L-shaped template used for motion search of the current block. Already decoded frame regions are shown in grey. The candidates (black circles) for template matching are derived from motion vectors of two neighbouring blocks using linear scaling.