EXPERIENCES IN UPGRADING OF LARGE DATABASES OF SATELLITE IMAGES
iesus iai AE x ‘ b : : Ph
G. Chirici *, M. Gianinetto ", M. Scaioni
? Università degli Studi di Firenze, geoLAB (DISTAF), Via S. Bonaventura 11-13, 50145 Quaracchi, Firenze, Italy —
gherardo.chirici@unifi.it
? Politecnico di Milano, Dept. LIA.R., P.zza L. da Vinci 32, 20133 Milano, Italy
{marco.gianinetto, marco.scaioni} @polimi.it
Commission IV, WG 5
KEY WORDS: Automation, Spatial Database, QuickBird, Spot-5, Eros-Al, Multitemporal, Image Registration
ABSTRACT:
Many technical aspects are involved in the upgrading of satellite image databases: geometric registration, resampling, radiometric
adjustment, mosaicking. In this paper, after an overview of all problems, we will focus on the automation of image geocoding. A
procedure to perform automatic co-registration of satellite images have been already proposed by the authors (GEOREF), based on
an image-to-image registration technique implementing the automatic extraction and matching of corresponding points in a robust
way. In case one of the pair of co-registered images is already geocoded, the second one will be as well. Recently the
implementation of GEOREF algorithms in an operational environment has been completed and its application to upgrade a database
of satellite images has become possible. Furthermore, GEOREF is also able to compute the co-registration of images acquired by
different sensors, involving also high resolution and multi-resolution imagery. In this paper, tests concerning high resolution and
multi-resolution data fusion from Eros-A 1, QuickBird and SPOT-5 satellites are presented.
1. INTRODUCTION purposes of this paper. Detailed information can be largely
found in literature. Here we would only to make some
In the last years many countries have been carried out an almost considerations about operational aspects of image registration,
complete coverage of satellite images of their own land, in and to propose a solution to this problem.
particular at mid scale (e.g. Landsat TM/ETM+). Images are Geocoding two or more images means to establish a geometric
registered to a cartographic reference system (national and/or transformation between them in order to perform their reduction
UTM) and play the role of geographic support for a spatial to a common reference frame. Obviously, if one of the images
database, suitable to be integrated by vector layers and by other is already calibrated to a given cartographic reference system,
kinds of raster data. after geocoding the other image(s) will be as well. In case of
The availability of this coverage is fundamental to investigate small and mid scale satellite imagery (but practically this is
and to detect changes in the use of the soil. An example is generally true), this transformation is estimated on the basis ofa
represented by the well known CORINE project (Perdigäo & set of control points (CPs) which are measured on both images.
Annoni, 1997). The availability of two co-registered remotely Usually, the measurement of CPs is carried out manually from a
sensed images of the same area at two different dates enables skilled operator, resulting in a largely time consuming task.
the development of studies regarding land cover and landscape — Nevertheless, to get a high quality on this process, the operator
dynamics (Forman & Godron, 1986). These are usually based must be very experienced, because in many cases the correct
on the development of diachronic land use/land cover maps and accurate measurement of homologous CPs is not so ease.
which are then analyzed with cross-tabulation techniques and On the other hand, different automatic procedures have been
landscape metrics. Multi-temporal maps are also the basis developed, based on image matching algorithms; among the
dataset to model future development of the landscape with others, methods proposed by Corvi & Nicchiotti (1995), Dare &
different techniques such as ccllular automata or Markov Dowman (2001) and Goncalves & Dowman (2003) cannot be
chains (Baker, 1989; Sklar & Costanza, 1990). Such application neglected. Unfortunately, results of these studies have not
are often based on old aerial photos that have to be scanned and followed up on the most widespread software packages which
co-registered with recent photo-planes or ortho-photomaps. are currently used to deal with remote sensing imagery. The
Obviously, the spatial database should be frequently updated by consideration which is easily addressed is that the most of the
introducing new images, either of recent acquisition (to know published algorithms have kept a very limited application,
the current use of the land) and from historical archives (to which have only concerned a small dataset and have not been
detect changes with respect to a given time in the past). implemented in a deliverable release.
Frequently, different kinds of images have to be fused together,
requiring the availability of geometric transformations which
are able to compensate for differences. 2. THE GEOREF SOFTWARE
Among problems involved in upgrading a large database of
images, data geocoding is of great importance, due to the fact A procedure to perform automatic co-registration of satellite
that this task is a prerequisite to any other geometric task. The images have been already proposed by the authors (Carrion ef
analysis of different techniques and algorithms that have been al, 2001, 2002; Gianinetto & Scaioni, 2003). The adopted
developed to perform image registration is out from the procedure, here referred to as GEOREF, is based on an image-
Us
oo
N
Interna
to-ima;
points
experi
been [
been a
images
fests, I
points
registr:
compa
Recent
operati
applic:
becom
enviro
divide:
showit
side ar
The m
sub-pa
24 I
Basic:
image
about
An e»
extrac
Spectr
be us
menti
rende
conce
enhan
stretcl
GEOEI
pixel,
image
In cas
frame
“worl
transf
