| Summary: | 碩士 === 國立中央大學 === 通訊工程學系 === 107 === 360-degree videos provide users immersive visual experiences. Since most video encoders take two-dimensional rectangular images as inputs, the three-dimensional sphere domain data of 360-degree videos must be projected onto the two-dimensional image plane before video coding. The eqirectangular projection (ERP) and cubemap projection (CMP) are the most commonly used 2D projection formats of 360-degree videos, where the cubemap projection enables better coding performance because of it has smaller geometric distortions. Since currently there are not any bit allocation schemes proposed for video coding of the cubemap projection, this thesis proposes a bit allocation scheme that consists of two parts for video coding of the CMP. First, high coding cost largest coding units (LCUs) of six faces of the CMP are detected using support vector machine (SVM), referring to the texture complexity, motion vector magnitude, motion density, and temporal variance of motion. Second, bit allocation between high coding cost LCUs and non-high coding cost LCUs is applied by functions attained by surface fitting of coding statistics based on HEVC/H.265. Experimental results show that our proposed performance is better than HM16.16 with R-λ model. with 2.256% BDBR decrease and 0.13 dB BDPSNR increase.
|
|---|
