基於遮蔽及色彩資訊之適應性深度視訊編碼;Adaptive

NCUIR > College of Electrical Engineering & Computer Science > Graduate Institute of Communication Engineering > Electronic Thesis & Dissertation > Item 987654321/65510

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/65510

Title:	基於遮蔽及色彩資訊之適應性深度視訊編碼;Adaptive
Authors:	林函萱;Lin,Han-hsuan
Contributors:	通訊工程學系
Keywords:	3D high-efficiency video coding (3D-HEVC);multi-view video coding (MVC);合成視角;遮蔽;垂直紋理複雜度;rate-distortion optimization (RDO);3D high-efficiency video coding (3D-HEVC);multi-view video coding (MVC);synthesized view;occlusion;vertical texture complexity;rate-distortion optimization
Date:	2014-07-18
Issue Date:	2014-10-15 17:02:20 (UTC+8)
Publisher:	國立中央大學
Abstract:	多視角視訊轉播漸成為趨勢，而自由視角雖可提供多視角供使用者選擇觀賞，但合成視角畫面品質仍有改進空間，此可由3D視訊編碼端提升合成畫面品質。因此，本論文提出針對深度視訊編碼，提升合成視角畫面中高複雜度垂直紋理及遮蔽之部分區域品質，其設計共包含兩個部分。從先前的研究成果得知，合成視角的失真除了受深度圖的品質影響，也和色彩視訊的垂直紋理複雜度有關聯性，因此第一部分參考色彩視訊區塊之垂直紋理複雜度，判斷深度視訊之目前區塊是否為易造成合成視角影像失真區域，以決定適用的深度視訊編碼模式決策之RD cost，此部分採用多視角視訊編碼架構MVC (multi-view video coding)。由於立體視覺影像中，遮蔽區域的深度資訊不確定性會影響合成視角畫面品質，因此在第二部分，本論文利用left right check (LRC)找出左右視角所對應的遮蔽圖，再結合由色彩視訊的重建畫面垂直紋理複雜度資訊，將區塊分成不同等級，對rate-distortion optimization (RDO)之RD cost採取不同的Lagrange參數調整，此部分採用3D-HEVC (3D high-efficiency video coding)編碼架構。由修改3D-HEVC之參考軟體HTM6.0的深度視訊編碼器實驗結果顯示，本論文所使用的演算法在第一部份實驗BDBR平均上升0.25%，而BDPSNR平均下降0.01dB，第二部份的實驗結果顯示平均BDBR少0.22%，合成視角的平均BDPSNR下降約0.01dB，而遮蔽區域且垂直紋理複雜度高之區域其合成視角畫面品質提升。;The tendency of TV service is broadcasting with multi-view videos. Although free-view TV provides optional views to users, there is still the room to improve the quality of color videos synthesized from 3D reconstructed videos. Thus, this thesis proposes adaptive depth video coding to improve the quality of regions which are occluded in the stereopsis and have highly vertical texture in the reconstruct color video. The research is composed of two parts. From the previous research result, we know distortions in synthesized views depend on not only coding distortions in the depth video but also the vertical textures in the reconstructed color video. Consequently, the first part consults the vertical complexity of textures in the reconstructed color video to detect regions with more distortions in the synthesized view. The appropriate RD cost function for depth video coding is selected. The implementation is based on multi-view video coding (MVC). Since the uncertainty of occluded regions in the stereoscopy reduces the quality of synthesized view, this thesis applies left-right check (LRC) based occlusion detection. The information of the occluded map and the vertical texture complexity map is fused to select proper Lagrange parameter in the RD cost function for each largest coding unit (LCU) in the left and right views. The implementation is based on 3D high-efficiency video coding. Our experimental results show that the first part of the proposed scheme has 0.25% BDBR increment and only loss 0.01dB BDPSNR in virtual views compared with the original JMVC 6.0.3. In the experiment of the second part of the proposed scheme, the result shows that the average BDBR decrease is 0.22% and loss of BDPSNR is 0.01dB, with better synthesized viewing quality in the blocks with high vertical complexity and occlusion.
Appears in Collections:	[Graduate Institute of Communication Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	477	View/Open

社群 sharing

Loading...