Research On Compression Techniques Oriented To Sign Language Based News Broadcasting System

People with hearing disability have obstacle in obtaining information, especially at news which is extremely real-time service. Therefore helping them with this issue using computer techniques becomes a very challenging research topic and also a very useful application.In this thesis, a computer graphics based sign language synthesis system using virtual human is designed as a new media oriented to people with hearing disability. Moreover, news broadcasting can be achieved by applying the system. Then redundant compression algorithms are proposed to optimize the throughput and usability of the system.First, in order to meet the requirement of letting the people with hearing disability to watch the news, a sign language synthesis system and a network-based sign language news broadcasting system is designed and implemented. Then further studies focus on the issue of enhancing the system's throughput quality using compression algorithms. Compression is done to different data involved in the system, including 3D mesh, sign language action driving data file, texture data and 2D sign language video.For 3D mesh, this thesis proposes a novel geometry sequence based progressive mesh compression scheme. It converts the series of progressive meshes into geometry images, which is line-up together to form a geometry sequence description of the original mesh. Video CODECs are used to compress the geometry sequence. Both objective and subjective reconstruction quality are achieved; also the usability and robustness of the compressed bit-streams are highly improved. For sign language action driving data file, this thesis introduced a disparity compensated Lempel-Ziv Markov-chain algorithm, which is applied to reduce the energy of the action script file. The algorithm yields significant performance gain compared to conventional compression software regarding on text compression. For texture, JPEG 2000 is applied to provide progressiveness to the compressed texture bit-stream. Moreover, an adaptive bit-plane based wavelet coefficient modeling scheme is applied to help better adapting to individual texture data. Performance over conventional JPEG 2000 CODECs have been gained by using the proposed scheme. For sign language videos, H.264/AVC is applied as the compression standard. Especially, intra frame prediction algorithm in H.264/AVC is analyzed, two improvement schemes over the standard intra frame prediction is proposed. The first one is block-based intra frame prediction scheme. It explores spatial redundancy in a single frame to help improve the objective and subjective of the reconstructed video frames. The other one is fast intra prediction mode selection scheme using mode fitness. It considers the energy distribution of the direction of the prediction to reduce the number of potential candidate prediction modes, which speeds up the process of intra prediction significantly while maintaining the compression performance. Conclusion is made and possible future work is also discussed at the end of the thesis.