Implementation and Optimization of an Enhanced PWD Metric for H.264/AVC on a TMS320C64 DSP

Abstract

A common method for selecting the best prediction mode based on block matching algorithm is to compare, for each source block, the associated distortions among the available prediction candidates. The human visual perception is sensitive to luminance contrast rather than absolute luminance values. In fact, the human eyes ability to detect the magnitude difference between an object and its background depends on the background luminance average value. The Perceptually Weighted Distortion (PWD) is a new distortion measure that can produce better image quality. In this paper, we propose to add a new feature to the PWD by introducing another diagonal component that yields to a significant quality improvement. The enhanced PWD metric actually outperforms the original PWD and the SAD metric, according to the experimental results, especially in the aspect of reducing block artifacts. An increase in terms of implementation complexity will be noticed as a result of this contribution. Therefore, optimized implementation of the Enhanced PWD exploiting the C64 DSP-Core assets will be presented. In fact, Standard Assembly (SA) is used to implement the different Enhanced PWD functions in order to exploit efficiently the C64 internal architecture and resources. Experimental results show more than 85% improvement in terms of cycle cost compared to C code.