Fast implementation of H.264 4x4 intra prediction

Abstract

This letter proposes a fast parallel architecture and redundancy reduction algorithm for H.264/AVC intra4x4 prediction to speed up intra frame coding. A significant reduction in execution time is achieved without losing video quality. Only 204 cycles are required to process a macroblock (MB). Compared with the dedicated intra prediction [1], processing speed is enhanced by 79%.

Fast H.264 intra prediction for network video processing

“…The decision-algorithm, responsible for selecting that prediction mode which is believed to result in the smallest bitstream, is not specified by the standard [2]. Though more complex algorithms have been developed in the past [5,7], for real-time implementation on dedicated hardware the brute force approach followed by summing of the residuals (Sum of Absolute Differences, SAD) to make the decision is popular [4].…”

“…Due to dependencies, simply changing the processing order, doesn't allow to fill a pipeline completely without skipping modes, hence the hardware is used inefficiently [4]. In [4], a solution is presented which prevents bubbles in the pipeline by clever disassembly, regrouping and scheduling of the required calculations.…”

“…In [4], a solution is presented which prevents bubbles in the pipeline by clever disassembly, regrouping and scheduling of the required calculations. In this work we propose to keep the pipeline filled by processing other streams while waiting for our data to be processed and thus, dependencies to be resolved.…”

“…In [4] an overview is presented of previously developed techniques to speed up the intra prediction process, by pipelining calculations, exploiting parallelism, changing the order in which blocks are processed [5] and even skipping prediction modes [5]. Furthermore, [4] presents a pipeline which supports the same modes as in this work, though no performance is mentioned.…”

“…Furthermore, [4] presents a pipeline which supports the same modes as in this work, though no performance is mentioned. Other solutions reduce the required number of computations by using an algorithm to determine and skip less likely modes.…”