Abstract: The paper deals with low-bitrate video coding using matching pursuit. Here,
it is proposed to augment the dictionary by use of some nonseparable functions derived
from the images obtained in motion-compensated prediction. The prediction is performed
in the exactly the same way both in the encoder and the decoder. Therefore,
these additional functions in the dictionary may be derived just in the same way in the
coder and the decoder. Thus, there is no need to send any description of them to the decoder.
These additional content-adapted dictionary elements improve video encoding as
proved by experimental results showing some improvement of compression efficiency
of the contemporary AVC/H.264 video codec.