57
Symposium on Remote Sensing for Resources Development and Environmental Management / Enschede / August 1986
Man-machine interactive classification technique
for land cover mapping with TM imagery
Shunji Murai, Ryuji Matsuoka & Kazuyuli Motohashi
Institute of industrial science, University of Tokyo, Japan
ABSTRACT: The conventional classification, for example, the maximum likelihood method for land cover
classification has limitations of lower accuracy because of an assumption of normal distribution for
respective classes. In order to improve the accuracy of classification, the paper deals with an improved
classification technique based on man-machine interactive system to enable to consider the actual distribution
and to enable to interactively study between the imagery, histogram and the existing topo-map using a computer
system with two color graphic displays and a digitizer. To assist human interpretability, HSI ( Hue,
Saturation and Intensity ) conversion and/or pricipal components can be applied to the conversation with
the computer system. The improved technique was tested in the vicinity of Tokyo Area with Themetic Mapper
data. As compared with the result of the conventional method, the result of the study was much improved.
The man-machine interactive classification technique developed by the authors can be said " a real time
and on-line classification ".
1 THE OBJECTIVE OF THE STUDY
In order to improve the following disadvantages of
conventional maximum likelihood method;
* misclassification due to the assumption of normal
distribution
* lack of consideration with detail information on
each pixel in the histogram or each geographic
position on a map, and
* lack of adoption of classification order from the
reliability or separability,
an interactive man-machine classification technique
has been developed to enable a real time or on-line
classification with use of image plane, two dimen
sional histogram plane and map plane located on a
digitizer tablet.
2 MAN-MACHINE CLASSIFICATION SYSTEM (MACSYS)
2.1 Hardware system
The following units are utilized.
1) Computer: FACOM M 170 F (8MB memory, 7.6GB Disc)
2) Color Graphic Display: NEXUS (512x480x27bits)
3) Digitizer: Oscon graphic digitizer
The following three planes are registered in order
to implement the man-machine classification system
as shown in Figure 1.
1) Image plane: generated in graphic display with
a work file
2) Histogram plane: generated in graphic display
with a work file
3) Map plane: generated on tablet digitizer with a
geometric transform to both of image
plane and histogram plane
2.2 Software system
The following softwares were developed in menu type.
1) FRQ-IMG: A point or polygon given in a histogram
plane is tranformed into the image plane
2) IMG-FRQ: A point or polygon given in an image
plane is transformed to the histogram
plane
3) IMG-IMG: A point or polygon given in an image
IMAGE PLANE
FRQ-IMG
\
HISTOGRAM PLANE
1
IMG-FRQ
MAP-IMS \ /
MAP-FRO
Figure 1 Man-machine interactive system (MACSYS)
4) MAP-IMG:
5) MAP-FRO:
6) ON IMG:
7) ON MAP:
8) CATEGORY:
9) DISPLAY:
10) RESET:
plane is classified in a given color
code
A point or polygon given in a map plane
is classified in a given color code
A point or polygon given in a map plane
is tranformed to the histogram plane
Classified map is edited in an image
plane
Classified map is edited in a map plane
A class to be classified can be given
A current work file will be displayed
A work file will be cleared
Both of image and histogram planes can be zoomed up
to 4 times and 8 times. The current results of
classification may be stored in a registered work
file of image and histogram plane respectively.
MACSYS also includes a function to detect which
pixels in an image plane or histogram plane are
not yet classified.
