👁️ Computer Vision

Handwritten Text Recognition

OCR System for Cursive and Block Handwriting

2 months Solo Project Completed

Overview

LSTM-based OCR system transcribing handwritten notes with 85% accuracy across diverse handwriting styles.

The Problem

Digitizing handwritten notes and forms is labor-intensive and error-prone when done manually.

The Solution

Implemented CRNN (CNN + LSTM) architecture with CTC loss for sequence-to-sequence transcription. Trained on IAM Handwriting Database with extensive augmentation.

Results & Impact

85% Character Accuracy On diverse styles
78% Word Accuracy Complete word recognition
0.3s Processing Speed Per line of text

Key Impact

  • Automated form processing pipeline
  • Reduced manual data entry by 70%
  • Enabled searchable digital archives

Technologies Used

PythonTensorFlowLSTMCTC LossOpenCV