annel for
DTM and
versity of
valuation,
da images
ed.
ance and
iversity of
Australian
> images,
mation on
atky who
1ake data
Seige, P,
tereo Im-
: Interna-
sing, Vol.
nstrained
1d Photo-
rful Tool
RS Jour-
ation Ex-
ial Map-
ammetry
, Part B4,
ching for
1944, pp.
Study on
:ct Using
chives of
C., USA,
mination
À ustrala-
Ground
tion Im-
ference,
ssing of
gramme-
the Spa-
national
J.
IMAGE-MAP-FUSION BASED ON LINE SEGMENT MATCHING *
Renate Bartl Werner Schneider Joachim Steinwendner
Institute for Surveying and Remote Sensing
Universitàt für Bodenkultur
(University of Agriculture, Forestry and Renewable Natural Resources)
Peter-Jordan-StraBe 82, A-1190 Vienna, Austria
e-mail: (renate,schneiwe,joachim Gmail.boku.ac.at
Commission IV, Working Group 3
KEY WORDS: Fusion, Matching, Registration, Landsat, Vector Data, Edge Extraction
ABSTRACT
Registration of cadastral information and the result of satellite image classification is a frequently encountered problem. The
approach taken in this contribution matches field border lines extracted from a given satellite image and lines from a digital
cadastre. In order to improve accuracy, edges and perceptual lines are extracted from an image in subpixel resolution obtained
by spatial subpixel analysis. The matching is performed by comparing characteristic features of line segments. The matching
result contains the information required for determining registration parameters as well as a set of explicit relations between
property borders and field borders on the image.
KURZFASSUNG
Die Überlagerung von Katasterinformation und dem Ergebnis einer Landnutzungsklassifizierung stellt ein in der Praxis wichtiges
Problem dar. In diesem Beitrag werden die aus einem gegebenen Satellitenbild extrahierten Linien, die Feldgrenzen entsprechen,
mit den digital vorliegenden Katasterkanten verglichen. Mittels ráumlicher Subpixelanalyse wird das Satellitenbild in Subpix-
elauflósung transformiert, um eine hóhere Genauigkeit bei der Extraktion von Linien zu erhalten. Das Matching basiert auf
einem Vergleich von bestimmten Linienmerkmalen. Basierend auf dem erzielten Ergebnis kónnen die Parameter für eine genaue
Registrierung eruiert werden. Darüber hinaus wird eine Zuordnung zwischen den Eigentumsgrenzen des Katasters und den
sichbaren Feldgrenzen explizit hergestellt.
1 INTRODUCTION and cadastre might be of sufficient accuracy. Otherwise one
can use a registration procedure as proposed by [3] which
might need further preprocessing. Then, edge detection is
performed on the image yielding edges with subpixel accuracy.
One possible approach, presented in [15], is the identification
of edgels with subpixel precision by detecting pixels which are
part of an edge. The location in subpixel precision is deter-
mined by interpolating along the gradient direction towards
the neighbouring pixel with higher gradient magnitude. A
subsequent chaining procedure collects neighbouring edgels
and produces lines by least squares adjustment. However,
we take another approach which not only delivers edges but
also perceptual lines, as described in section 2.2. Correspon-
dences between image edges and cadastral boundaries (also
available in vectorized form) are found in a matching process
presented in section 3. By this strategy, the "distortion" of
cadastral boundaries relative to the geometry of the image is
A frequently encountered problem is to fuse [2, 9] the result of
satellite image classification with cadastral information. Po-
tential fields of application are GIS input conversion [14, 4],
map generation and updating [16] or agricultural land use
monitoring which is the topic of this contribution. In order
to solve these tasks, for standard satellite image data (e.g
Landsat TM with a resolution of 30 m x 30 m) subpixel
accuracy is required which can hardly be attained by con-
ventional methods of image registration due to the following
problems:
e A large number of control points of high accuracy is
necessary which are difficult to find in the image auto-
matically.
e High subpixel accuracy cannot be attained for image
data geometrically preprocessed with nearest neigh-
A determined.
bour resampling.
e |f the terrain is not flat, a digital elevation model 2 INPUT DATA
(DEM) is required causing additional problems of avail-
ability and costs. For the matching of property borders of a cadastral map
: with corresponding image information, visible field bound-
In this work, we start with a rough registration of the digi- ^ aries must be identified on the Landsat image. Agricultural
ta cadastral boundaries to the satellite image. Although, in fields in Central Europe are typically long and narrow due to
t Is contribution, this is done manually, one could automate the repeated splitting between the children of the farmer for
this step. Occasionally, the orientation information of image generations. Thus, the spatial resolution of Landsat TM im-
; This work is supported by the Austrian "Fonds zur Fórderung der ages of 30 m. x 30 m is not sufficient for the identification
wissenschaftlichen Forschung" (project $7003). of such fields. Spatial subpixel analysis helps to alleviate this
*
147
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996
