INTEGRATED TECHNIQUE FOR AUTOMATED DIGITIZATION OF RASTER MAPS
Serguei Levachkine and Evgueni Polchkov

(continue...)

1 Introduction

The process of developing an applied Geographical Information System (GIS) can be represented by the following set of steps.

1. Pose of the problem.
2. Selection of the methods for solution of posed the problem.
3. Preparation of the input information, which is required for solution of the posed problem by the selected methods.
4. Definition of the composition and logical structure of the cartographic data.
5. Analysis of existing cartographic information.

5.1 Collection of the vector maps, which can be used in GIS.
5.2 Definition of the vector map layers, which can be directly elaborated by the geodesic data.
5.3 Selection of the paper and vector maps for vectorization.

6. Elaboration of the cartographic base for the GIS.

6.1 Inclusion of existent vector maps in GIS.
6.2 Digitization of new cartographic materials.
6.3 Elaboratio of vector map layers by geodesic data.
6.4 Unification of the vector map layers (conjunction of fragments, unification the cartographic projection, elimination of thelogical contradicted data, etc.).

7. Elaboration of thematic maps.
8. Solution of the posed problem.
9. Publication of the results.

Table 1. Novelty scores of an applied GIS and relative digitization task.

Clearly the work required for each of the stages listed in Table 1 depends on the qualitative and quantative characteristics of the cartographic materials used. These can be combined into a single index; novelty. We classify an applied GIS into one of four novelty levels:

1. Production of a GIS requiring new vector cartographic materials. For the territory under consideration and the desired scale, only paper maps exist.
2. Production of a GIS when basic vector cartographic materials exist for the territory under consideration, but not thematic maps of the desired type.
3. Production of a new GIS when thematic vector cartographic materials have been produced in a previous GIS for the territory under consideration. The new GIS is to be developed under the assumption of some fundamental changes.
4. Updating and development of an existing GIS without essential changes.

Table 1 shows the results of an expert group analysis (28 experts from Mexico, Russia, USA and Europe) of the relative work involved in each of the stages in producing an applied GIS.

These results show that GIS's with novelty level A and B require relatively more work for the digitization of paper and raster maps than the others. In Mexico, topographic and thematic vector base maps have been produced for the entire country with scales 1:1,000,000 to 1:2,000,000. These have been developed thanks to the efforts of the Institute of Gegraphy and Statistics (INEGI), the National Mineral Resource Council (COREMI) and the National Water Resource Council (CNA) as well as other bodies. Current work in the production of GIS is directed towards problems in urban planning, emergency situation monitoring, election enumeration and other government activities requiring maps with scales 1:25,000 and 1:10,000. To carry out these projects, the volume of work involved in the digitization of paper and raster maps is considerable, even when only the largest urban areas of the country are considered. Clearly the development of these problem-oriented GIS can only be completed if the task of vectorizing traditional cartographic materials is automated to the fullest extent possible.

Two main technologies are currently used for map vectorization ([2], [7], and [16]):

The process of digitizing paper maps cannot be automated, hence the authors propose that the only practical approach to produce applied GIS's is the development of new methods and software to vectorize automated raster maps.

Raster map digitization technologies can be divided into four intersecting groups:

In practice, manual raster map digitization methods (D1) coincide with electronic paper map digitization methods (V1). A few examples will serve to illustrate this. In the case of point objects ([7], [16]), the operator visually locates graphic symbols and fixes their coordinates. In the case of linear and polygonal objects ([7],[16]), the operator uses rectilinear segments to approximate curvilinear contours. The manual digitization rate is one to two objects per minute.

Interactive digitization uses special programs which, once the operator indicates the starting point on a line segment, automatically follow the contours of the line. These programs are capable of tracing relatively simple lines. If the program cannot solve a graphical ambiguity on the raster map, it returns a message to the operator. Recently, vector editors capable of carrying out this digitization process have appeared, reducing the time required by a factor of 1.5 to 2. These can be called semiautomatic systems ([7], [3]—[6]).

In theory, automatic digitization vector editors automatically digitize all objects of a given class, leaving the operator to correct errors in the resulting vector layers. The most popular vector editors use this system. However, in practice, the high error level resulting from any small complication in the raster map means that alternative methods must be sought in order to reduce the high volume of manual corrections that are required. It should be noted that the use of increasingly complex methods and algorithms for machine recognition of cartographic objects does not materially improve the results ([7]-[13]).

We suggest that the solution lies in the development of an integrated technique for automated map digitization which uses the basic domination principle: "And he said unto them, 'Render therefore unto Caesar the things which be Caesar's, and unto God the things which be God's.' " In other words, methods and software should be developed which leave to the human operator only those tasks which the computer cannot carry out. In the present case, this implies a detailed analysis of all existing map digitization processes, and the development of software which automates all technological vectorization operations based on formal, heuristic and interactive algorithms which can effectively be used by the computer.

The above thesis is especially suitable for analytical GIS in contrast to register GIS, as the former do not require an extremely high level of geometric exactness in the cartographic materials, whereas they do require fast processing of a large number of vector layers. An example of such a GIS is a GIS developed to solve territorial planning problems, while an example of a register GIS is one developed for a cadastral system.

Application of such an integrated technique greatly enhances the outcome not only of processing, the main stage of cartographic image recognition itself but also of pre-processing (preparation of paper maps and their raster analogues) and post-processing (final processing of the results of automated digitization).

Dirección General de Servicios de Cómputo Académico-UNAM
Ciudad Universitaria, México D.F.