Project leader: Prof. Ing. Jan Flusser, DrSc.

Department: ZOI

Supported by (ID): GA102/08/1593

Duration: 2008 - 2010

Abstract:

This project concerns superresolution (SR) of digital images and videos. SR means an improvement of spatial resolution of images beyond the physical limit of the sensor. The central idea of SR is combining a sequence of low-resolution images in order to produce higher resolution image. Current SR techniques use very simplified image acquisition models only, which prevents them from being applicable in practice. We propose a realistic acquisition model, comprising constant sensor blur, unknown volatile blur(s), downsampling (by non-integer factor in general), and geometric deformations. We show the SR problem is equivalent to a minimization of certain functionals with proper regularization terms. We focus on SR of a single gray-level image, SR of color images, SR of video, and SR of images of 3D scenes. Each of these cases requires specific treatment. The theoretical results will be applied to enhance visual quality of ground-based and satellite astronomical observations. In addition to astronomy, we envisage numerous applications wherever cameras/imaging systems having insufficient resolution are used (security and surveillance systems, cell phones, web cameras, low-cost video cameras, etc.).