Video Super-Resolution Using High Quality Photographs

Abstract

This paper introduces a technique that increases the spatial resolution of a given video. The method is built on the fundamentals of super-resolution techniques that aim to reconstruct high-resolution frames from a low-resolution input sequence. Different than classical super-resolution methods, besides using the information of adjacent frames, we take advantage of several reference high quality (resolution) photographs of the same scene. The method is purely image-based, and does not require depth estimation. The additional information extracted from the reference photographs is used to construct several high resolution seed frames added with a constant step in the initial video sequence. Therefore, the seed frames but also the adjacent low-resolution frames provide important information to define priors that are considered in the probabilistic interpretation of the generative model. The estimated solution is obtained based on a standard maximum a posteriori (MAP) approach. Objective tests on real and synthetic video sequences demonstrate the utility and the benefits of the proposed technique over related methods.