Rule-Based graphics recognition on unconstrained maps
James Z. Xu
Cohtmmt Research Inc., One Adler Drive, East Syracuse, New York 13057
Abstract
This paper presents a graphics recognition system for processing scanned maps with
degraded image quality. The system extracts complete object information and uses it to
form a reliable data structwe for better representing the original shapes. Its knowledge
sources are organized in a hierarchy of independent modules for inawsed adaptability; its
spatial reasoning can be executed in a bottom-up or top-down manner. The system also
employs a spatial query technique for efficiently processing spatial objects.
Introduction
Paper-based technical drawings and utility maps are complex documents containing a wide
variety of graphical and textual information. The main objective of an automatic data
conversion system is to captwe the information in these paper documents in an intelligent
form. There m specific issues surrounding the conversion application that place additional
demands on the platform that must support an automated conversion system. Since
automation must always be augmented by human intervention, the system must be robust
enough to deal with partial information and intelligent enough to detect inconsistencies
covering abroad range of cases.
Two critical matters must be addressed in the design of automated conversion systems: the
recognition of graphics in a conversion system, and the ability to work with domain-independent
information. Specifically, these systems must be able to classify severely
degraded, “real-world” graphics and process shape information independent of what the
graphical objects actually represent. As intelligence is added to the objects the system
produces partially-instantiated domain objects, which maintains them until the conversion
is complete. At the same time, the system must reason and process spatial objects as
efficiently as possible.
Over the last few years, considerable progress has been made for recognizing graphics.
However, in many popular systems some drawbacks are common, such as partial use of
raster information, inadequate adoption of simple architectures, lack of high-level
processing knowledge and mishandling of spatial operations. This paper presents a
graphics recognition methodology which addresses these problems.
Domain Objects and Knowledge
The recognition system discussed in this paper is based on a multilevel object model that is
comprised of four layers (Figure 1). In decreasing order of semantic meaning, these layers
anx
Figure 1 Four layer object model for maps
- Domain objects
- Primitive graphics
- Primitive vectors
- Raster images
Domain Objects m all the objects in the target system data model. In the case of
Automated Mapping /Facilities Management (AM/FM) systems, these objects are the
facilities and land features. Primitive Graphics are shape @ consisting of objects such as
circles, x-shapes, squares and triangles. Primitive Vectors are line and polyline structures
without any other meaning attached to them. Raster Images are pichues of the hardcopy
document to be converted.
