Temporal Coding

The goal of temporal coding is to remove the temporal redundancies among frames. The basic unit is a macro block of 16x16 pixels. The process consists in searching the neighboring macro blocks of the given macro block in the different subsequent frames to find the location of the current macro block in these frames. If a good match (some times approximated when privileging the compression rate to the quality) is found, then instead of encoding this macro block in the current frame and all the frames containing a good match of it, simply a motion vector indicating the location of that block is encoded for each matching frame. Search windows can be any size. However, the larger the window size is the better the motion estimate, but the higher the computational cost and hence longer encoding delay (too much delay is not acceptable for real-time applications). This process reduces considerably the bit rate. This is because most of natural video scenes have many redundant parts spread into several frames. The situation where the redundant blocks are very small occurs when there is scene cuts (the picture changes abruptly and is replaced by another one). In addition, it occurs when the scene contains too much motion with motion frequency is very large compared to the encoding frame rate. There are some approximations of that step depending on the compression factor required, the tolerated encoding latency, etc.