Imern 
rinzip 
n es 
1 die 
sticks 
sowie 
mmern 
lungs- 
ehler 
durch 
ldung 
det, 
nkte 
iden 
Oder 
ind, 
irch 
INPUT PROCEDURES FOR TM-LANDSAT PHOTOGRAPHIC PRODUCTS INTO A GIS ENVIRONMENT 
Julio Cesar Lima d'Alge 
Diogenes Salas Aives 
Image Processing Division - DPI 
National Institute for Space Research - INPE 
Brazil 
ISPRS Comission No. IV 
ABSTRACT: 
Visual interpretation of high resolution images is an important tool because 
digital processing techniques do not completely replace the human capacity of 
understanding spatial features over an image. System corrected images present 
internal geometries which are compatible to medium scale works, but 
photographic processing can introduce small scale changes for the whole image. 
Therefore, a sequence of heuristic local 
adjustments is normally carried out 
by the interpreters with the purpose of transferring image information into 
the topographic map which represents the GIS database. As a consequence, the 
resulting thematic database is badly georeferenced. This work presents two 
alternatives which were used for the georeferencing of TM-LANDSAT photographic 
products over Brazilian Amazonia intended to detect and control deforestation 
process. 
KEY WORDS: GIS/LIS, Image Interpretation, 
1, INTRODUCTION 
1.1 Geodetic accuracy of TM-LANDSAT 
imagery 
System corrected TM-LANDSAT pictures 
still represent basic products for 
various environmental research projects. 
Although fully corrected images can be 
preferable, they are not so widely used 
for visual interpretation, due to their 
processing costs. Instead of using them, 
it is less expensive and more common to 
use system corrected ones. Many studies 
were presented in the last two decades 
showing geometric correction procedures 
and the results of geometric evaluations 
over LANDSAT images. 
  
Welch, Jordan and Ehlers (1985) presented 
a geometric evaluation of system 
corrected digital TM data. Internal 
accuracy of around 30 m was obtained 
using ground control points extracted 
from large scale maps. They emphasized 
the need of having image coordinates of 
control points within 1 pixel of 
accuracy. 
Machado e Silva e d'Alge (1986) showed 
slightly different results --40 m - for 
the internal ^ accuracy in a similar 
geometric evaluation procedure over 
1:1,000,000 TM diapositives. This 
discrepancy could be explained by less 
accurate image coordinates. It was not 
possible to ensure 1 pixel of accuracy 
for the control points determined over 
that analogical product. 
Similar geometric evaluation reports were 
presented by Borgeson, Batson and Kiefer 
(1985), and by d'Alge (1987). All these 
Works comment, in some extent, the 
influence of control points accuracy in 
geometric correction models and geometric 
701 
Image Matching 
evaluation results, and also agree that 
1:100,000 must be considered the largest 
Scale for the use of these images. 
Besides this fact, there is an important 
Source of errors that can easily affect 
the geometry of a photographic product. 
Successive enlargements performed by 
precise photographic equipment must not 
introduce local distortions, but. can 
originate a general scale discrepancy. 
Therefore, the photointerpreter will 
probably deal with some small matching 
problems while trying to register 
perfectly the whole Scene over a 
cartographic base. For instance, the base 
is: a 1:250,000.. scale map. and . the 
interpreter has a 1:248,600 scale image. 
This work assumes this hypothesis, 
discusses the conventional approach for 
visual interpretation, and presents 
alternatives aiming to an accurate input 
of image content into a GIS database. 
1.2 Conventional photointerpretation 
procedures 
  
It is convenient to explain what is meant 
here as conventional photointerpretation. 
Photointerpretation means a set of 
procedures that represent the human. 
capacity of understanding and extracting 
features from an analogical image. The 
term "conventional" carries the meaning 
of absence of GIS or Computer Cartography 
technologies. In other words: a 
cartographer receives an analogical 
overlay with image features and has the 
task of drawing a map manually. 
A typical working session is assumed as 
an example. Suppose a Forest Engineer is 
interested in mapping the extension of 
deforestation in a Park which appears 
entirely in a 1:250,000 topographic map, 
and consider the existence of a nominal