Single image super-resolution reconstruction

Several deep learning-based models for single image super-resolution (SISR) have been presented in recent years that achieve superior performance compared to the traditional vision-based approaches. However, the performance of each \ac{SR} model varies for different image content. For example, one model may outperform the others on an image with high and intricate textures, while another may perform better in reconstructing images of structured scenes. To utilize the best capability of each model for SR reconstruction, we propose a system that uses the image content to select the most suitable model for SR reconstruction by taking advantage of each image's dominant content. The proposed system is tested on the standard benchmark datasets for SISR. The obtained results indicate that the proposed system delivers comparable results with the state-of-the-art methods in terms of the overall quantitative metrics and visual quality, and confirmed that it delivers better results for scene specific content.