MMSR: Multiple-Model Learned Image Super-Resolution Benefiting from Class-Specific Image Priors

Assuming a known degradation model, the performance of a learned image super-resolution (SR) model depends on how well the variety of image characteristics within the training set matches those in the test set. As a result, the performance of an SR model varies noticeably from image to image over a test set depending on whether characteristics of specific images are similar to those in the training set or not. Hence, in general, a single SR model cannot generalize well enough for all types of image content. In this work, we show that training multiple SR models for different classes of images (e.g., for text, texture, etc.) to exploit class-specific image priors and employing a post-processing network that learns how to best fuse the outputs produced by these multiple SR models surpasses the performance of state-of-the-art generic SR models. Experimental results clearly demonstrate that the proposed multiple-model SR (MMSR) approach significantly outperforms a single pre-trained state-of-the-art SR model both quantitatively and visually. It even exceeds the performance of the best single class-specific SR model trained on similar text or texture images.

1. INTRODUCTION

Image super-resolution (SR) aims to generate the latent high-resolution (HR) images from low-resolution (LR) ones by generating information that does not exist in the LR images [1], [2]. Single-image SR (SISR) by learning from examples dates back to the seminal works of Freeman [3], [4], which implicitly learned features from a database of HR and LR image patch pairs. These example natural image patch pairs were employed as data-driven priors to extrapolate high frequency image details. Example-based SISR has later been applied to target specific classes of images, e.g., face images [5], [6]. These class-based SISR algorithms have yielded superior results, because they were able to better capture class-specific image priors.