing.
• Cascading multiple classifiers and using a correlation coefficient to differentiate similar hand postures.

Main Results:

• The system successfully classifies 21 distinct hand postures representing Thai finger-spelling.
• Parameter tuning through three experimental setups (FPR, TPR, N) led to optimized training models.
• An average accuracy of approximately 78% was achieved across all classifier experiments.
• Correlation coefficients effectively distinguished between visually similar hand postures.

Conclusions:

• The proposed system demonstrates a viable approach for automatic Thai finger-spelling recognition.
• The combination of HOG features and AdaBoost provides a robust method for hand posture classification.
• The system's accuracy suggests potential for practical applications in HCI and assistive technologies.