Abstract:
We propose a super-resolution algorithm based on local adaptation. In the proposed
algorithm, the mapping function from the low-resolution images to high-resolution
image is estimated by adaptation. Moreover, the property of the high-resolution image is
learned and incorporated in a regularization-based restoration. The proposed
regularization function is used as a general directional total variation with adaptive
weights. The adaptive weights of the directional total variation are estimated based on
the property of the partially reconstructed high-resolution image. The regularization
function can be thought as a linear combination of smoothness in different directions.
The convexity conditions as well as the convergence conditions are studied for the
proposed algorithm.