Fast Video Transcoder Based On Data Mining

Advancement in computer sciences has catalyzed development of multimedia telecommunication industry to an extraordinary pace. Multimedia telecommunication has become indispensible in our daily lives. The enormous amounts of data storage and transition have posted new challenges on compression and transition of multimedia data. The importance of video transcoding technology is gradually being realized, becoming the key to multimedia telecommunication. Video transcoding enables video conversion in terms of bit rates, spatial resolution, frame rates and video format, in order to achieve seamless connectivity between various network, terminal devices, as well as different compressed data format. With the current boom in mobile Internet sector, video transcoding will play an increasingly crucial role and its future is promising.In this paper, three classical video transcoding architectures, namely cascade, open-loop transcoder and close-loop transcoder was analyzed. Full-decode-full-encode classical pixel field close-loop transcoder was chosen based on application requirement. Since the input and output of the transcoder is merely a different compression method of the very same video, the two are closely related, and enables us to employ data mining in the field of video transcoding to analyze the connection between coded message of input and output code streams. The transcoder developed in this paper substantially accelerated the process of transcoding to cater for practical needs, through extracting information from decoding stage to boost the most time consuming macro block decision stage-the re-encoding stage. Based on the algorithms, this paper enclosed two transcoders:iso-resolution and reduced-resolution transcoder from MPEG-2to H.264and H.263to H.264. Test results have demonstrated these transcoders are capable of improving transcoding time by at least30%, with less than0.5reduction in PSNR.