Manfred H. Günzl
A NEW SEGMENT SHAPE PARAMETER FOR GRID DATA AND ITS APPLICATION TO LAND
USE SEGMENTATION.
Manfred H. GÜNZL*, Olaf HELLWICH**
* Institute for Biomathematics and Biometrics
GSF-National Research Center for Environment and Health
Manfred @Guenzl.de
** Chair for Photogrammetry and Remote Sensing
Technical University of Munich
Olaf.Hellwich @photo.verm.tu-muenchen.de
KEY WORDS: Shape Parameter, Segmentation, Land Use, Algorithms, Mathematical Models, Radar
ABSTRACT
Various land applications require an image or image series to be divided into segments corresponding to areas of homo-
geneous land use. Image segmentation is needed to generate and update such GIS-stored geometric land use information.
Regarding agricultural land use the shape of segments is a major feature to determine the correctness of a segment.
Most remote sensing data is derived and stored as grid data. Due to the fact that most region segmentation approaches
are operating on such grid data, segments are handled as groups of grid cells. The shape of such segments is given by a
polygon within that grid. The topic of this paper is a new approach to parameterize the shape of segments of grid data.
A common approach to parameterize the shape is the compactness determined as the ratio of the area and the square
of the boarder length. For polygons within a grid this parameter is highly dependent on the orientation of the segment.
Regarding the application to remote sensing data, the major drawback of the parameter is that it depends on the sensor
position and orientation. Due to the stairway-like shape some parts of the boundary usually are longer than the real
boarder they approximate. In existing approaches, this effect is reduced by previously generalizing the polygon. This has
two drawbacks. First of all the result depends on the way the generalization is conducted. The generalization of polygons
enhances the compactness of the enclosed segment by simplifying its shape. To stay as close as possible to the shape
of the original object, it is necessary to pay attention to the way the shape was effected by the image grid. Thereby the
second drawback can be recognized: The generalization approach is computationally expensive.
In this work, a different, new approach is suggested. The goal was to determine the compactness of an object out of
grid data by compensation of the influence of the image grid. The compensation was found to be possible using a set of
geometrical parameters. This concept lead to a shape parameter with several advantages:
e The parameter is completely independent of the orientation of an object regarding the grid.
e The geometrical parameters of a segment derived out of merged segments can be determined from the parameters of
the original segments. Thus, applied to region growing by merging, the computational effort is independent of the
shape, size and girth of the regions and therefore constant. This enables shape-controlled region growing with the
same order of computational effort as without shape control.
e The same set of parameters also enables the compensation of the lateral ratio of rectangular objects. This can be
carried out independent of the orientation compensation.
Summarizing, the parameter describes the deviation of the shape of a segment from a rectangle. It derives to 1.0 whenever
the segment is rectangular — independent of the location and orientation of the segment with respect to the grid, as well as
of the ratio of side lengths of the rectangle. When the segment increasingly deviates from rectangular shape, the parameter
increases continuously. As a result this new, computationally efficient shape parameter is perfectly suited to segment areas
of agricultural land use usually consisting of rectangles.
The shape parameter was implemented as part of a computationally highly efficient region growing algorithm. Shape as a
parameter for segmentation control is of special interest when applied to noisy data. Therefore, results of the segmentation
of speckle influenced, simulated and real SAR data are given.
1 INTRODUCTION
Segmentation of remote sensed data of agricultural areas is the attempt to recognize homogeneously used regions. If
someone evaluates such data manually and she or he is in doubt about the geometry of a field, she/he will usually make
some assumptions:
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 351