Patent US5020121: Neighborhood block prediction bit compression 

 Inventor: Rosenberg; Charles J., Hewlett, NY
Applicant: Hewlett-Packard Company, Palo Alto, CA

Application Date: Aug. 16, 1990
Publication Date: May. 28, 1991

Abstract:

A method for compression of data used to describe an ideographic
character, or a set of such characters such as Japanese, Chinese,
Hebrew, Arabic or Tamil characters or a set of two-level images, on a
computer screen. The computer screen is divided into an array of
non-overlapping pixel sub-blocks, each J pixels wide by K pixels high,
and a neighborhood of L adjacent pixels is defined for each target
sub-block. The known pixel configuration (white versus black or off
versus on) of the neighborhood of pixels is used to predict the rth
most probable pixel configuration for each target sub-block, for r=1,
2, . . . , R-1 (R>=2), based on a statistical analysis of the
sub-block pixel configurations for all characters in the set. Where a
particular target sub-block pixel configuration cannot be predicted
from the R-1 pixel configurations associated with the neighborhood
pixel configuration, the exact pixel configuration is used. Use of the
exact sub-block pixel configuration is not often required. This method
allows an average percentage compression or bit savings for the bit
map representation that ranges from about 30 percent to about 70
percent, for a well known set of 6802 Kanji characters.