Progressive Encoding of Facsimile Images.
01 January 1988
The increasing popularity of facsimile, desktop publishing and electronic document retrieval suggest a substantial amount of non-voice traffic carried by future networks could be raster scanned images of two level (black and white) documents. For these applications rapid output to a high-resolution document monitor is frequently desirable and sometimes even necessary. In an earlier memorandum we described a scheme for decomposing a bi-level image into a half-resolution image (essential) plus supplementary information. This separation of information into non-droppable (essential) and droppable (supplementary) is crucial for efficient congestion control in a packet network carrying simultaneously non-bursty, delay-sensitive (voice, video) traffic and bursty, delay-tolerant (computer data, facsimile) traffic. If this decomposition is applied recursively, an image can be decomposed into a super-low resolution image plus successive fields of supplementary information each providing a resolution doubling. Encoding an image in this way costs 5 to 10 percent in compression efficiency, but allows the comparatively rapid presentation of a low resolution image followed by resolution enhancements over time. On a very high bandwidth channel there is no reason to make this tradeoff since final resolution is reached in a time subjectively instantaneous (less than a second). On a very low bandwidth channel it is also uninteresting, this time because even the low-resolution image requires a subjectively intolerable transmission time. The subjective benefit of progressive presentation is likely to be greatest with a transmission bandwidth of about 64 Kbit/sec (i.e. ISDN basic rate).