Learning wavelet-transform based sparsity knowledge for blind infrared images restoration in students' training in the wild

Abstract

This paper proposes a novel image restoration model with wavelet-transform-based regularization for recovering high-resolution infrared images in students' training in the wild. The degraded infrared images are often contaminated by random Gaussian noises. The proposed method integrates the wavelet-transform-based prior, togethers with the adaptive L2 norm constraint into the image reconstruction framework. The major novelty of this study is that wavelet-transform regularization is proposed for infrared blindly deconvolving images, which can save the image structures effectively. The famous alternation minimization algorithm is introduced to estimate the latent infrared images and instrument function. Experimental results demonstrate that the proposed BRWTR method can remove the random noise as well as produce the smooth results simultaneously. The qualitative comparison on the real infrared images with the state-of-the-art methods show the good performance of the proposed method.