Data compression and the 8×8 integer transform

Data compression is a kind of power concentrating process. Complex data could be transferred into eigenvalues and sparse data concentrated directly. DCT is a very efficient transform but hard to implement. 8×8 integer transform derived here is easily mapping into a parallel accelerating system, and proved to earn a better performance than existing 4×4 ones. This work implemented 8×8 integer algorithm on the REmus system, a novel reconfigurable system designed by Tsinghua, verified the architecture with several media applications. A 46.0% speedup comparing with 4×4 integer ones and a 60.1% average boost comparing with other processors are achieved. Different dividing of frames also has an impact on other algorithms, such as deblocking, the most significant process in h264 codec. It gets 62.5% boost, too. Comparing with 4×4 ones, when quantity level is smaller than 10, 8×8 restores better pictures and a 10% improvement is always found on compress ratios.