Abstract
This project deals with restoration of images corrupted by additive gaussian noise. The algorithm used is based on optimal estimation using the properties of MRF’s (Markov Random Fields).
The Problem
While optimal estimation ( ) is the theoretically best restoration method, assuming a probabilistic model and under least mean square error, it demands very high calculation complexity, and is therefore not practical to implement due to limited computational resources.
The Solution
MRF’s are 2-D analogies of Markov Chains. Using MRF’s we can simplify the complexity of calculations required for an approximation of the optimal estimation.
Block Diagram
Tools
The algorithm was implemented using Matlab 5.3
Data analysis was done using MS Excel 2000.
Example
In picture B we can see a magnification of the marked area from the clean picture A. gaussian noise was added to the picture which was then restored using MRF restoration (picture C) and Wiener restoration (picture D). we can see the wiener restoration over-smoothed the image, while the MRF restoration kept the original characteristics of the image.
Conclusions
Though using an MRF-based algorithm reduced calculation complexity, the execution time remained very high compared to standard restoration techniques.
The algorithm achieved good results for some images. In some cases, however, the parameter estimation per block was inefficient due to the varying characteristics of real images.
Acknowledgement
We would like to thank Tsachy Weissman for supervising this project. Our acknowledgement goes also to the laboratory staff and to the Ollendorff Minerva Center Fund for their support.
