Related Experiment Video
Updated: Aug 18, 2025
08:27
Author Spotlight: Efficient Image Recognition Using Directional Gradient Histogram Technique and Support Vector Machines
Published on: January 5, 2024
1.2K
Vision-based Pakistani sign language recognition using bag-of-words and support vector machines
Muhammad Shaheer Mirza1, Sheikh Muhammad Munaf2, Fahad Azim3
1Department of Biomedical Engineering, Faculty of Engineering, Science, Technology and Management, Ziauddin University, Karachi, Pakistan. shaheer.mirza@zu.edu.pk.
Scientific Reports
|December 9, 2022
Summary
This study developed a vision-based system for recognizing Pakistani Sign Language (PSL) alphabets. The system achieved high accuracy for both static and dynamic signs, aiding communication for the deaf community.
Area of Science:
- Computer Vision
- Human-Computer Interaction
- Linguistics
Background:
- Communication is essential for daily activities, posing challenges for the deaf population who use sign languages.
- Pakistani Sign Language (PSL) is used by over 250,000 deaf individuals in Pakistan.
- A robust sign language recognition system can significantly improve accessibility and communication for PSL users.
Purpose of the Study:
- To collect a dataset of static and dynamic PSL alphabet signs.
- To develop and evaluate a vision-based system for PSL alphabet recognition.
- To utilize Bag-of-Words (BoW) and Support Vector Machine (SVM) techniques for sign language recognition.
Main Methods:
- Collected 5120 static images and 353 videos (45,224 frames) of PSL alphabets from 10 native signers.
- Preprocessed images by resizing, converting to grayscale, and segmenting using Thresholding.
- Extracted features using Speeded Up Robust Features (SURF) and clustered descriptors with K-means for Bag-of-Words (BoW) creation.
- Employed fivefold cross-validation for training and testing the Support Vector Machine (SVM) classifier.
Main Results:
- Achieved 97.80% classification accuracy for static PSL signs at 750x750 image dimensions with 500 Bags.
- Obtained 96.53% classification accuracy for dynamic PSL signs at 480x270 video resolution with 200 Bags.
- Demonstrated the effectiveness of the BoW and SVM approach for PSL recognition.
Conclusions:
- The developed vision-based system effectively recognizes static and dynamic Pakistani Sign Language alphabets.
- The system shows high accuracy, offering a promising tool to facilitate communication for the deaf community in Pakistan.
- This research contributes to the advancement of sign language recognition technology.
