Bell Telephone Laboratories

Kenneth A Walsh

Arun N Netravali

Frank W Mounts

A series of bits which represent the intensity values of elements of a facsimile picture are preprocessed to eliminate runs of length one, and then applied to a predictive encoder which treats values for alternate pels in two distinct ways. Specifically, the value of every odd pel is predicted, based upon the intensity values of some surrounding picture elements. Each prediction is compared with the true intensity value to yield an error signal. The encoder is also arranged to generate a reference signal which indicates whether the confidence in each prediction for odd pels is high or low. The error values for odd pels within a predefined reordering interval are then sorted or reordered into two groups, dependent upon the reference value. On the other hand, the intensity values of even pels are predicted only when the values cannot be precisely determined from the two previous and subsequent values; the error values for even pels are assembled into a third group. When data for an entire reordering interval has been processed, each of the groups is run length coded for transmission. By encoding even pel values only when they are needed, the amount of data required to represent the facsimile picture is reduced, at the expense of only a slight loss in detail, which is not objectionable. Further, since the statistics of the groups formed by reordering are different, different code dictionaries can be used for run length coding.

Run length encoding of facsimile pictures

A facsimile signal encoding technique reduces the amount of information needed to represent a picture by (a) generating an error signal (on line 1120) which indicates whether an intensity value prediction corresponds to the true intensity value, (b) generating a reference signal (on line 1005) indicating whether the confidence in each prediction is high or low, (c) reordering the error signal values (logic 1010) into high and low confidence groups in accordance with the reference signal, and (d) assigning code words (encoder 1016) to represent the run lengths of the reordered data. Advantageously, different code dictionaries are selected (line 1019) for code assignment in each of the groups, each dictionary being tailored to the type of runs expected in each group. Also, code words for certain runs may be dropped, and the lengths of other runs may be combined (by logic 1015) before run length coding, again increasing encoder efficiency. A receiver (FIGS. 16-18) for decoding the run lengths is described.

Growing community of inventors

Atlantic Highlands, NJ, United States of America

Kenneth A Walsh