309
ON A NEW CLASSIFICATION METHOD
BY MULTI-DIMENSIONAL HISTOGRAM
Y. Haba, T. Iizuka and N .Hakada
Kanazawa Institute of Technology, P.0. Kanazawa-South,
Nonoichimaci, Ogigaoka 7-1,Ishikawa 921, Japan
ABSTRACT
It has been generally accepted that the remotely sensed MSS
data are the most efficient way immediately to obtain the
information of a ground surface. They have universally been
used as the economical means for environmental pollution
monitoring, natural resource monitoring, land-use map,
ecological map and so on. It is generally accepted that
they should be basically classified prior to various kinds
of their application.
There is the unsupervised classification method which is
superior to the supervised one from points of accuracy and
universality. But in the conventional unsupervised
method,there are such problems that the distribution of the
clusters are assumed to be normal one and that it is time
consuming to classify the MSS imagery due to the complicated
computation proportional to the number of classes per pixel.
Thus, a new classification method by using multi-dimensional
histogram (MDH) table has been proposed. The MDH method has
mainly three advantages. First,it has no constrains on the
cluster distribution. Second, it is possible rapidly to
classify the MSS cmagery owing to the table search method.
Last.it requires the small capacity of the main storage for
stowing the MDH table. As results, it has such more another
advantage that a mini-computer is available for the
unsupervised classification system which has been limitted
only to a large scaled computer.
INTRODUCTION
Various approaches to the resource monitoring ,environmenta1
pollution monitoring, ecological map, land-use map and so on
have been reserched and developed by using the imageries
remotely sensed by the space-boarn or air-boarn multi-
spectral scanner(MSS) . It is necessarily important and
inevitable that MSS data are classified by any means in
advance of their application to the actual various problems.
Classification of the MSS data should be analized not only
by using the spector data per pixel but also by using the
areal data as used in the texture analysis and the temporal
analysis. Futhermore the high level pattern recognition
technique like 'Production System' is desired to be applied
to the classification of the MSS data. But it is also
important to make the great efforts in order to obtain the