Design and Implementation of a Fast DVC-SVC Transcoding System in Cloud Computing

• Main Authors: Kang-Chih Chang, 張剛止
• Other Authors: Chang-Ming Lee
• Format: Others
• Language: zh-TW
• Published: 2013
• Online Access: http://ndltd.ncl.edu.tw/handle/37358693423373254526

碩士 === 國立中正大學 === 通訊工程研究所 === 101 === With the development of Internet and personal mobile devices, the multimedia demands of users become various and vast. Therefore, the traditional single coding is impractical. Fortunately, we can use cloud computing, which can provide dynamic and massive computing resources, to effectively improve the efficiency of existing video coding systems. Usually, users need to transcode the image sequences to fit different video specifications while capabilities of their personal devices are limited. Designs of a low-complexity, low-cost, and low-power coding/decoding system are considerable while maintaining a good coding efficiency. Based on the Wyner-Ziv theorem, the concept of Distributed Video Coding can transfer the computational complexity from the encoder side into the decoder side to meet the low complexity demand in the user side. Moreover, to broadcast the video to other users in different scenarios, we design an architecture to transcode the DVC-coded data to the SVC stream. Since the high-complexity components in DVC decoder and SVC encoder are centralized in the transcoder, we provide a cloud-based solution with the file allocation strategy that can improve the coding efficiency according to the characteristics of the cloud computing. In this thesis, the temporal and spatial scalabilities are considered to reuse the MV information (from the DVC decoder) to increase the throughput of SVC encoding instead of the full motion estimation/compensation. Finally, these strategies not only can save the cloud resources, but also can provide large-scale and high-fidelity quality of service. The proposed fast DVC-SVC transcoding system can save 396.03% coding time in average compared with traditional single computer coding. The transcoding throughput can be 4.08 frame per second and can be improved in the future.