Map generalisation is the process of reducing the scale and complexity of map detail whilst maintaining the important elements and characteristics of the location. When creating a map using traditional manual techniques, a cartographer aims to achieve a balance between the amount of real-world information required to make the map useful and avoiding confusion for the user. This is a time-consuming and expensive process.
GIS has led to the realisation that the efficiency of the cartographer could be increased through the automation of some of the more time-consuming techniques such as line and polygon simplification. Current off-the-shelf GIS software packages contain tools that allow basic generalisation to be performed.
Probably the most famous line generalisation algorithm was developed by Douglas and Peucker in 1973. The Douglas-Peucker algorithm simply filters the number of vertices along a digitised line to create a representation suitable for the specified depiction scale.
Although these algorithms go some way to help in the automated production of smaller-scale maps, generalisation technologies are very much in their infancy. The challenge of replacing an experienced cartographer with a computer that can make the same decisions to produce a map is significant.
The main problem that needs to be addressed in generalisation is how to resolve the conflict between different map features when they are displayed at smaller scales. As there is not enough space to display all of the information in an uncluttered manner, methods to typify the data in an intelligent, consistent and coherent way at smaller scales need to be developed. It is for this reason that the move from map generalisation as a manual art to a computerised scientific process is a distant dream. For the foreseeable future the process will be a semi-automated collaboration between cartographer and machine.
It is difficult to place text on a map so that it is both legible and clearly associated with the feature that it is annotating. Text placement refers to the complex challenge of achieving this in an efficient manner to generate high-quality results. Text can of course be placed by simple manual methods, although this is a time-consuming and inconsistent process. The automatic generation and placement of text can result in savings in time and labour together with a more repeatable result. Although seemingly simple in concept, this automatic process is remarkably difficult to achieve in practice and is the subject of widespread research interest.The manner in which text is placed on a map depends largely upon the cartographic symbols that are chosen to represent the points, lines and polygons of the source data and how they relate to each other in a spatial context. One layer’s text or symbols may have a dramatic impact in determining the placement of another layer’s text. It is therefore necessary to assemble all the data layers required within the final map, then symbolise their features according to the map’s extent and scale before text placement takes place. Modern automatic text placement software offers flexible placement options. There are now choices of font styles, sizes and colours; preferred location of a piece of text; weightings as to which text is more important and takes preference over other text (for example, road text might be more important than building text); and minimum and maximum allowable distances between different labels. A predetermined set of rules can therefore be created applying to any source data for any location at any given scale. This results in a map product that is generated more consistently and also more efficiently, thereby greatly reducing the amount of manual effort required in its production. See below for an example of automatically generated text.
When we talk of automated cartography, what we are trying to achieve is a fundamental GIS goal of capture once, use many times. In other words, it’s inefficient to go through the process of manually creating an aesthetically pleasing map every time something changes. It’s far more desirable to automatically represent and display the source data as often as required and in an infinite variety of ways. Modern GIS software can be used to rapidly and efficiently generate highly complex maps from basic point, line and polygon features.
Automated cartography can become a highly sophisticated business. In addition to overcoming the problems of scale differences and placing text appropriately, we may also wish to generate different kinds of map for different users. For instance, to create a map oriented in the direction in which someone is travelling or with colours that aren’t affected by one person’s colour blindness.
Electronic data and electronic displays enable new forms of cartography to be developed. For instance, standard data formats such as Virtual Reality Mark-up Language (VRML) allow maps to become virtual three-dimensional worlds that you can explore as if you were flying through the landscape. Furthermore, geographical data is not necessarily best represented as a map. In many cases we are more interested in direct information, such as navigation instructions, which might be delivered as text or as synthesised voice. Where will this lead? Smelly and tasty GIS?
Data from imagery
Imagery – usually from aerial or satellite sensors – is widely used on GIS platforms as a backdrop to vector mapping. An image may contain an abundance of visual information that is not conveyed by the points, lines and polygons of a vector map. As far as GIS software is concerned, however, an image is a dumb background. A key research challenge is to derive vector objects from imagery. An image is a raster dataset: it is a grid of squares or pixels. Each pixel has a numerical value that may relate to colour, height or indeed virtually anything measurable.
Human interpretation is often used to derive data from imagery. An operator traces lines over the on-screen image in a technique known as heads-up digitising. This process remains very labour intensive, however, and significant efforts are being made to find ways to automate it.
One approach is to look for abrupt changes or discontinuities in the image that will equate to a line feature in a map. This can be achieved using an edge detection algorithm that applies a mathematical function to each pixel and its immediate neighbours in turn. The result is an image of lines that can simply be converted into vectors. These lines are often very messy, however, and this method is best used where the discontinuity itself is distinct and separate. An alternative approach is to use software to look for similar clusters of pixels and thereby classify the image into distinct areas. Where successful, this will identify real objects such as buildings, fields and bodies of water within a classified image, which may then be converted into a vector map. Accurately and appropriately deriving vector data in this way is a complex activity at the forefront of research. The ability to automatically generate a map from an aerial photograph or satellite image is a holy grail of GIS because it would help make data far more inexpensive and up to date.