Fast mode decision on H.264/AVC baseline profile for real-time performance

Abstract

In this paper a new fast mode decision (FMD) algorithm is proposed for the recent H.264/AVC video coding standard, aiming to reduce its computational load without loosing coding efficiency. This algorithm identifies redundancy and selects the minimum sub-set of modes for each macroblock (MB) required to provide high rate-distortion (RD) efficiency. It is based on a fast analysis of the histogram of the difference image between frames which classifies the areas of each frame as active or non-active by means of an adaptive thresholding technique. More coding effort is devoted to active areas with the selection of a large sub-set of Modes, as these areas are expected to be the most relevant in terms of RD cost. Results show reduction values around 35-65% of motion estimation (ME) time, preserving the RD cost for the Baseline Profile, by using PSlices and without needing B-Slices. Moreover, the strategy works as an intelligent tool for real-time applications with constrained number of operations per frame: it wisely uses the given operational resources distributing them among those MBs that need it.