ts 
er 
th 
~ 
J 
et 
Q 
of 
be seen from publications of results by IGN-France 
(Villet, Leconte, 1995). 
3.2 Classification accuracy 
No significant misclassifications may show up, unless 
larger scale photographs are used. The analysis is 
based on themes like roads, buildings, vegetation; for 
each theme a separate misclassification matrix will be 
created. 
Here also some statistical testing procedures can be 
applied. 
a, : (main diagonal): correctly classified features 
a, : (line elements - main diagonal): commission 
errors, (features erroneously included into a class) 
a, : (column elements - main diagonal) : omission 
errors (features erroneously excluded from a class) 
3.3 Completeness 
Checking for completeness and classification goes 
generally in parallel. The same strategy can be 
applied: visual inspection according to object classes 
using superimposition; however the results must be 
presented separately. 
4. QUALITY REPORTING 
4.1 Introduction 
Quality information collected through the various 
processes need to be filtered and properly stored for 
easy access and retrieval. There are a number of 
unanswered questions about quality reporting : 
how much information is required by the user ? 
how to structure and store it ? 
how to display and present it ? 
Various attempts for modelling and storing quality 
information of spatial data can be found in the 
literature. (Faiz, Boursier, 1994) 
4.2 Multi-level approach 
The management of quality information is seen from 
a different perspective by the supplier and by the 
user of data. This may lead to separate quality 
models, one which is mainly process-oriented (for the 
supplier) and one which is data-oriented (for the 
user). This implies that only part of the quality 
information has to be transferred to the database. 
The basic units for feature extraction are models 
which will be dissolved in layers of a database; 
quality information can best be organized in a 
separate quality layer, using a multi-level approach 
(figure 3). 
15 
DATABASE M.D. 
PROJECT M.D. 
6 
LAYER M.D. 
A 
CLASS Attributes 
A 
OBJECT Attributes 
(M.D. = Meta Data) 
Figure 3: Quality information in a multi-level 
approach. 
Quality information is somewhat heterogeneous and 
therefore has to be organized partly in the form of 
meta-data and partly in the form of attributes. 
This information is to be stored at various levels : 
. Project : data source 
camera type 
photo scale 
date of flight mission 
determination of GCPs 
method of AT 
method of data collection 
. Layer : lineage (data source, 
method) 
logical consistency (after editing) 
positional accuracy (after field check) 
classification accuracy (after field 
check) 
temporal accuracy 
. Class: model-lay out and model-ID 
control points 
data from field check and field 
completion 
accuracy (estimated) 
. Object : reliability 
4.3 Visualization of quality information. 
Quality information can be displayed, graphically or 
numerically. The more quality attributes that are 
added at the object level, the more quality analysis 
can be performed based on quality criteria. 
Typical questions which may be of interest for the 
user are the following: 
which models cover a certain area of interest 
(defined by a window) and what is the photo scale 
and the date of the flight mission ? 
in which areas has field check/completion taken 
place ? 
what is the positional accuracy (in planimetry) of 
a class of objects ? 
This type of question can easily be answered through 
queries. 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996