Video Shot Boundary Detection And H.264 To MPEG-4 Video Transcoding

With rapid development of information technology, video collection becomes easier, and massive video data and all kinds of video application show up. People put forward higher requirements for video retrieval and transmission. Video shot boundary detection is the basis of video analysis and retrieval. Video transcoding is the critical technique for video transmission with network adaptation.Video shot boundary detection and H.264 to MPEG-4 video transcoding are studied and implemented in this paper.1. Video shot boundary detectionBased on the analysis of present main methods of shot boundary detection, this paper concludes three points of shot boundary detection performance affection factors: visual content description, picture sequence context construction, pattern classification and recognition. Considering feature description, detection strategy of CUT and GT, and detection system framework, a real-time shot boundary detection algorithm based on multiple-level feature description and support vector machine is proposed. In order to evaluate the performance of the algorithm, we took part in TRECVID2007 SBD evaluation. The results show that detection speed of the algorithm is much faster than real-time, average precision and recall of CUT detection is over 96%, and GT detection needs to be improved.2. H.264 to MPEG-4 video transcodingAn effective cascade pixel domain transcoding architechture is proposed for H.264 to MPEG-4 video transcoding in this paper, based on the analysis of similarity and difference between H.264 and MPEG-4 and popular video transcoding architechtures. Considering transcoding functional needs, following key technology modules are implemented and improved:(1) An improved rate control algorithm with GOP / frame / macroblock three layers control based on linear sourceρdomain model is proposed, considering the characteristic of H.264 to MPEG-4 video transcoding and input bit stream coding type. Experiment results show that, compared with "single pass" rate control in Xvid encoder, output bit rate produced by the proposed algorithm is closer to target bit rate, bit rate curve is smoother, and PSNR of output bit stream is closer.(2) This paper also proposes methods of macroblock type conversion and motion vector mapping by statistical analysis. Experiment results show that compared with complete cascade pixel domain transcoding method, PSNR are both less than 0.5dB lower, and transcoding rate are promoted over 30% and about 20% respectively in no spatial resolution reduction transcoding and 2:1 spatial resolution reduction transcoding from H.264 to MPEG-4.(3) For transcoding with arbitrary ratio spatial resolution down-sampling, an arbitrary ratio spatial resolution down-sampling filter with 8 tap coefficients is designed, and weighted average method of macroblock coverage area is adopted for macroblock type conversion and motion vector mapping in this paper. Experiment results show that compared with complete cascade pixel domain transcoding method, PSNR is about 1dB lower in most cases.