3.3 OPERATIONS BETWEEN VIEWS 
Some operations must inevitably act between views, forcing view transformation of some sort. As an example, 
consider the case of retrieving all forests that are co-incident with a line. The line is a geometric primitive, whereas 
forests are features. The operation demands access to the geometry of the features, so a geometric view onto the 
forest objects must be provided. The projection of data between views is achieved by extending the concept of a 
field, from the field view. Instead of restricting a field to refer to a point in space, we allow it to represent any 
arbitrary geometry. So the shape of any class or feature may be projected into the field view and associated with the 
co-incident field data. This is fully described in Part Two. 
One of the difficulties in formulating a model describing the transformation of geographic data is the complexity of 
the tasks involved, particularly with regard to object formation. Figure 2 shows some of the transformations that 
must be supported for the integrated analysis of image data within GIS. These transformations not only change the 
dataset, but also involve the migration of the dataset from one data model into another. 
Figure 2: State transition diagram for geographic data 
The classification operation labels the primary (image) data to give thematic data in the form of overlays. From 
these overlays, geographic features may be formed by some kind of labelling procedure. Alternatively, primary data 
may be transformed directly to features via a process of feature extraction, using a combination of spectral and 
structural evidence. In the reverse direction, geographic features can be transformed into classified data by 
aggregation (by class) to produce an overlay. Classes and features may be interpolated or extrapolated to produce an 
estimated primary data source, hence projected back to the image or field view. Where the classes and features are 
derived from primary data, then their reduction back to primary data is error free, since it represents the inverse of 
their formation. Where the classes and features have been supplied ready made, then their projection to field or 
image data may involve a great deal of uncertainty. 
In forming an overlay from an image or extracting a feature from an overlay, it is entirely possible that the spatial 
data type may change as a result. For example the application of a crack edge detector (e.g. Prager, 1980) 
effectively converts image data into an overlay in vector form consisting of a series of coordinate pairs with an 
associated edge strength. Similarly, a line (or boundary) following algorithm will change a raster overlay into a 
series of vectors labelled with an object identifier (e.g. Kass et al. 1987). To re-iterate, changes in data 
representation should be invisible to the user.