Decimated sliding windows for multi-reference prediction in video coding

摘要

Decimated sliding windows for multi-reference prediction are disclosed. Reference frames for coding using motion prediction are stored in two or more buffers, which can be arranged as first-in-first-out stacks. As new reference frames are added to a first buffer, the least recently used reference frame is either moved to a second buffer or discarded depending upon a decimation factor. More than two buffers can be used, and each of the buffers can have the same or a different decimation factor. This permits both short duration and long duration availability of reference frames while minimizing memory usage and overhead in the encoded video bitstream.

主权项

1. A method for encoding a video stream having a plurality of frames, comprising:
allocating a first plurality of buffers, each buffer of the first plurality of buffers configured to store one frame of the plurality of frames; allocating a second plurality of buffers, each buffer of the second plurality of buffers configured to store one frame of the plurality of frames; selecting a reference frame for encoding at least one of the plurality of frames; assigning a picture ID number to the reference frame; if the first plurality of buffers is full of reference frames:
A) determining a first least recently used reference frame from the first plurality of buffers;B) determining, based on the picture ID number of the first least recently used reference frame and a first decimation factor, if the first least recently used reference frame is to be stored at a buffer of the second plurality of buffers;C) storing the first least recently used reference frame at the buffer of the second plurality of buffers if it is determined that the first least recently used reference frame is to be stored at the buffer of the second plurality of buffers; andD) removing the first least recently used reference frame from the first plurality of buffers; storing the reference frame at a buffer of the first plurality of buffers; encoding the at least one of the plurality of frames using the reference frame into an encoded video bitstream; and including the picture ID number of the reference frame and the first decimation factor in the encoded video bitstream.