Video Super-Resolution Based on Multiple Networks Merging

2018 4th International Conference on Environmental Science and Material Application

生态环境科学;材料科学

Shao, Weiye^1 ; Xu, Zhihai^1 ; Feng, Huajun^1 ; Li, Qi^1

School of Zhejiang University, Hangzhou

310027, China^1

关键词: High definition video;    High resolution image;    Image super resolutions;    Low resolution video;    Multiple networks;    Superresolution methods;    Temporal relation;    Video super-resolution;   

Others  :  https://iopscience.iop.org/article/10.1088/1755-1315/252/4/042121/pdf
DOI  :  10.1088/1755-1315/252/4/042121

来源: IOP

【 摘 要 】

Video or image super-resolution technology is designed to recovery a high-resolution image from a low-resolution video or image. In recent years, deep neural networks have developed rapidly and have been applied in many digtal image processing tasks. In this paper, we choose an optical flow network to effectively exploit temporal relation within multiple consecutive video frames. In addition, we propose a weight distribution network which gives weight images of different high-resolution images obtained by various super-resolution network methods. This architecture combines advantages of different methods and provides a more accurate high-resolution image. We build a dataset with high-definition video, and use this dataset to train and test our networks. We compare our algorithm with other super-resolution methods and show that it performs a state-of-the-art results.