Bayesian Image Denoising using MRF Priors

A Maximum-A-Posteriori (MAP) Bayesian approach for image denoising that uses Markov Random Fields (MRFs) as the prior model

Usage:

See code/main.mlx and run it cell by cell.

code/main.mlx: The main script/driver program
General Functions:
- code/rrmse.m: Returns the RRMSE (Relative Root Mean Squared Error) between two images
- code/objectiveAndGrads.m: Returns the objective function value and the corresponding gradients
- code/adaptiveGradientDescent.m: Gradient Descent with a dynamic learning rate/step size
MRF Prior:
- code/fourNeighborSystem.m: The standard four-neighbor system for each pixel
- code/quadraticPrior.m: The quadratic loss function
- code/huberPrior.m: The Huber loss function
- code/dafPrior.m: A discontinuity adaptive function
Noise Model:
- code/weightedNoiseModel.m: Contains implementations of the Gaussian and the Rician noise models

(This was done as a course assignment for CS736: Medical Image Computing, Spring 2021, IIT Bombay)