Project Overview

The exercises focus on:

1. Generating a handwritten '8' using a fully connected neural network.
2. Denoising noisy images of the digit '8' using gradient descent optimization.
3. Reconstructing reduced-resolution images of the digit '8' while adding noise.

Structure of the Exercises

1. Problem 2.1 - A generative model creates a handwritten '8' using two fully connected layers with ReLU and sigmoid activations. The input is a random Gaussian vector.
2. Problem 2.2 - Reconstructing noisy images of a '8' by retaining a subset of pixel values.
3. Problem 2.3 - Reconstructing images with reduced resolution and added noise using matrix transformations.

Files in this Repository

• data21.mat, data22.mat, data23.mat: MATLAB data files containing matrices and vectors used for the neural network transformations and image reconstructions.
• ex1.m, ex2.m, ex3.m: MATLAB scripts corresponding to each problem.

Usage Instructions

1. Setup

   • Ensure MATLAB is installed on your system.
   • Load the data files using the scripts provided (e.g., load('data21.mat')).

2. Running the Scripts

   • Each script (ex1.m, ex2.m, ex3.m) corresponds to a specific problem in the exercise.
   • You can modify input parameters (e.g., N values for pixel retention) as described in the code to observe different results.

3. Visualization

   • The generated images, denoised images, and reconstructed images will be displayed using MATLAB's imshow function.