以DenseNet / Denoising Autoencoder演算法則應用於HEVC和VVC畫面內解碼端後處理研究;Post-Processing for HEVC and VVC Intra Prediction With DenseNet / Denoising Autoencoder Algorithms

NCU Institutional Repository > 資訊電機學院 > 通訊工程研究所 > 博碩士論文 > Item 987654321/92829

請使用永久網址來引用或連結此文件: http://ir.lib.ncu.edu.tw/handle/987654321/92829

題名:	以DenseNet / Denoising Autoencoder演算法則應用於HEVC和VVC畫面內解碼端後處理研究;Post-Processing for HEVC and VVC Intra Prediction With DenseNet / Denoising Autoencoder Algorithms
作者:	陳彥均;Chen, Yen-Chun
貢獻者:	通訊工程學系
關鍵詞:	高效率視頻編碼;多功能視訊編碼;影像後處理;HEVC;VVC;Image post-processing
日期:	2023-01-16
上傳時間:	2024-09-19 16:21:40 (UTC+8)
出版者:	國立中央大學
摘要:	現今的科技發展日新月異，人們的生活跟科技已經息息相關，而高解析度的影像也成為多數人的日常需求，舉凡人們在看電影或者打電動時都希望追求更高畫質的影像品質，因此在這些高解析度的影像品質背後，所花費的資料量無庸置疑是很龐大的，為了更有效的因應這些高解析度的影像，HEVC以及VVC都運用了許多方式可以有效的降低位元傳輸。 HEVC(High Efficiency Video Coding)採用四分樹(QuardTree)的編碼劃分架構，在編碼端會造成影像的失真，所以本論文採用三種在畫面內預測後處理的方式來增強影像的品質，第一種藉由Add Fusion替代Concatenate Fusion的方式去比較兩者的之間的差異，第二種是透過DenseNet的架構讓影像可以還原得更好，第三種是在模型的最後面再加上Denoising Autoencoder讓整體性能有更進一步的提升。最終整體架構在HEVC畫面內預測後處理可以提升0.35(dB)的BDPSNR與降低6.37(%)的BDBR。 VVC(Versatile Video Coding)相比HEVC則不僅是採用四分樹的劃分方式，還多了多類型(Multi-Type)的編碼劃分架構，相較於前一代的HEVC可以提供4K~16K高解析度的影像，讓使用者可以有更好的視覺體驗。本論文一樣採用上一段所提到的模型架構在VVC進行增強影像，最終整體架構在VVC畫面內預測後處理可以提升0.288(dB)的BDPSNR與降低5.25(%)的BDBR。 ;Nowadays, technology is developing rapidly, people′s lives are closely related to technology, and high resolution images have become the daily needs of most people. For example, people want to pursue higher image quality when watching movies or playing video games. However, behind the quality of these high resolution images, the amount of data consumed is undoubtedly huge. In order to respond to these high-resolution images more effectively, both HEVC and VVC use many ways to effectively reduce transmission bit. HEVC(High Efficiency Video Coding) adopts a quadtree encoding partition structure, which will cause image distortion at the encoding end. Therefore, this paper uses three post-processing methods of intra-screen prediction to enhance image quality. The first method uses Add Fusion instead of Concatenate Fusion and compare the differences. The second is to reconstruct the image better through the DenseNet architecture. The third is to add the Denoising Autoencoder at the end of the model to improve the overall performance. In the end, the post-processing for HEVC intra prediction can improve the BDPSNR by 0.35 (dB) and reduce the BDBR by 6.37 (%). Compared with HEVC, VVC(Versatile Video Coding) not only adopts the quad-tree division method, but also has a multi-type (Multi-Type) encoding division architecture. It can support 4K~16K high resolution images, to provide better experiences of media consuming. This paper also uses the model architecture mentioned in the previous paragraph to improve and enhance the image in VVC. Final, the post-processing for VVC intra prediction can improve the BDPSNR by 0.288 (dB) and reduce the BDBR by 5.25 (%).
顯示於類別:	[通訊工程研究所] 博碩士論文

文件中的檔案:

檔案	描述	大小	格式	瀏覽次數
index.html		0Kb	HTML	9	檢視/開啟

在NCUIR中所有的資料項目都受到原著作權保護.

社群 sharing

資料載入中.....