Handwritten Zero Generator – README

This project explores the generation and classification of synthetic handwritten digits—specifically focusing on the digit 0—using simple geometric sampling, pixel conversion, dataset creation, and an adversarial learning setup.

The notebook contains 5 main sections, each representing a building block of the full pipeline.

🚀 1. Zero Curve Generator (x, y coordinates)

The first part of the notebook generates a rough handwritten “0” shape using parametric equations.
A random ellipse is created using:

random scaling (a, b)

random rotation angle

wobble effect to simulate handwriting

translation noise

Output: an array of (x, y) coordinates representing a distorted zero.
This acts as a simple synthetic handwriting generator.

🖼️ 2. Converting Coordinates to Pixels

The second section takes the generated coordinates and converts them into a 28×28 grayscale pixel image, similar to MNIST.

This includes:

coordinate normalization

rasterization into a pixel grid

optional noise

visualization with matplotlib

Output: a black & white image of the generated zero.

🔢 3. Zero & Non-Zero Generator

In this stage, the dataset is expanded to include two classes:

✔️ Zero class

Generated using the parametric ellipse method.

❌ Non-zero class

Created by:

random strokes

noise-only patterns

or shapes that do not resemble a zero

This produces a simple but effective binary classification dataset.

🤖 4. Adversarial Classifier (Simple GAN-like Setup)

This section introduces an adversarial learning mechanism.
Instead of a full GAN, the idea is:

Generator: produces zeros and non-zeros

Classifier (Discriminator-like): tries to classify the images into zero vs non-zero

The generator tries to create samples that “fool” the classifier.
The classifier tries to correctly classify them.

This gives a first taste of adversarial training, even if lightweight.