5, ACCURACY CONTROL OF THE CORRELATED HEIGHTS
5.1 Control procedure
For a detailed control of the correlation results two main aspects have
to be considered, First, to achieve a valuable information about the
really attained accuracy all these heights the correlation procedure has
marked as questionable have to be rejected, and second, the comparision
of the results with the reference DTM should be accomplished in separate
classes to reflect the influence of the object content.
The correlation procedure provides three informations to mark a height
determination as doubtful. The correlation coefficient, the y-parallax
and the difference to a second height determination. With an appropriate
selection procedure for the last two criterions /Boochs 1984a/ it is
possible to distinguish between reliable and less reliable heights,
while for the correlation coefficient itself the minimum of r=0.6 serves
as separator between good and bad calculations.
To differentiate in the results for the influence of the object content
three major categories have been established :
- contrast : Depending on the image contrast exteriour influen-
ces like noise, for example are of more or less
importance. Therefore it is interesting to evaluate
if this is reflected in the accuracy.
- geometry : The complexity of the object geometry fixes if the
correlation procedure will be able to approximate
the object surface or if greater deviation during
adaption to the surface are to expect, what should
affect the accuracy too.
- object type : The object type comprises the above mentioned
influences with some additional characteristics like
edges, for example. However, for small scale images
this has to be expected as of minor importance.
The following subclasses are established for the above categories :
- contrast : without affection (WA), hazy (HY), elody (CD)
- geometry : no or low eurvature (LC), steep hills (SH)
- object type : fields (FD), forests (FO), build-up area (BA)
The class (SH) comprises surface regions with a slope of more than 100%.
For the examination of the correlation results each point in the test
sites was given a value of the above type for each category. These
values then served for the classification in the control procedure, so
that only points with the desired characteristics were examinated.
The judgement of the accuracy is based on the values of the average
height difference (AZ) and the mean square residuals (mary), which are
calculated comparing the correlated and the corresponding reference
heights.
5.2 Attained accuracy
Table U4 shows the results of the control procedure.
Examinating the values for mary in test site Deggendorf it first has to
be stated that the object type is of major influence for the quality of
- 88»