v0.0.4

📝 Word Recognition System – v0.0.4

Fourth release of the Word Recognition (OCR) System, introducing a confidence-driven ensemble inference engine that dynamically selects the most reliable prediction across multiple CRNN architectures. This release significantly improves real-world accuracy, robustness, and interpretability.

Highlights:

• Confidence-based multi-model selection

  • Both Baseline CRNN and Transfer-Learning CRNN are executed per input
  • Final prediction is selected using CTC negative log-likelihood confidence
  • Eliminates heuristic thresholds and fallback logic

• Likelihood-aware ensemble inference

  • Prediction confidence derived directly from ctc_decode log probabilities
  • Ensures statistically grounded model arbitration
  • Higher-confidence prediction consistently correlates with higher accuracy

• Model-specific preprocessing pipelines

  • Dynamic preprocessing based on model architecture
  • Grayscale variable-width handling for baseline CRNN
  • Fixed-width RGB pipeline for transfer-learning CRNN

• Improved recognition accuracy

  • Challenging real-world word images
  • Mixed-case and stylized fonts
  • Low-contrast and noisy inputs
  • Short and long word sequences

• Deployment-ready inference layer

  • Unified decoding interface
  • Clean separation of preprocessing, inference, and confidence evaluation
  • Stable Flask deployment with confidence visualization support

This release upgrades the system from a confidence-aware single-model OCR pipeline to a likelihood-based ensemble recognition system. By leveraging CTC log-probability scores as a decision criterion, the system achieves more reliable predictions without increasing model complexity or inference fragility.

v0.0.3

📝 Word Recognition System – v0.0.3

Third release of the Word Recognition (OCR) System, focusing on prediction quality, model selection intelligence, and deployment-grade inference enhancements. This version introduces multi-model evaluation, confidence-aware decoding, and improved validation behavior for real-world text images.

Highlights:

• Dual-model inference pipeline
  • Baseline CRNN and transfer-learning CRNN evaluated per input
  • Automatic selection of the best prediction based on confidence scoring
• Confidence-aware CTC decoding
  • Token-level probability aggregation for reliable confidence estimation
  • Enables confidence visualization in UI and downstream decision logic
• Improved sequence decoding stability
• Inference-only optimized backbone
• Enhanced real-world robustness
  • Improved handling of:
  • Mixed-case text
  • Low-contrast word images
  • Short tokens and acronyms
• Confidence-enabled web interface

This release elevates the system from a single-model OCR pipeline to a confidence-driven, multi-model recognition engine. The improvements significantly enhance prediction reliability and transparency, making the system more suitable for real-world deployment, user-facing applications, and academic demonstrations.

v0.0.2

📝 Word Recognition System – v0.0.2

Second release of the Word Recognition (OCR) System, featuring major architectural and training improvements through transfer learning, data augmentation, and enhanced sequence modeling for higher accuracy and robustness.

Highlights:

• Pretrained VGG16 backbone for feature extraction (New)
• Transfer learning–based CRNN architecture with deeper representation
• Improved CTC-based sequence modeling and decoding
• Fixed-size RGB image pipeline for stable inference (32 × 256 × 3)
• Advanced data augmentation:
  • Random brightness and contrast
  • Geometric rotation using KerasCV
• Improved generalization on challenging fonts and casing
• Higher recognition accuracy on unseen word images
• Clean separation of training-only CTC logic and inference backbone
• Optimized inference pipeline for Flask deployment
• Production-ready TensorFlow .keras model

This release significantly improves recognition quality and model robustness by leveraging pretrained CNN features and modern augmentation strategies, while maintaining a lightweight and deployable Flask-based inference system.

v0.0.1

📝 Word Recognition System – v0.0.1

Initial release of the deep learning–based Word Recognition (OCR) System, capable of recognizing single-word images using a CRNN architecture with CTC decoding, deployed via a Flask web application.

Highlights:

• Image-based single-word text recognition
• CNN + Bidirectional LSTM (CRNN) architecture
• CTC (Connectionist Temporal Classification) decoding for sequence prediction
• Variable-width image handling (no fixed padding required)
• Trained on Synth90k (100k) synthetic word images
• TensorFlow .keras production model
• Lightweight Flask web application
• Simple HTML + CSS user interface
• Real-time inference on uploaded images

This release provides a complete end-to-end OCR pipeline, including model training, decoding logic, and a web-based inference interface, designed for educational and experimental purposes.