Efficient Algorithm and System Architecture for the Suppression of MPEG Artifacts
Michael Smed Kristensen, Soren Birk Sorensen, Yannick Le Moullec, Peter M. Koch · 2006
This paper presents i) the development of an algorithm that enhances the visual quality of MPEG compressed image sequences and ii) the design of a system architecture for implementation of the algorithm in a battery powered handheld device. In the spatial domain a combination of classification, Kalman smoothing and sharpening filtering has been used to suppress blocking, ringing and blurring artifacts. In the temporal domain a Kalman filter in conjunction with a motion detection method has been used to suppress temporal flicker and Mosquito noise. Furthermore, a system architecture has been designed at two levels of granularity. At the coarse level, the SpecC methodology has been used to determine an overall system architecture consisting of a GPP and a hardware accelerator. At the fine level, time constrained scheduling has been used to find a suitable architecture for the accelerator data-path. Several optimization techniques have been applied to the accelerator in order to minimize the area consumption. A subjective and quantitative evaluation shows that the algorithm improves the visual quality of the compressed image sequences. The final architecture of the accelerator consists of 6 register files, 1 register, 3 multipliers, 2 adders, 1 subtracter, 1 divider, 1 comparator and 1 magnitude calculator.