Continuous Digital Zoom with Cross Attention for Dual Camera System
Yang Yifan,Li Qi,Xu Zhihai,Feng Huajun,Chen Yueting
DOI: https://doi.org/10.1007/s11042-021-11688-0
IF: 2.577
2021-01-01
Multimedia Tools and Applications
Abstract:Reference-based super-resolution (RefSR) aims to recover realistic textures, when a reference (Ref) image and low-resolution (LR) image are given. Because the Ref images are selected randomly, the quality of RefSR will degrade when the Ref image has less similar content with LR input. In this article,we propose a dual camera system to unleash the potential of RefSR. Additionally, we presents a cross attention mechanism to realize a high-quality digital zoom by using two camera modules with different focal lengths. In dual camera system, shorter focal length module produces the wide-view image with the low resolution. On the other hand, the longer focal length module produces the tele-view image via optical zoom. The long-focal image contains more details than short-focal image and can be used to guide short-focal image to reconstruct high-frequency part. Since the two images are taken from the same scene, we can get better image matching correlation in dual camera system. Inspired by the recent work on reference-based image super-resolution (RefSR), we propose a cross attention mechanism to fuse two images with different focal length and generate more feature correlations within them by texture transferring. Besides, we use segmentation information to improve match accuracy. Instead of using a direct matching between different images, the attention module fully utilizes texture of different levels. Additionally, we present a feature restoration module to reconstruct more image details. Extensive experiments show that Our method achieves state-of-the-art results both quantitatively and qualitatively across different datasets.