sherif1313/Arabic-English-handwritten-OCR-v3
The sherif1313/Arabic-English-handwritten-OCR-v3 is a 3 billion parameter multimodal model, built on Qwen/Qwen2.5-VL-3B-Instruct, specifically fine-tuned for Arabic and English handwritten Optical Character Recognition (OCR). This model achieves an average Character Error Rate (CER) of 1.78%, outperforming commercial solutions like Google Vision API by 57% on similar tasks. It features adaptive sharpness enhancement, 99.2% skew correction accuracy, and specialized processing for connected characters, making it highly effective for digitizing diverse handwritten documents.
Loading preview...
Arabic-English-handwritten-OCR-v3: Advanced Multilingual Handwriting Recognition
The Arabic-English-handwritten-OCR-v3 is a 3 billion parameter multimodal model, fine-tuned from Qwen/Qwen2.5-VL-3B-Instruct, designed for highly accurate Optical Character Recognition (OCR) of Arabic, English, and multilingual handwriting from images. This model represents a significant advancement in the field, achieving an average Character Error Rate (CER) of 1.78%, which is a 57% improvement over commercial solutions like Google Vision API.
Key Capabilities
- Superior Accuracy: Achieves an average CER of 1.78% on diverse handwritten documents, significantly outperforming industry benchmarks.
- Advanced Image Pre-processing: Incorporates adaptive sharpness enhancement to improve accuracy on blurred text by 15-20% and boasts 99.2% accuracy in skew correction.
- Cursive/Connected Text Handling: Features special processing for connected characters, reducing error rates by 12-18% on such text.
- Efficient Processing: Includes auto-resolution reduction for images larger than 1200x1200 pixels, speeding up processing by 3-5 times while maintaining quality.
- Multilingual Support: Primarily optimized for Arabic and English, with enhanced English vocabulary and support for other languages like Persian, Urdu, and Turkish.
- Dynamic Equilibrium Theorem: Developed based on a scientific discovery in model training, ensuring optimal predictive accuracy and resource usage.
Good For
- Digitizing Historical Manuscripts: Excellent for converting handwritten Arabic and other scripts from historical documents into searchable text for digital archives and cultural preservation.
- Academic and Linguistic Research: Analyzing handwriting styles and digitizing student work or notes.
- Commercial Applications: Automating the processing of handwritten forms, financial documents (e.g., checks), medical records, and invoices in various industries.
- Developers Seeking High-Performance OCR: Provides a free, high-accuracy alternative to commercial OCR APIs for Arabic and English handwritten text.