Study For The Key Techniques In Mpeg Video Real-Time Transcoding

Video compression coding technique is one of key techniques for multimedia applications. Because of great capacity of video imformation efficient compression must be done. International Standard Organizations such as ISO/IEC and ITU etc. have been successful to release a series of video coding standards aiming at different application realms based on current video coding techniques.With the rapid development of multimedia communications, terminal users hope to enjoy real-time video services conveniently. At that time they can exchange and operate on video data freely. However, at present many video contents have been encoded with different standards and techniques. What's more, the conditions of different networks and the processing abilities of different terminals are so distinct. Direct Transmition of these video contents would lead to mismatching of video coding bit stream with channels or terminals. Therefore, it is highly desirable to implement real-time and efficient video transcoding.In this dissertation, the main research objective focuses on the real-time transcoding from MPEG-2 to MPEG-4. The key techniques in the real-time transcoder of MPEG-2 to MPEG-4 are discussed and some significant results have been also obtained.The reason why MPEG-2 DCT domain texture data of I frame could not be directly used by MPEG-4 quantization is demonstrated. A quick compensation method for DCT domain texture data is proposed and it is also validated by the simulation results.The theories of field DCT and frame DCT in MPEG-2 are analyzed in detail and a quick conversion method of field DCT to frame DCT is proposed. The simulation results show that it can implement the conversion of field DCT to frame DCT efficiently.Without considering resolution down sampling, a detailed method of motion vector mapping is put forward aiming at conversion of frame types in transcoding. The simulation results indicate that video quality of transcoding is lower than the case using direct MPEG-4 coding at the same target bit rate.The theories of field prediction and frame prediction in MPEG-2 have been analyzed briefly. A method based on directly discarding the top field motion vectors is adopted to implement the quick conversion from field prediction to frame prediction in macroblock. The simulation results show that acceptable transcoding quality can be obtained under the condictions of slower motion in the video.At last, DCT domain texture down sampling method based on filtering in [16] is deduced and the results show the good effect of transcoding makig use of this method.