Single Image Super-Resolution Based on Capsule Neural Networks

• George Corrêa de Araújo UNICAMP
• Artur Jordão USP
• Helio Pedrini UNICAMP

Resumo

Single image super-resolution (SISR) consists of obtaining one high-resolution version of a low-resolution image by increasing the number of pixels per unit area. This method has been actively investigated by the research community, due to the wide variety of problems ranging from real-world surveillance to aerial and satellite imaging. Most of the improvements in SISR come from convolutional networks, in which approaches often focus on the deeper and wider architectural paradigm. In this work, we decided to step up from the traditional convolutions and adopt the concept of capsules. Since their overwhelming results in image classification and segmentation problems, we question how suitable they are for SISR. We also verify that different solutions share similar configurations, and argue that this trend leads to fewer explorations of network designs. Throughout our experiments, we check various strategies to improve results, ranging from new and different loss functions to changes in the capsule layers. Our network achieved positive and promising results with fewer convolutional-based layers, showing that capsules might be a concept worth applying to the image super-resolution problem. In particular, we observe that the proposed method recreates the connection between the different characters more precisely, thus demonstrating the potential of capsules in super-resolution problems.