Abstract: Serial runlength data obtained by scanning an image includes "slices" that each include the length and an ending pixel number of a successive string of connected black pixels. The runlength data is operated upon, line-by-line, by a decision tree classifier that creates a software "object" including a first linked list of a plurality of further linked lists each of which contains all of the slices of a simple shape of the image. The slices of the object are entered into a "frame" in the same order in which they are scanned and are "filtered" to delete insignificant horizontal and vertical gaps between slices. The slopes of edges of the shapes are tested for linearity to determine which shapes can be represented as trapezoids that can be represented by four corner points and which must be represented as irregular blobs containing all of their slices.