This paper describes the result of a research and development effort focused on developing technologies supporting rapid extraction of simulation databases. The specific goal of the project during this period of time was to significantly reduce the time required to extract features [roads, contours, streams, etc.] from graphic hardcopy sources [i.e, maps and charts]

This problem is significant to overall database construction cost and timelines. Currently, attempts are being made to use large maps scanners and commercial vectorization software to improve extraction efficiency. Unfortunately, the result of the use of only color or intensity to separate objects is that substantial interactive editing of the final product is necessary. This restricts the use of maps as an effective information source.

A new process was defined as a result of this task. It represents an integration of insights gained through the technologies of image processing, pattern recognition and neural network based learning. It represents two kinds of improvement: (1) A reduction in setup time [the operator need only identify typical objects, not define a complete color lookup table] and (2) reduction in interactive editing [by on the order of 90%] due to the higher quality of the output.

Examples are presented of images which illustrate the new process. They show the very significant capability which has resulted. In addition, possibilities for extension of the process to multi-spectral image data are defined.

Gregory DeLozier

Affiliation

Refine Results

Rapid Simulation Database Build using Hardcopy Input