Fast Mode Decision Based on Activity Segmentation in H.264/AVC Encoding

The recent H.264 video coding standard has been developed to increase video coding efficiency, achieving better Rate-Distortion results than any other standard. This is mainly due to the new features included at the motion estimation process, as tree structured motion compensation and multiple reference frames. However, the motion estimation stage is the most expensive part of a video encoder, and these improvements greatly intensify this computational load. To solve this situation, in this paper we propose a fast mode decision scheme, that analyses the activity between frames and classify the picture into active and non-active areas. This classification is performed with and adaptive thresholding technique that minimizes the area set as active and the activity energy of the area set as non-active. For each macroblock, a sub-set of inter-prediction alternatives is selected according to the amount of active and non-active area inside it. Results show motion estimation time reduction of 38% to 51% while achieving Rate-Distortion values very similar to the ones obtained with the complete analysis of inter-prediction alternatives.