On the use of orientation information for improved contour difference coding

Contour difference coding represents Freeman chain codes using only the differences between consecutive links. Improved contour difference coding is possible through the use of information about the orientation of the links forming the differences. It is shown analytically that grid-intersect quantization of curves inherently results in different underlying probability distributions for diagonally and vertical-horizontally oriented chain differences. This difference can be exploited to produce models with reduced source entropy. Results are presented demonstrating the effectiveness of this method on real contour data