Data fusion techniques for super-resolution imaging

Abstract

In this paper, we present two new techniques of using data fusion, based on the modality of the data generation process, to generate a super-resolved image from a sequence of low-resolution image intensity data. First, we develop a generalized interpolation scheme wherein an image is decomposed into appropriate subspaces, interpolation is carried out in individual subspaces and subsequently the interpolated values are transformed back to the image domain. Various structural properties of the image, such as 3D shape of an object, regional homogeneity, local variations in scene reflectivity, etc., can be better preserved during the interpolation process. The issue of preserving the structure in an image has received very little attention in the information fusion literature and we concentrate on this issue. In the second method, the data to be fused consist of a sequence of decimated, blurred and noisy versions of the high-resolution image. The high-resolution image is modeled as a Markov random field (MRF) and a maximum a posteriori (MAP) estimation technique is used to fuse the data to obtain a super-resolved image. The proposed technique does not require sub-pixel registration of given observations as most other techniques in the literature do. A simple gradient descent method is used to optimize the functional. The discontinuities in the intensity process can be preserved by introducing suitable line processes for improved result, albeit at the cost of computation.