st
of
ne
ill
next node will be selected according to the criterion of
minimum cost / maximum benefit.The setting of
parameters depends on the evaluation results of previous
processing or experience.
In our system, frame-based knowledge representation
provides flexiblity and inheritance of common knowledge
over a set of subroutines and parameter values.
In execution, firstly, we try to find some key factors
affecting the performance of a processing algorithm, then
to specify the different values of factors to a specific
problem-solving, we can store these promising values into
the legal values slots (see section 3). Our representation
structure allows us to store these values in advance. On
the other hand, we can also store less important factors
into default slots.
A user wil pay more attention to the specific
application problem instead of to the know-how about
problem-solving, i.e. he is interested in the final
result(state), but not in how to reach this. We provide
users with all possible final states for choosing. In the
system, we organize those states and some intermediate
layer states to form a search space. On each state node
there are different direct outputs according to input and
state. We give different weights to the state nodes in order
to arrive at a final node along a minimum cost path.
Tracing of the problem space search provides an
explanation facility, prviding answers to questions of the
types why(X)? and why. not(Y)?.
5. SUMMARY
In this paper, we present a knowledge based method
to solve some problems in image analysis. We embed
human understanding and experience about subroutines
in image processing packages into the system as a kind of
prior knowledge to guide the setting of parameter, and
organize the knowledge about techniques of image
processing to plan the process sequence. Although the
system cannot produce anything new to users, our aim is
to develop a more intelligent system as an expert system
for image processing. We regard the expert system for
image processing as a new flexible software environment
for developing image analysis. It facilitates the
development of image analysis for users . At the same
time, the increasing knowledge of image analysis can be
represented in the system to enlarge the systems
knowledge base and to expand application areas of image
processing.
In this section we would like to present some problems
to be solved.
(1). Evaluation of the analysis result. To go to a
next step in processing, we need to evaluate the result of
the last step. This analysis result will be represented in the
system as a kind of heuristic knowledge or analytical cost
function. In our method, a minimum cost evaluation
function is used.Alternatives for cost function definition are
being investigated.
(2). Description of ‘visual’ information. Our basic
strategy is to model the relation between three dimensional
scenes and two dimensional images on the basis of
313
physical systems analysis. However, some knowledge, like
that of a trained picture interpretor, is difficult to represent.
To facilitate the inclusion of this kind of unstructured
knowledge we are going to develop a friendly interface for
users to assist the description of visual information in
terms of geometric, radiometric, dynamic models.
(3). The integration of RS with GIS can be
achieved in a natural way by storing likelihoods and prior
probabilities in a (new)GIS, let a knowledge based system
generate hypotheses from the GIS and evaluate them
against evidence derived from RS data. This goal of
integration is approached through an overall research plan
"model based image analysis" being executed at ITC and
the UT in cooperation with members of the Dutch society
for pattern recognition and image processing.
REFERENCES
[1] A.Barr and E.Feigenbaum, The Handbook of Artificial
Intelligence, Vol. |, Stanford, CA: HeurisTech Press,1981.
[2] SPIDER USER'S MANUAL, Joint System Development
Corp., Tokyo,1982.
[3] Katsuhiko Sakaue and Hideyuki Tamura, Automatic
Generation of Image processing Programs by Knowledge-
based Verification.
[4] Keith Weiskamp and Terry Hengl , Artificial Intelligence
Programming with Turbo Prolog, John Wiley & Sons, Inc.
1988.
[5] X.S.Cheng,Design and Implementation of an Image
Understanding System: DADS, Ph.D Thesis, Delft
University of Technology, Netherlands, 1990.
[6] Mulder N.J. and H.P.Pan, Panorama: An Experimental
Image Understanding System, in Proceedings ISPRS VI
Symposium, Wuhan, China ,May 1990.
[7] Mulder N.J. and K.Schutte, Model Based Image
Analysis: Using Solid Modelling and Ray Tracing,
Symposium Remote Sensing & Space, Hat Yai, Thailand,
Jan. 1992.
[8] Sharon A. Stansfield, ANGY: A Rule-Based Expert
System for Automatic Segmentation of Coronary Vessels
From Digital Subtracted Angiograms IEEE Trans. Pattern
Anal. & Machine Intell., Vol.PAMI-8,No.2, pp188-199,
March 1986.
[9] Martin D. Levine and Ahmed M. Nazif, Dynamic
Measurement of Computer Generated Image
Segmentation, IEEE Trans. Pattern Anal. & Machine Intell.,
Vol. PAMI-7, No.2 pp155-164, March 1985.
[10] David M.Mckeown, Rule-Based Interpretation of Aerial
Imagery, IEEE Trans. Pattern Anal. & Machine Intell. Vol.
PAMI-7, No.5 pp570-585,Sept. 1985.
[11] Takashi Matsuyama, Knowledge-Based Aerial Image
Understanding System and Expert Systems for Image
Processing, IEEEE Trans. Geoscience and Remote
Sensing, Vol. GE-25, No.3,PP305-316,May 1987.
[12] Takashi Matsuyama, Expert System for Image
Processing Knowledge-Based Composition of Image
Analysis Processes, CVGIP, Vol.48, pp22-49,1989.
[13] A. Nazif and M.Levine, Low-Level Image
Segmentation: An Expert System, IEEE Trans. Pattern
Anal. & Machine Intell. Vol.PAMI-6, No.5 pp555-577, 1984.
[14] J.J.McKeown, D Meegan and D Sprevak, "An
Introduction to Unconstrained Optimisation”, Adam Hilger,
1990, ISBN 0-7503-0025-6.
