VIDEO STREAMING

摘要

A video sequence to be encoded is divided into a plurality of temporal portions, and analysed to determine (at least for each portion after the first), in accordance with a plurality of encoding quality settings, (i) a quality metric for the portion and (ii) the number of bits generated by encoding the portion at that quality setting. These data are analysed, for example using a Viterbi-like trellis, to choose a set of quality settings, one per portion, that tends to minimise a combined quality cost for the sequence. This the combined quality cost is the sum of individual quality costs each of which is a function of the quality metric of the respective encoded portion. The sequence is encoded using the chosen quality settings. In order to determine each individual quality cost, despite not knowing precisely what network throughput will be available at any given time in the future, one proceeds by estimating receiver buffer fullness and its standard deviation. From said estimates, the probability of buffer underflow and or overflow is obtained, the individual quality costs being a function also of the underflow and/or overflow probability.