各項科技隨著時間蓬勃發展，現代人的生活與科技有密不可分的關係，而在多媒體方面也不例外，畫質不斷地提高、色彩也著墨不少，但是在這些高解析度影像背後，需要龐大的資料量，為了有效的壓縮高解析度影像的巨大資料量，HEVC(High Efficiency Video Coding)運用的許多方式有效的降低位元傳輸。 近期已經有許多研究將深度學習應用於HEVC後處理中，而這些研究的目的是要解決因為壓縮的過程中，在編碼端會造成影像的失真，本篇論文也是以增強影像畫質為目的，然而許多後處理方式仍然是著重在單張影像內的資訊，並沒有考慮到畫面與畫面之間其實有很大的相關性。在此篇論文中，我們提出SEResNeXt運算法的時間序列模型在HEVC解碼端增進影像品質，藉由觀察畫面之間的的特性以及利用時間序列的模型，在連續幀之間的相似性中提升模型性能。在我們結合整體架構後，最終在HEVC畫面間預測與參考程式HM16.0相比，可以達到BDBR減少7.084%，在BDPSNR增加0.244dB。 ;Technologies have flourished over time. Modern people’s live are closely related to technology, and multimedia is no exception. The resolution of pictures is much higher than before, but amount of data is required behind these. In order to effectively compress the huge amount of data, HEVC (High Efficiency Video Coding) are used to effectively reduce transmission bit. Recently, there are many researches applying deep learning to HEVC post-processing, and the purpose of these researches is to solve the problem of image distortion caused by the encoding side during the compression process. This paper is also aimed at enhancing image quality. However, there are many post-processing algorithms which still focus on the information in a single image. Thus they fail to take the advantage of the inter-frame correlation in the video. In this paper, we propose the post-processing for HEVC inter prediction with SEResNeXt algorithm. By observing the characteristics between the images and using the time series model, the performance of the model is improved. After we combine the overall architecture, the experiment result of our algorithm compared with the reference program HM16.0 achieves up to 7.084% BDBR reduction and 0.244dB BDPSNR increase.
