文献の詳細
| 言語 | 英語 |
|---|---|
| 著者 | Xi Huang |
| 論文名 | A Study on Video Superresolution Using Reference Frames |
| 年月 | 2019年8月 |
| 要約 | In this paper, we proposed a reference-based super resolution method to tackle the super resolution task. The existing super resolution methods could be roughly divided into three types: Single Image Super Resolution (SISR), Multi-frame Super Resolution (MFSR) and the Reference-based Super Resolution (Reference-based SR). Recently, SISR and MFSR have been widely studied. However, these SR methods have limited performance due to the loss of the information in the low-resolution images. This limitation will degrade performance when the under-resolved images are at a very low-resolution level. As a solution of this limitation, the proposed reference-based SR method has three advantages: 1) With the support of the high-resolution reference frame, the performance of the SR model is able to be further improved; 2) With only two input images, the calculation cost can be cut off compared with the MFSR methods; 3) With the help of the reference, it allows the proposed model to obtain superior results even when dealing with the very poor quality images, which beneficial to the data reduction. To evaluate the proposed model, we did a lot of experiments and compared our model to the baselines with both quantitative and visual estimations. We experimented under both scaling factors of $\times 2$ and $\times 4$, and found that the proposed method could recover high quality super-resolved images from a very low resolution level. This experiment results showed the superiority of the proposed model on the very poor quality images. In the analysis, we assumed a video compression system and conducted video super resolution experiments using the proposed method. Finally, we found a better fashion for the video super resolution and verified the practicability of this video |
@MastersThesis {Huang2019,
author = {Xi Huang},
title = {A Study on Video Superresolution Using Reference Frames},
year = {2019},
month = aug
}