In generative adversarial network based single image superresolution, most of the networks are using residual-in-residual dense block. However, it cannot improve the content and perceptual quality simultaneously. To improve upon, this paper presents a two-stage internal prior generative adversarial network which improves both reconstruction accuracy and visual quality in the superresolved image. The first stage uses an optimized block for image content, whereas the second stage uses an optimized block for visual content. The first stage is further improved by using attention block and fractal connections to leverage prior information in the network. The experimental results highlight the effectiveness of the proposed network in terms of the peak signal-to-noise ratio, structural similarity index measure, and learned perceptual image patch similarity and demonstrate its superiority over other state-of-the-art models.
Singla et al. (Mon,) studied this question.