A Frame-to-Frame<i>Picturephone</i>® Coder For Signals Containing Differential Quantizing Noise

Abstract

The frame-to-frame coder described in Ref. 1 used an 8-bit PCM signal for input. If, instead, the signal is obtained by digitally integrating the output of an element difference coder, the quantization noise may be misinterpreted as motion, and cause unnecessary transmission. In the particular example of the Phase I coder, <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> the quantization noise loads the frame codec to the extent that it produces an unacceptable picture. In this paper, a frame-to-frame coder for Picturephone® signals is described which is capable of coding the digital output of a Phase I codec for transmission over a 2-megabit/second channel. Improved methods are used to segment the noisy picture into moving areas and background areas. The moving areas are then transmitted using a number of data reduction techniques. During periods of slow movement, clusters of frame-to-frame differences in the moving area are transmitted. For moderate movement, frame differences are sent only in every other field, the moving areas of intervening fields being transmitted by a conditional field interpolation technique. For rapid movement, 2:1 horizontal subsampling is used, and, finally, during violent motion when the buffer fills, frame repeating is used. The picture quality obtained from a laboratory simulation of this system is believed to be satisfactory even for a very active subject. With small amounts of motion the subjective quality is actually improved because the visibility of the quantizing noise from the Phase I codec is reduced by the inherent frame repeating action of the coder.