Utilizing Optical Character Recognition for Handwritten Arabic Alphabet Learning
Click the Poster to View Full Screen, Right click to save image
Grant: IUSE Scholars Academy
Mysara Elsayed
College:
The Dorothy and George Hennings College of Science, Mathematics, and Technology
Major:
Computer Science
Faculty Research Advisor(s):
Daehan Kwak
Abstract:
The emergence of interactive language learning tools like Duolingo or Rosetta Stone has revolutionized the process of learning new languages, making it easier than ever before. While these tools excel in teaching spoken and basic written language skills, they face limitations when it comes to recognizing and deciphering handwritten texts.
This challenge creates a divide between people's willingness to engage in handwritten language learning, particularly in alphabets like Arabic. The different styles of handwriting, strokes, and character shapes pose difficulties for language learners, often discouraging them from continued learning. To address this gap, this project is underway to develop a program utilizing optical character recognition (OCR) techniques to assist language learners in deciphering and distinguishing letters from photos of handwritten texts, specifically focusing on the Arabic language.
Utilizing OCR unconventionally and breaking away from traditional translation applications, this project breaks down individual characters from handwritten Arabic texts, enhancing readability for users. To train the application on handwritten Arabic characters, the Arabic Handwritten Characters Dataset (AHCD) and the Hijja dataset were employed.
The chosen approach involves exploring various convolutional neural network (CNN) training models, assessing dataset compatibility, and optimizing accuracy. TensorFlow was selected as the preferred model for its accuracy and effectiveness in recognizing individual characters as separate entities. Future steps involve implementing TensorFlow training into an accessible OCR learning tool, offering a dynamic style of learning for recognizing individual letters from word compositions.
Beyond language learning, the tool's capabilities can extend to diverse applications. For instance, it could be employed in product design to automatically break down 3D objects into exploded 2D views, potentially streamlining the design process, offering benefits to design professionals and contributing to broader educational initiatives. The project signifies a step towards unlocking the versatile potential of OCR beyond its traditional applications.