Entropy Measurements for Nonadaptive and Adaptive, Frame-to-Frame, Linear-Predictive Coding of Video Telephone Signals

In coding television pictures for transmission over a digital channel, it is well known that the required bit rate can be significantly reduced by removing various redundancies that exist in the signal, and in recent years methods for removing frame-to-frame redundancy have been investigated. 1 In a conditional replenishment 2 system, only the picture elements (pels) that have changed significantly since the previous frame are transmitted. Their amplitudes as well as their locations must be sent; however, most of the transmission capacity is used in sending the amplitudes. During periods of rapid motion, only every other moving-area pel need be transmitted, i.e., the moving area of the picture can be subsampled 3,4 at half-rate with the unsampled pels being replaced by the average of their neighbors. 1155 Linear predictive coding is an efficient method of transmitting these amplitudes. Channel rates of 1 bit per pel and below have been obtained. 4-7 With this technique, a prediction is formed of each pel to be sent by computing a linear combination of previously transmitted pels. The difference between the actual value and prediction is then quantized and coded for transmission. Since the differential signal is small usually and large only occasionally, variable word-length coding can be used to good advantage in reducing the overall bit rate. The entropy of the quantized moving-area differential signal provides an estimate of the average number of bits required to transmit a pel.