Infrared Image Super-resolution by Using Dual-dictionary

This paper addresses the problem of using a single low-resolution infrared image to generate a highresolution infrared image. In this paper, we approach a problem from the point of dictionary-training. The traditional dictionary-training model has some problems. First, the image features, which the traditional fllters extract, are not enough for representing images. Second, traditional super-resolution model hardly reconstructs accurate high-frequency component from the prior information trained by the training images. In order to solve the flrst problem mentioned above, this paper proposes additional fllters to extract image features. The fllters can extract more meaningful image features from training images than traditional fllters. To solve the second problem, a dual-dictionary training model is proposed. The dictionary-training model, which consists of the main-dictionary training and the residual-dictionary training, aims to reconstruct the high-frequency component. In this paper, the image high-frequency component, which is considered as a combination of main high-frequency component and the residual high-frequency component, can be accurately recovered via the dual-dictionary training model. Extensive experimental results demonstrate that the proposed method achieves much better results in terms of both PSNR and visual perception compared with the conventional super-resolution method. In Section 2, the algorithm of sparse representation super-resolution will be introduced. Section 3, the dual-dictionary training modeling is proposed. Section 4 presents experimental results.