Image prior combination in super-resolution image registration & reconstruction

Abstract

In this paper a new combination of image priors is introduced and applied to Super Resolution (SR) image reconstruction. A sparse image prior based on the 1 norms of the horizontal and vertical first order differences is combined with a non-sparse SAR prior. Since, for a given observation model, each prior produces a different posterior distribution of the underlying High Resolution (HR) image, the use of variational posterior distribution approximation on each posterior will produce as many posterior approximations as priors we want to combine. A unique approximation is obtained here by finding the distribution on the HR image given the observations that minimize a linear convex combination of the Kullback-Leibler (KL) divergences associated with each posterior distribution. We find this distribution in closed form. The estimated HR images are compared with images provided by other SR reconstruction methods.

A l1 Norm Based Image Prior Combination in Multiframe Superresolution

“…It is easy to see that the 1 prior method in [20] is a special case of our proposed method. We have not compared our method with the one in [21], because how to determine the combination coefficients is not given in [21].…”

“…Later the 1 prior [20] was introduced and proven to have better performance. In [21], the combination of the sparse prior and nonsparse SAR prior was proposed, but the combination coefficient had to be determined by the hand empirically. In [17], the author suggested a nonstationary image prior combination method based on a general combination of spatially adaptive image filters, and the combination coefficient can be estimated automatically.…”

A Parallel Framework for Video Super-Resolution

“…Around 1990, the term "super-resolution" was incorporated in the literature by Irani and Peleg [3]. Since then, many super-resolution algorithms have been proposed, using different approaches [4], [5], [6], [7].…”

“…In this class of methods, different approaches may be used, such as reconstruction of non-uniformly spaced samples [9], backprojection [10], [7], [11], and stochastic models [12], [6], [13].…”

Posterior mean super-resolution with a compound Gaussian Markov random field prior

“…In a Bayesian framework, what prior we use is quite important. For example, various Markov random field (MRF) priors [4]- [10], the total variation (TV) prior [11], [12], the Huber prior [13], and patchbased priors [14] have been used in image processing. These can represent image properties well and have good performance in SR, image restoration, and other applications.…”