Super-resolution reconstruction of thermal infrared images

Abstract

In this paper a high-resolution (HR) thermal infrared image is reconstructed from a sequence of subpixel shifted, aliased low-resolution (LR) frames, by means of a stochastic regularized super-resolution (SR) method. The Huber (H) cost function is employed to measure the difference between the projected estimate of the HR image and each LR frame. The bilateral Total Variation (TV) regularization is incorporated as a priori knowledge about the solution. The proposed HTV super-resolution approach that employs the Huber norm in combination with the bilateral TV regularization exhibits superior performance to former SR method. Thus the effect of outliers is significantly reduced and the high-frequency edge structures of the reconstructed HR thermal infrared image are preserved. The proposed technique is also tested on frames that are corrupted by Gaussian noise and proves superior when compared to existing regularized SR method.