Role: Independent student project
Tools: Python, NumPy, Matplotlib, scikit-image

Background

Image denoising can be viewed as a simple inverse problem: recover a clean image from an observed image corrupted by noise. This project starts with a deliberately small scope so the mathematical assumptions are easy to inspect.

Method

The MVP compares a baseline smoothing method with a regularized reconstruction objective. The main idea is to balance fidelity to the noisy observation with a penalty that discourages unstable or overly rough reconstructions.

Results

The early experiments show the expected tradeoff: stronger regularization removes more noise but can blur edges and fine details. Visual comparisons and simple metrics help make that tradeoff concrete.

Technical Stack

  • Python for experiment scripts
  • NumPy for array operations
  • Matplotlib for visual comparison plots
  • scikit-image for sample images and basic quality metrics

Next Steps

  • Add total variation regularization
  • Compare Gaussian, salt-and-pepper, and Poisson noise
  • Write a short note explaining the optimization objective
  • Build a small interactive demo for parameter tuning