The major expense in establishing a geographic information system (GIS) is converting paper maps to digital. This paper describes a system that scans and semi-automatically digitizes Italian Land Register Authority maps, allowing their replacement by a GIS. Because of the accuracy required, manual conversion or tracing was estimated at 20 hours per map; the system described cuts this to half an hour of automated processing and three hours of interactive processing, for a total saving of 5.1 million person-hours over the 300,000 sheets.
The system works in four phases. First, a map is scanned and the lines found are converted to a graph structure of edges and nodes. Second, the graph representation is searched for shading (multiple parallel lines), continuous lines, and dashed lines. Third, the line candidates are vectorized as lines or symbols. Fourth, high-level entities are recognized. Testing indicates that over 90 percent of map features are correctly identified automatically, and operator intervention deals with the remainder.
The paper clearly describes the approach and high-level design of the system, but does not give algorithm details; it includes no general discussion of the problem of automatic recognition or consideration of alternative approaches. Judging by the example figures in the paper, Italian Land Register maps are relatively clean drawings with uncomplicated semantics. This system appears to be a major success, however, and will be of interest to other government instrumentalities and utilities with large backlogs of maps to digitize.