Television Coding Using Two-Dimensional Spatial Prediction

Three classes of differential coders have been built which exploit the two-dimensional spatial correlation present in television pictures. The three classes are distinguished according to the technique used to predict the value of the picture element (pel) to be coded. In the first class, the prediction is the value of the previous element along the scan line averaged with one or more elements in the previous line (of the same field). In the second class, the prediction is the sum of the value of the previous pel with the local element difference in the previous line. In the third class, the transmitter and receiver choose the prediction from either the previous element or some combination of the previous element with the elements in the previous line. The choice is based on differences measured between elements already coded. These are available both at the transmitter and receiver, so no extra information need be transmitted. In the absence of transmission errors, the pictures resulting from all three classes of coder are markedly improved over those in which only the previous element is used as a prediction. In particular, vertical and sloping edges are now well defined. The effect of a transmission error on a single frame of the received picture is scarcely visible in the first class of coder but is much more visible in the second and third classes.