levation
DTMs),
uctivity
11 scale
tomatic
ation is
rofiling
| to the
studies.
uired at
apan or
e of the
h; with
‘ategies,
enough
1996).
ess and
. These
tomated
he bare
1 added
ditches
Trors in
sting an
levation
calling
several
e job or
Multiple
> or less
) decide
ceed to
pyramid
em uses
1atically
kes the
d terrain
nd more
system,
nbers of
ge areas
Hon and
er while
intuition
k usage.
d higher
stems 18
yr image
forward,
provided a DTM is available, customers have had a difficult
time producing products that are of top quality in a minimum
amount of time and steps. Some of the typical problems that
often must be addressed are:
1. The images to be mosaicked have "hot spots" that are often
caused by the sun angle to camera/sensor relationship.
2. The images are of poor quality and/or they are very
different from one another for mosaicking.
3. Systematic effects such as hot spots and vignetting make
simple mosaicking unsatisfactory.
4. Multiple processing steps such as orthophoto generation
and then mosaicking result in multiple resampling and too
much user interaction.
5. Multiple steps result in large disk usage, low productivity,
and are unfriendly.
5.1 One-Step Image Maps
To improve this situation, several steps have been taken to
exploit automation and simplification for the user to arrive at a
more cost effective process. To this end, a one-step process for
the production of the image maps is being made available. This
process can start at various points in the typical production flow
and continue to its completion. This process can start as early
as before image triangulation. In this case the triangulation,
DTM production, radiometric balancing, seamline generation,
orthophoto correction, and seamline feathering are performed in
a single batch process.
5.2 Radiometric Processing
As mentioned previously, one of the biggest time wasters in
image map generation and the cause of poor aesthetic quality is
acute radiometric problems such as hot spots and vignetting.
Interactively setting up differing tonal curves for each image in
a mosaic can also be a big time waster. It has been typical for a
user to adjust each image while viewing the other images to
obtain a "visual" balance. This can be very frustrating and is
seldom optimal. Automation can greatly reduce these problems.
One method that is being used to fix the majority of these
problems is to adaptively correct each image to a common
standard. This is done by first characterizing each portion of
each image in terms of its radiometry. Care must be taken in
the algorithms so as not to allow "artifacts" in the digital image
to reek havoc in the process. For example, lakes and the film
border must not be allowed to drive the algorithms. Once each
part of each input image has been characterized a common
radiometric goal is automatically computed or user entered.
The radiometric process in mosaicking can then alter each
portion of the output image in a direction toward the goal. This
is performed on single image based orthophotos as well as
mosaics. The result of this processing is that the seams or
differences between map sheets or orthophotos are drastically
reduced. This permits a much better result and makes seamlines
of very high quality.
53 Seaming Methods
Depending on several factors such as image scale, image
content, and project requirements, it may be important to have
a variety of methods for mosaicking the images together. In
Some cases, only the user can define the seam line because all
buildings and trees must be avoided for accuracy and aesthetic
l'asons. To increase productivity, seam lines can be extracted
253
as vector polygons in the original images in advance of the
mosaic production. Many polygons can be extracted and edited
to cover the desired "orthophoto sheets". Once these are
extracted, the "sheets" can be created in batch processes
without further user interaction. This includes the radiometric,
and seam feathering processes.
For productivity, we would like to avoid having the user
manually draw seamlines. This has been accomplished in two
ways. In one method, the seams between images are determined
automatically. This is accomplished by using a "cost" function
that examines the overlap regions, radiometric characteristics,
and other factors to derive a seam line automatically. This
method attempts to avoid buildings and picks a path which is
usually well hidden.
In a second method, the input images are seamed at the
transition from one image to another based on the most nadir
sensor view. This also permits an automatic seam line and at
the same time reduces errors due to relief displacement that is
not correctly modeled by the input DTM. Work is continuing,
however, to introduce an editing facility to handle the situation
where the user wishes to change a seamline which the
automatic system has placed injudiciously, for example through
a building.
6.0 VECTOR EXTRACTION
Perhaps the most studied and researched area of automation is
that of extracting vector data from images. Yet, to our
knowledge, in practice this is one of the least automated areas
of photogrammetric work. Unfortunately, it is probably the
largest time user in practical production. After many years of
our own research in this area and following the work of others,
we are starting to see some practical semi-automated operations
reduce the production time line. Completely automated
operations continue to develope very slowly.
6.1 Semi-Automation in Vector Extraction
Our main emphasis for vector extraction has been to develop
semi-automated tools to be used under user direction and which
operate in near-real-time. New tools for the "refinement" of
points and lines that are quickly measured by the user have
been a major area of development. These tools take a "quick
and dirty" measurement from the user and refine it using a
variety of image, geometric, and photogrammetric processes.
Figure 1 illustrates the types of points or "seeds" the user places
for a building. The user does not worry about being precise or
about squaring, thus user fatigue is reduced and extraction
speed is improved.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B2. Vienna 1996
