A low-bit-rate interframe coder for videotelephone

It has been suggested that customers for videotelephone service may be more interested in graphical information and in views of stationary objects than in head-and-shoulder views of people engaged in conversation. For this reason, an interframe coder simulation was constructed of a system that transmits graphics with full seven-bit pcm resolution, but displays scenes containing much movement with visible smearing in the moving areas. With the coder operating at 200 kb/s (0.1 bit per pel for a 1-MHz signal), a very usable (somewhat reduced-resolution) graphics picture can be transmitted in about one-half second, which is about as fast as the human eye can assimilate the information. A full-resolution picture is built up after 3 to 5 seconds but, except for high-detail scenes, it is very difficult to tell the difference between the half-second picture and the 5-second picture. Head-and-shoulder views of people engaged in low-key conversations are transmitted with quite adequate picture quality. Moving lips appear somewhat smeared, but it may not be enough to be objectionable if the audio is suitably delayed. However, large area movement is very visibly smeared–even to the point of being unrecognizable at moderate speeds. Whether or not this feature makes the coder unusable depends upon the value the user places on high-quality animated face-to-face conversation. Briefly, the coder works as follows: First, the signal is temporally prefiltered. Then moving-area peals are sent as line-to-line differences of frame-to-frame differences. As the buffer fills, field-to-field, pel-to-pel, and frame-to-frame subsampling as well as adaptive quantization are brought in as needed to reduce the data rate.