This was different for the Indonesian data set where 
sophisticated sensor models in combination with a DEM 
were applied to geocode the images (Cheng et al., 1995). 
The GCP’s used in the polynomial model as well as in 
the sensor model, to improve the model parameters, were 
identified in existing topographic maps at 1:50,000 
scale. This limited the accuracy of the geocoded data to 
the geometric accuracy of the map material used. As a 
result of further investigation the Indonesian data set was 
geocoded to an estimated precision of + 50-100 m. 
4. 3 Multisensor Image Fusion 
The techniques considered in the research presented in 
this paper were pixel based because of the purpose of 
image map production using multisensor VIR and SAR 
data. The fusion of only VIR data was not considered for 
two reasons: 1. The subject has been studied extensively 
by other researchers (e.g. Carper et al., 1990; Chavez et 
al. 1991; Cliche et al. 1985; Franklin and Blodgett, 
1993; Mangolini et al, 1993; Pellemans et al, 1993; 
Rothery and Francis, 1987; Shettigara, 1992; Welch and 
Ehlers, 1987) and 2. VIR image fusion is not very likely 
to solve the cloud cover problem. The actual 
implementation of image fusion techniques consisted of 
colour composites, arithmetic band combinations, IHS 
colour transformations, mosaics and a mixture of 
techniques. It was anticipated to optimise image 
combinations and fusion techniques for the purpose of 
topographic map updating. 
S. RESULTS 
This section summarises the scientific findings. They 
refer to the image fusion environment as well as to the 
advantages and disadvantages of techniques and image 
combinations examined. 
5. 1 Pre-Processing 
All sensor-specific corrections and enhancements of 
image data have to be applied prior to image fusion since 
the techniques refer to sensor-specific effects. After 
image fusion the contribution of each sensor cannot be 
distinguished or quantified in order to be treated 
accordingly. A general rule is to first produce the best 
single sensor geometry and radiometry (geocoding, filter, 
line and edge detection, etc.) and then fuse the images. 
Any spatial enhancement performed prior to image 
fusion is of benefit to the resulting fused image. An 
advantage is the possibility of filtering and enhancing 
the data during the geocoding process to avoid multiple 
resampling. The data has to be resampled to the pixel 
spacing required for the desired image fusion. 
The type of radiometric enhancement required depends 
on the nature of the area being studied. An example is 
the high pass filtering, successfully implemented in the 
processing of the Dutch data set. However, it led to poor 
results in the case of Bengkulu in Indonesia due to the 
extreme variations in elevation and the ground cover. 
The importance of geometric accuracy to avoid artefacts 
and misinterpretation should not be underestimated. 
Pixels registered to each other should refer to the same 
object on the ground. This implies that the data should 
be geocoded with an accuracy of < 1 pixel. Therefore the 
DEM plays an important role in this process. As found 
for the Indonesian example, the SAR data still contains 
shifts of one or more pixels as a result of the poor DEM 
and map quality. The need for DEM's of high quality 
and appropriate grid spacing is therefore evident. 
S. 2 Techniques and image combinations 
The following list of statements categorised by technique 
conclude the findings of the research discussed. 
RGB 
The following list provides a summary of the main 
features of image fusion using the RGB method: 
O Digital numbers in single images influence the 
colours in the RGB composite. This implies the 
following considerations: 
1. Histogram stretching of single channels 
influences the visibility of features in the 
final colour composite. 
2. Inversion of image channels might be 
desirable to assign colour to features. 
3. RGB channel assignment significantly 
influences the visibility of features and 
visual interpretation by a human interpreter 
(blue = water, green = land, etc.). 
C The technique is simple and does not require CPU 
time-intensive computations. 
CO RGB overlay prevents the contribution of the optical 
imagery from being greatly affected by speckle from 
SAR. 
Band Combinations 
The following conclusions were drawn by visual 
interpretation from the image fusion using band 
combinations: 
C) The fusion of ERS-1 SAR with SPOT PAN data 
improves the interpretation of the SAR data. 
C) Subsequently, it helps applications that benefit from 
the interpretation of up-to-date SAR data (urban 
growth, coastal zone monitoring, tidal activities, soil 
moisture studies, etc.). 
C) The resulting fused image depends very much on the 
appearance and content of the SAR data. 
C As a result, the SAR data have to be selected 
according to the interest of the application. 
658 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996