DigitEye

This project implements a neural network using TensorFlow and Keras to recognize handwritten digits from the MNIST dataset. The model can also predict custom handwritten digit images.

---

Features

• Customizable Neural Network: Dynamically set the number of layers, neurons, activation functions, and epochs.
• Data Augmentation: Increases the training dataset using techniques like rotation, shifting, and zooming to improve generalization.
• Custom Predictions: Load your custom digit images and let the model predict them.
• Model Saving and Loading: Train a new model or load an existing one.

---

Installation

Prerequisites

• Python 3.x
• TensorFlow
• NumPy
• OpenCV
• Matplotlib

Install the required libraries with:

pip install tensorflow numpy opencv-python matplotlib

Note for Linux users: It is recommended to create a virtual environment to install these libraries, as direct pip installation may lead to system-wide conflicts. Use the following commands to set up a virtual environment:

python3 -m venv digit_eye_env
source digit_eye_env/bin/activate
pip install tensorflow numpy opencv-python matplotlib

Usage

1. Clone the repository

git clone https://github.com/AleksaVukadinovic/DigitEye
cd DigitEye

2. Run the script

python digit_recognition.py

3. Follow the console prompts

Materials Used

This project draws inspiration from and builds upon the following resources:

• DeepLearning Series by 3Blue1Brown
  An excellent visual introduction to the concepts behind neural networks and deep learning.

• Handwritten Digit Recognition by NeuralNine
  A practical guide to implementing a digit recognition system.

• Veštačka inteligencija (Artificial Intelligence)
  A comprehensive textbook by Mladen Nikolić and Predrag Janičić, providing theoretical insights into artificial intelligence.

• Using Neural Networks to Recognize Digits
  A detailed article explaining the mathematics and algorithms behind digit recognition using neural networks.