A Matrix based implicit prior for joint demosaicing and super-resolution

OBJECTIVE

To implement joint demosaicing and super-resolution for color images:

1. To perform demosaicing of missing color components using interpolation algorithm
2. To super-resolute RGB color images obtained from Bayer image
3. To learn the image level information between color class using efficient matrix based priors

The proposed method involves three steps:

i) New calculation of gradients
ii) Edge classification based on color differences and gradient directions
iii) Refinement stage using a weighted sum strategy for artifacts

The proposed method for super-resolution involves:

i)Implement a example-based super-resolution algorithm to obtain the implicit prior
ii)Regression analysis is done to obtain a relation between the corresponding patches in low resolution and high resolution image from the training set
iii)Matrix-based regression is used instead of vectors to preserve structural information.