Convolutional Neural Network for Text Scene Recognition (TSR)

This project implements a Convolutional Neural Network (CNN) for Text Scene Recognition (TSR), designed to detect and recognize text in natural scene images.

Project Overview

Text Scene Recognition (TSR) is a computer vision task that involves detecting and recognizing text within images of natural scenes, such as street signs, billboards, and product labels. This project utilizes a CNN architecture to achieve accurate text recognition in challenging environments.

Files

• TSR_final-checkpoint.ipynb: Jupyter notebook containing the implementation of the Text Scene Recognition model with training and evaluation code
• main.pdf: Main documentation or research paper related to the project
• README.md: This file

Features

• CNN-based architecture optimized for scene text recognition
• Preprocessing pipeline for scene text images
• Training and validation modules
• Model checkpoint saving and loading functionality
• Evaluation metrics for assessing model performance

Requirements

The project likely requires the following libraries (specific versions may vary):

• Python 3.x
• TensorFlow/Keras or PyTorch
• NumPy
• OpenCV
• Matplotlib
• Pandas

Usage

To run the model:

1. Open the TSR_final-checkpoint.ipynb notebook in Jupyter
2. Follow the instructions in the notebook to train or evaluate the model
3. Adjust hyperparameters as needed

Dataset

This project may utilize standard scene text datasets such as:

• ICDAR dataset
• Street View Text dataset
• COCO-Text dataset

Model Architecture

The CNN architecture likely includes:

• Convolutional layers for feature extraction
• Pooling layers for dimensionality reduction
• Fully connected layers for classification
• Activation functions (ReLU, Softmax)

Results

For detailed experimental results and performance metrics, please refer to the main.pdf document.

Contributing

Contributions to improve the model performance or extend functionality are welcome.

License