---

Handwritten Digit Recognition System

This project implements a system for capturing, processing, and recognizing handwritten digits using Python, OpenCV, and Keras. It includes functions for image capture, preprocessing, dataset creation, and model training.

Features

• Image Capture: Captures images of handwritten digits through a webcam.
• Image Preprocessing: Processes images to the required size and format.
• Dataset Splitting: Splits images into training and test datasets.
• Model Training: Trains a convolutional neural network (CNN) on the dataset.

Setup

To run this project, you need to have Python installed along with the following packages:

• OpenCV
• NumPy
• Keras
• scikit-learn

You can install these packages using pip:

pip install opencv-python numpy keras scikit-learn

Usage

1. Capture Images: This script will use your default webcam to capture images of digits. Make sure your webcam is accessible and correctly configured on your system. Run the capturing function with:

   capture_images()

2. Preprocess Images: To preprocess the captured images for model consumption, run:

   preprocess_images("dataset", "preprocessed_dataset")

3. Split Dataset: Split the preprocessed images into training and testing datasets:

   split_dataset("preprocessed_dataset", "train", "test")

4. Train Model: Train the CNN on the training dataset:

   train_model("train", "test")

File Descriptions

• capture_images(): Captures images of digits 0-9 using a webcam.
• preprocess_images(input_dir, output_dir): Resizes images to 28x28 pixels and converts them to grayscale.
• split_dataset(input_dir, train_dir, test_dir): Splits the dataset into training and testing sets.
• train_model(train_dir, test_dir): Loads the dataset, creates the CNN model, and trains it.

Model Architecture

The CNN model used includes the following layers:

• Conv2D: 32 filters, 3x3 kernel size, ReLU activation
• MaxPooling2D: 2x2 pool size
• Flatten
• Dense: 128 units, ReLU activation
• Dense: 10 units (output layer), softmax activation

The model is compiled with the Adam optimizer and categorical crossentropy loss function. It evaluates the accuracy of the trained model on the test dataset.

Conclusion

This script provides a basic yet comprehensive approach to digit recognition from handwritten images. It demonstrates the end-to-end process from image capture to model training and evaluation.

---