Handwriting Recognition with CRNN

View on Kaggle
Check out a video demonstration here!

Overview

Developed a Convolutional Recurrent Neural Network (CRNN) model with Connectionist Temporal Classification (CTC) loss for recognizing handwritten text. The project achieved high accuracy by integrating attention mechanisms and advanced preprocessing techniques.

Key Features

• 85% Character Accuracy: Achieved high recognition accuracy by optimizing the CRNN model.
• Attention Mechanisms: Improved focus on relevant features for precise recognition.
• Preprocessing Techniques: Enhanced data quality through normalization and cropping.

Tools Used

• Programming Languages: Python
• Libraries/Frameworks: TensorFlow/Keras, OpenCV
• Dataset Management: Kaggle