Buffering of Data Generated by the Coding of Moving Images

Although most coding schemes for pictures have considered only a single frame at a time, it has long been realized that if only the changes 239 240 T H E BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1972 between frames were transmitted the average bit rate could be reduced significantly. 1,2 For Pidurephone® service this is particularly true because most of the time the person using a Pidurephone set sits quite still or only moves his mouth. However, extreme movement can occur when a camera is panned, when a user leans forward to adjust the controls, or when he stands up and moves out of the field of the camera. In such instances the amount of movement can exceed even that which would be expected in movies or broadcast television. But even if such extremes were unimportant, it would still be necessary to buffer the data generated during normal human movement in order to take advantage of the large reduction in average bit rate which occurs when one transmits only those parts of the picture which change. Movements exceeding one second in duration are probably not unusual during videotelephone use. To smooth the peaks in data generated during such movement would require a large buffer, capable of storing a significant fraction of the data generated during the movement. Even if buffers were cost-free items, it would probably not be feasible to smooth, completely, the flood of data generated during movement because of the delay introduced into the signal path by the buffer. (The maximum round-trip delay that can be tolerated in a conversation is between one-half second and one second.