This page contains information on the Research Project TIN2007-65533 "Super-resolución bayesiana de imágenes aplicada a vigilancia y seguridad", funded by the Ministerio de Educación y Ciencia de España from 2007 to 2010 with a budget of € 99,200.00.
The research team consists of 6 doctors and a graduate from the University of Granada with extensive experience in project theme.
This page will provide information on the project results and publications that derived from it.
If you are interested in using the software developed for the project please check this page.
The aim of the project is to obtain spatial high resolution images from one or a set of observed low resolution images which share the following characteristics: the blurring process of the observed images is not known and the objects of interest in those images belong to restricted domains.
The above described problem appears in many real situations like:
- Images captured by surveillance cameras. The goal is to improve the quality of, for instance, faces, registration plates, text and particular objects that may appear in the images.
- Images of means of transport: cars, military vehicles, aircrafts, ships, installations. The goal is to improve the visual quality of the objects of interest in those images for recognition or classification purposes.
To tackle the above problem we use the Bayesian paradigm:
- First, we model our prior information on the motion (if any) between images, the high resolution image and the blurring function or functions using prior probability distributions on these unknown variables and formulate the observation process using conditional probabilities. We will analyze and develop new blurring, motion and image models. For the image models, we will distinguish between general models and those based on learning the prior image distribution using examples from similar images.
- Once the Bayesian modelling has been completed, we will approximate the posterior distribution of the unknown variables given the observations using variational approximations of those distributions based on the Kullback-Leibler divergence measure. These approximations will allow us to obtain not only point estimates of the unknown variables in our problem: image, motion (if it exists) and blurring function or functions but also to simulate those distributions in order to analyse, for instance, different possible reconstructions of the objects of interest in the images.
The super-resolution methods to be developed will be integrated in a software application to be downloaded from the project web site which can be used to improve the quality of the observed images.
- Introduce, into the modeling of the super-resolution from low resolution image sequences, the blurring function of the procedure, a function that is unknown in real problems.
- Study image models, both general and aimed at improving the images of concrete objects, e.g. faces, license plates, text, cars, military vehicles, aircraft, ships and installations that appear in images of surveillance and security.
- Jointly address the estimation of the image, the displacement vectors and the blurring function or functions under the Bayesian problem formulation.
- Get into the estimation process, in addition to point estimates of the image, motion and blurring function, probability distributions of these estimators that allow their simulation.
- Deploy in a friendly graphical user interface based on Matlab® all algorithms that are developed in the project so they can be used by researchers and companies. Apply the results to real images provided by the companies interested in the project.
The project has produced results in different aspects:
- Modelling the image super-resolution using Bayesian models and inference. Those models and inference has been scientifically contrasted in the project publications.
- A software application implementing the best SR methods developed for the project has been placed at disposal for the scientific and reserach comunity.
- This website has been developed to disseminate the results of the project.