performed by the method as mentioned above, the clusters are 
unnecessarily very large in number. It is meaningless to 
obtain clusters of the cell with low frequence. So the 
threshold of the frequence is introduced into our system. 
Cells with the lower frequence than the threshold are 
allocated to the already classified cluster when a cell 
belonging to the cluster is contained within the rather 
larger vicinity than normal one. Otherwise,they are 
allocated to the unclassified class. It gets rid og the 
meaningless increase of the clusters. 
Fig. (3) Procedure for the cluster analysis 
CLASSIFICATION 
Index. It is important and desirable that the 
classification is executed in short time. Information about 
the relationship between the pixel data and the class number 
is stored in the MDH table in our MDH method. So the cell 
which contains just the same pixel data as ones at the pixel 
to be classified has to be accessed swiftly as possible. 
To make it possible to access the MDH table at high speed, 
the index has to be created again by the same way as the 
histogram table was generated from the actual MSS imagery. 
It is easy to link the each cell with the same key in 
descending order of the frequence because the MDH table is 
already sorted in descending order of the frequence in 
advance of the cluster analysis. 
Procedure, To classify MSS data,a key has to be obtained 
by using a part of quantumized pixel data of the actual MSS 
imagery. V/ e can always find out the histogram cell 
containing the data quite equal to the pixel data. The 
number registered in the class column of the cell is 
allocated to the pixel as its class number. These procedures 
are sequentlly applied to every pixel to obtain the 
classification map. 
Example. The local area of 256 x 240 (covering the 
central Japan) extracted from the data obtained by the 
U 
316