ncy of
Iracy.
A new
of CIPA
qualita'
iplinare,.
erritorio,
onero.
Metodi di
. Saluzzo
walita nei
Analisi
mite un
nferenza
-Arnold,
ork,.
mation —
SOME EXAMPLES OF TABULAR ALGORITHMS IN SYNTACTIC PATTERN RECOGNITION
T.Bellone*,E. Borgogno *, G. Comoglio *
*DIGET, Politecnico di Torino, Torino, Italy
tamara.bellone@polito.it, enrico.borgogno@polito.it, giuliano.comoglio@polito.it
Commission VI, Working Group 3
KEY WORDS: Pattern Recognition, Image understanding, Algorithms
ABSTRACT:
Image understanding is a very important step in the course of data analysis in GIS. Syntactic Pattern Recognition of
models is a process, as used in Cartography and Remote Sensing, which makes possible the matching of parts of maps,
of images and 3D models with archetypes and patterns (parsers). The main item of the entire process is the so-called
Parsing, a device used in Linguistics, successively used in Cognition Disciplines. Simply, it must help decide whether a
data string (a phrase) can be a part of an existing pattern (language).
A number of algorithms are available for Parsing, for the needs of specific grammars: although not suited for any
grammars, tabular methods help save time, as the Kasami method, remarkably simple to use: it works well in the case of
context-free grammars, as reduced to the so- called Chomsky’s normal form.
1. Foreword
The so-called computer science revolution has also taken
place in the area of geodetic sciences, which eventually
are playing a key role in it (Geomatics).
Even common people know that many mental preocesses
are based upon evaluation of contrast and difference: this
is, for instance, the art of seeing, and generally a good
approach in all cognitive sciences, also in Linguistics.
Language, indeed, is based upon blending of discrete
parts (phonemes, morphemes). Also, vision and speaking
are based upon principles not quite different. This is why
some present procedures of Geomatics may be referred to
logical and symbolic structures proper for Mathematical
Logic and Linguistics. Some improvements in GIS and
Digital Photogrammetry are referred to as Computer
Vision, Image Processing, Image Understanding,
Machine Learning, which are linked to developments of
Artificial Intelligence.
An easy case of this cultural melting is Syntactic Pattern
Recognition: it is a procedure, widely used in
Cartography and Remote Sensing, that trusts upon
matching of sections of maps and/or images or 3D
models with archetypes or objects (parsers). Also, parsing
is a topic proper of Linguistics, which has been borrowed
from cognitive sciences; Artificial Intelligence in turn is
based upon Logic and Linguistics (Mathematical Logic
and Mathematical Linguistics).
A Syntactic Pattern Recognition system consists with
three parts: pre-processing, pattern description and syntax
analysis.
Pre-processing includes pattern encoding and
approximation, filtering, restoration and enhancement.
The pattern representation procedure consists of pattern
segmentation and feature (primitive) extraction, in order
to represent a pattern in terms of its sub-patterns and
43
pattern primitives: each sub-pattern is identified by a
given set of pattern primitives.
The decision whether or not the representation belongs to
the class of patterns described by the given grammar or
syntax (is syntactically correct) is made by a “parser”.
Parsing is then the syntax analysis: this is an analogy
between the hierarchical (treelike) structure of patterns
and the syntax of language. Patterns are built up by sub-
patterns in various ways of composition, just sentences
are built up by words and sub-patterns are built up by
concatenating primitives (features) just words by
concatenating characters (Sester, 1990).
Parsing adaption to the GIS context can be thought for
selecting reference shapes from numerical cartography or
raster geocoded images. Phrases are cartographic objects
and language is the set of rules which define these objects
satisfy or not selecting criteria.
A double use of Parsing is possible in the framework of
GIS: both at the generation stage, and for the process of
query and data analysis.
The shape reconnaisance by parsing is of evident interest
for:
> map production, implementation and updating
through automatic extraction of shapes from raster
data;
» editing of existing maps (as association of codes to
not-still-coded shapes or perceiving of difference
between a closed polygon and an open line in order
to close it);
» query purposes, aimed to the extraction of raster
features or not-coded shapes from the database; an
ample variety of questions may be posed, as one
deals with new geometrical features which can be
classified as recursive shapes according to a
reference pattern.