Multimedia compression algorithms usually take DCT or wavelet transformed input data and produce a stream of bits as output. It has been shown that compression algorithms can be parallelized in a data driven way so that a sequence of bit-stream parts is produced by each processing element (PE) corresponding to the PE-local part of the partitioned and distributed input data. Unfortunately, the collection and assembly of these bit-stream parts by a single PE turns out to be a major bottleneck because it is sequential. This work addresses this problem by sequential optimizations and parallelization of the assembly process itself.
Special Issue Papers