The realization of fast gradient adjusted prediction method for lossless video compression with low memory requirement
Yun Hua, Bo Hu · 2004
Gradient adjusted prediction (GAP) is an efficient prediction method for lossless compression of video. By calculating its gradient in latitudinal and longitudinal orientation with the known context, the pixel value can be well estimated and used for predictive coding. However, when realizing this method in a real-time compression system, it needs a large amount of memory space. The number of operations for each pixel is also large. In this paper, we introduce an optimized algorithm to realize GAP in a digital signal processor (DSP) named Analog Device Blackfin/spl trade/ DSP BF533. The GAP calculation is divided into two parts: gradient calculation and prediction calculation, and carefully designed to utilize the direct memory access (DMA) operation and parallel instruction provided by BF533. Worthless repetition and stalls in the DSP pipeline are avoided.