RG-DW IMAGE MATCHING SCHEME
AND ITS APPLICATION
Yan Lue, Kurt Novak
Department Of Geodetic Science and Surveying
The Ohio State University
1958 Neil Ave.
Columbus, Ohio, 43210-1247, USA
Commission III
Abstract
In this article we report on research performed at the Ohio State University to
automatically extract digital elevation models and orthophotos from satellite images and
digital aerial photographs to implement the integration of digital data into GIS. The focus of
this research was to develop an operational image matching technique to automate the
whole process for acquiring digital elevations. In contrast to many others, our RG-DW
(Recursive Grid—D ynamic Window) matching scheme does not need any a priori
knowledge about image orientation nor any information of object space, it takes advantage
of the parallelism of the matching scheme and makes a comprehensive utilization of a series
of simple concepts as elementary tools, such as pyramid images, geometric transformation,
multiple dynamic window size, cross-correlation and least squares matching. The only
approximations needed can be taken from four corresponding points of the overlapping
area of a stereo model. Matches are obtained over the whole area of interest by recursively
densifying a regular grid through scale space. This matching scheme was applied to
develop the package MATCH for automatic image matching for both SPOT and aerial
photos, which is the key part of the Digital Ortho Module, the new commercial digital
photogrammetric module recently released by ERDAS Inc. MATCH has been fully
integrated in a raster GIS, together with a variety of functions to compute the orientation of
satellite and aerial images, to perform the matching, to interpolate a grid DEM, and to create
digital orthophotos. Thus, digital image data can be directly analyzed in the GIS.
Key words: Image Matching, Digital Photogrammetry, GIS, SPOT Image, Aerial Photo
1. INTRODUCTION
Image matching, one of the central problems for
automating photogrammetry, is the key to simulating the
function of human vision to replace an human to do stereo
observation and measurements by computer. An
overwhelming diversity of approaches proposed by many
researchers who tried to tackle the matching problem shows
its very high complexity and difficulty. It seems that the
proposed methods are getting more and more complicated.
However, whether area-based/signal matching (cross-
correlation or least squares correlation), feature-based
matching, relational or global matching approaches, and
whether simple or complex, the real goal of matching is to
utilize a computer to collect a mass of corresponding points
from a stereo pair of images, just like a human operator
observes the stereo pair on a plotter. Once the successful
matching results are available, the reconstruction of the 3-D
object space can be done more easily by using traditional
photogrammetric concepts and techniques. The Recursive
Grid—Dynamic Window (RG-DW) technique introduced in
this paper is such a matching technique which simulates the
function of a human operator to do matching without
involving any other processes. The only difference is that
with the RG-DW technique, computers find matched points
much easier, faster and more cost-effective than human
operators do.
The RG-DW utilizes a series of simple concepts as
elementary tools, such as pyramid images, geometric
transformation, multiple dynamic window size, cross-
correlation and least squares matching to perform a
successful matching for both SPOT and aerial photos. The
RG-DW matching scheme, unlike many other matching
approaches, is free of any ground approximations or image
orientation, and achieves rather high success rates.
Especially, if array processors or parallel computers are
available, it will be easy to implement this technique, and
match hundreds or even thousands of points in a second.
The RG-DW matching scheme was proposed by the
authors at the beginning of 1991 (Lue and Novak 1991). At
813
the ASPRS Annual Convention in March, 1991, in
Baltimore, the first matching results for SPOT images with
the RG-DW technique were exhibited and provoked interest
there. At the GIS/LIS and ASPRS Fall Convention in
October, 1991, in Atlanta full matching progress and
successive products—DEM and orthophotos—were
demonstrated by ERDAS Inc., our commercial partner in
this joint research project, again drew interest. This was
followed by the sophisticated process of multiple
experiments to optimize the software and finally make it
operational and user friendly. MATCH, based on the RG-
DW technique, as the key component of the new commercial
digital photogrammetry module Digital Ortho Module, was
released by ERDAS Inc. at the beginning of this year
(ERDAS, 1992, Lue and Novak, 1992).
In the following chapter, the concept of RG-DW is
briefly outlined, and then some characteristics of RG-DW
are discussed. The final part gives some matching results of
Digital Ortho Module. The images used to get matching
results in this paper include three types: SPOT images,
digitized normal aerial photographs and digital aerial photos
taken by the MapCam System developed by the Center for
Mapping of the Ohio State University (Novak, 1992). The
other parts of digital photogrammetry, such as
aerotriangulation, intersection, DEM interpolation,
orthophoto generation are not described.
2. THE RG-DW MATCHING SCHEME
In the early research stage of image matching, area-based
matching with simple concepts was predominantly used. The
normal area-based matching methods have a rather small
pull-in range, and the required approximations are usually
taken from previous matching results or existing data. As
long as the images to be matched, however, have a good
quality, contain enough information, and are free of water or
clouds, reasonable results can be expected. Otherwise
systematic matching errors or even failure to match may
Occur due to inaccurate or even wrong approximations.
Theoretically, least squares matching (LSM) and its variants
such as multi-point LSM or LSM with geometric constraints,
