DC 欄位 值 語言 DC.contributor 通訊工程學系 zh_TW DC.creator 羅國軒 zh_TW DC.creator Kuo-Hsuan Lo en_US dc.date.accessioned 2020-1-17T07:39:07Z dc.date.available 2020-1-17T07:39:07Z dc.date.issued 2020 dc.identifier.uri http://ir.lib.ncu.edu.tw:88/thesis/view_etd.asp?URN=106523003 dc.contributor.department 通訊工程學系 zh_TW DC.description 國立中央大學 zh_TW DC.description National Central University en_US dc.description.abstract 視訊編碼標準為高效率視訊編碼(High Efficiency Video Coding, HEVC)，比H.264/AVC有更佳的編碼效率。HEVC的畫面內預測中，使用了35個模式來增加預測的精確度，但同時也大幅增加其編碼複雜度。因此本篇論文探討卷積神經網絡和約略模式決策所預測的候選模式來跟全模式搜索的最佳模式比較，準確率方面約略模式決策比卷積神經網絡來的高11.48%，而效能方面卷積神經網絡所運行的時間比約略模式決策高了5.109%，而BDBR卻多上升了0.59%。由我們剛才所討論的結果，我們可以知道卷積神經網絡的預測候選模式沒有約略模式決策來的好，但是兩者之間所選的候選模式從紋理方面可以看出是有相關性的，所以接下來我們會使用卷積神經網絡輔助約略模式決策，此處我們會使用卷積神經網絡的模式機制與機率機制來輔助約略模式決策，在模式機制方面，將會用約略模式決策與卷積神經網絡的候選模式做重疊，如果候選模式沒有重疊則會刪除；而在機率機制方面，將會比較候選模式的機率與閥值的大小，如果過小則會刪除候選模式，藉此減少候選模式個數已達節省時間的效果。在只進行8x8編碼的情況下，實驗結果顯示，當使用模式機制，可以在BDBR上升0.014%下，節省9.76%的時間；使用機率機制，可以在BDBR上升0.008%下，節省10.753%的時間。 zh_TW dc.description.abstract High efficiency video coding (HEVC) is the latest video coding standard. To improve predict more accurately, using 35 prediction modes in intra prediction. This process which is meant to improve the efficiency in HEVC intra prediction however leads to a significantly higher computational complexity. In this paper , we discuss candidate mode predicted by CNN and RMD to compare with the best mode for full mode search in terms of accuracy, performance, and texture. We can know that the prediction mode of the CNN is not as good as the RMD, but the candidate mode selected between the RMD and CNN has a correlation, so we will use CNN to assists RMD. Here we will use the CNN′s mode and probability to assist RMD. In terms of the mode, RMD and CNN candidate modes will be overlapped. If the candidate modes do not be overlapped, they will be deleted. In terms of the probability, the probability of the candidate modes will be compared with the threshold value. If value is too small, the candidate modes will be deleted, thereby reducing the number of candidate modes to save time. When only 8x8 encoding is performed, the experimental results show that when the mode is used, BDBR can be increased by 0.014%, saving 9.76% of time; using the probability, BDBR can be increased by 0.008%, which saves 10.753% of time. en_US DC.subject HEVC zh_TW DC.subject 預測單元 zh_TW DC.subject 畫面內預測 zh_TW DC.subject 深度學習 zh_TW DC.subject RDO zh_TW DC.subject RMD zh_TW DC.subject HEVC en_US DC.subject Prediciton Unit en_US DC.subject Intra Prediction en_US DC.subject Deep Learning en_US DC.subject RDO en_US DC.subject RMD en_US DC.title 利用深度學習以降低HEVC模式決策之運算複雜度的研究 zh_TW dc.language.iso zh-TW zh-TW DC.title A CNN-Assisted Technique for Computation Reduction of HEVC Intra prediction en_US DC.type 博碩士論文 zh_TW DC.type thesis en_US DC.publisher National Central University en_US