have been already filled in the cluster column of are
found within its vicinity and when only one kind of the
cluster numbers is allocated to them,the cell is
recognized to belong to the same cluster. The cluster
number is registered in the cluster column of the cell,
step 3. When there are more than two kinds of the
clusters in its vicinity, the cell is considered to be
located on the bottom of the valley, that is, the
boundary between the clusters, the cluster number of the
mountain nearest to the cell is registered in the
cluster column of the cell.
Radius of the vicinity. From the same reason as the
histogram space was quantumized in its generation , the size
of the vicinity,that is,the radius drv of the hyper-sphere
should vary according to the distance from the origin.
The radius drv of the vicinity should be reasonably
described as followings,
drv(x)=a*drq(x) (9)
where a is a constant deciding the size of the vicinity. The
constant a should be rather larger than one. Otherwise,any
other cells except oneself under consideration would not be
found within the vicinity.
The searching time. It is basically possible
sequentially by checking all the distance between the cell
and ones with higher frequence to find whether there may be
any other cells within the vicinity or not. When the cell
is especially stored in lower part of the MDH table, there
is a great problem of searching time because the number of
the cells to be compared with is rather large. To reduce
the searching time, a key k presented by the following
eq.(10) is introduced to our system.
By checking whether its key may be in the following range
or not,
Z xi-SQRT(N) * drv <= k <= Z xi + SQRT(N ) * drv (11)
i i
the cell can be known of the potential presence in its
vicinity.
Clustering process. For the simplicy, a diagram is shown
in Fig.(3) to demonstrate the procedure of the cluster
analysis when the histogram space is one dimensional. The
small figures on the shoulders and capital ones inside the
figure show the processing order and the allocated cluster
numbers, respectively. The diagram teaches us that the one
dimensional histogram is classified into two clusters 1 and
2 through 15 processes. 1st and 4th processes generate new
clusters and 15th process discovers the bottom of the
valley.
Threshold of the radius. When the cluster analysis is
315