205 
GEODATABASE CENTRIC ORTHOIMAGE PRODUCTION 
USING ARCGIS IMAGE SERVER 
P. Becker 3 
d ESRI, 380 New York St., Redlands, CA, 92373, USA - pbecker@esri.com 
Commission IV, WG IV/2 
KEY WORDS: Imagery, Management, Production, Orthoimage, Parameters, Metadata, Distributed, Online 
ABSTRACT: 
In traditional image production workflows, imagery is processed in multiple steps, with the imagery being sampled multiple times, 
and being read and written to disk multiple times. This reduces image quality and production efficiency. The linear workflows also 
are susceptible to bottlenecks when a single parameter in the workflow is not available or needs to be changed. ArcGIS Image server 
enables a geodatabase centric workflow by which the parameters and models for processing imagery are stored within a database 
and the imagery product is generated on demand as required directly from the base imagery. This methodology provide a number of 
advantages. For example, it enables serving of dynamic image services that can be updated as revised parameters become available. 
It enables the creation of graded products that change over time with the revised parameters and enabling improvements in quality 
assurance processes. The same image services can also be used to generate imagery products in the form of caches or the traditional 
tiled images that are required by most orthoimage mapping projects. The quality of the imagery is superior due to reduced sampling 
of the images. By optimizing the processing and reducing disk access, the production system is very efficient and scalable, enabling 
high production rates. 
1. TRADITIONAL ORTHOIMAGE PRODUCTION 
Traditional orthoimage production workflows, result in imagery 
being processed in multiple steps. Typically, an image will go 
through the following steps: radiometric correction, pan- 
sharpening, aerial triangulation, orthorectification, mosaicking, 
reprojection, then product generation. Depending on the 
software used, the order of these steps may change, but 
typically each step is performed separately with imagery being 
read, sampled, or enhanced and then written to another location. 
With each radiometric or geometric enhancement some 
information is lost, resulting in the final image quality being 
non-optimum. Even processes, such as mosaicking multiple 
images together, result in unnecessary sampling if pixels of all 
the input and output images are not aligned. Additionally, if the 
sampling of the input and output imagery is a very similar 
resolution (which is often the case when re-projecting) then 
aliasing artefacts can also become apparent especially in 
imagery covering areas that are near featureless, but have good 
textures such as over water or gravel desert. In some areas the 
imagery appears to have slightly higher contrast than in other 
areas. These artefacts are caused by pixels in some regions 
being sampled to be close to the average of four neighbouring 
pixels (which reduces the local contrast), while in other areas 
the pixel is the nearly solely derived from a single input pixel 
(maintaining the local contrast). Although different sampling 
methods may be defined to reduce these effects, they are 
generally at the cost of accuracy. To reduce the data volumes 
some workflows apply lossy compression methods to the 
intermediate products, which further increases the creation of 
artifacts and degrades quality. 
Such orthoimage generation workflows where the imagery is 
read and written multiple times are also not truly scalable. 
Traditionally, such workflows are scaled using technologies 
such as CORBA, enabling distributed processing over multiple 
machines. The processing gains are quickly mitigated by 
network and harddisk bottlenecks, which are caused by the 
multiple reads and writes of large image data volumes to disks 
saturating the available bandwidth as well as fragmenting the 
disks. 
1.1 Why do we need a geodatabase centric approach? 
Linear process workflows can also cause substantial bottlenecks 
in production. A delay in one step will stop subsequent steps. 
For example, if the accurate orthoimage product is not created 
due to the non-availability of a terrain model, the color 
balancing cannot be performed, since color balancing in such 
workflows are dependant on the orthoimages. As projects 
become larger, the chances of one part of the process being 
delayed increases which in turn increase the delay and risks for 
the complete project. 
Such production workflows can be considered a set of 
independent tasks and not all tasks need to be performed 
necessarily in a fixed order. Each of the production steps can 
be considered to consist of two components: The determination 
of process parameters, and/or the application of these 
parameters on the imagery. The process of aerial triangulation 
is a typical example of parameter determination with no pixel 
processing being applied. Orthorectification is a process that 
utilizes the parameters of orientation and a terrain model to 
apply a pixel process. During the complete image production 
workflow, there are many parameters that affect the resulting 
product. These include parameters of orientation, radiometric 
enhancement, pan-sharpening, terrain models, and mosaic 
seamlines. These production stages are actually quite loosely 
coupled. For example, the parameters of pan-sharpening have 
no effect on aerial triangulation. The determination of color 
balancing parameters for imagery is often dependant on the 
orthoimages, but does not require accurate orthoimages nor