Adaptive global motion temporal prediction for video coding

摘要: Depending on the content of a video sequence and settings used for encoding it, amount bits spent transmission motion vector information can be enormous in some cases even take largest fraction bit rate. This is not always necessary since often wide areas, i.e. background or large foreground regions, fit same global motion. Additionally, model using sophisticated interpolation techniques better representation movement these regions than that has only quarter-pel accuracy. true especially if scaling, rotation perspective transformation occur. paper presents novel prediction technique based compensation temporal filtering previously decoded pictures. The new approach incorporated into an H.264/AVC reference software. encoder outperforms by up to 14%.