Denoising of noisy MNIST dataset images using Conditional Random Fields

Noisy images of a particular digit are denoised using a trained CRF model. The CRF model is implemented using the PyStruct library

The problem statement requires the denoising of images of a particular digit using CRF's.

• CRFBasic.py Basic model with unary potentials defined by node pixel value.
• CRFBasic_Iterative.py Basic model with iterative saving of models at different stages of model training.
• CRFEdgeFeatures.py Model with unary potentials defined by node pixel value and pair-wise potential defined pixel values of connecting nodes.
• CRFNeighborhood.py Model with unary potentials defined by pixel values of node and neighboring nodes.
• CRFNeighborEdgeFeatures Combination of CRFEdgeFeatures and CRFNeighborhood.