The output of the Canny edge detector, composed of a set of non-overlapping contiguous regions covering the whole image, is precisely of the format provided as input to the expert system, constructed by Coherent Research, Inc. in their SmartMaps system. This expert system performs such high-level tasks as object grouping, proximity analysis, Hough transforms, and so on. The output of an RGB clustering and/or neural network can also be structured in such format. Probably the best strategy at this point is to extend this expert system so that it would select the best ultimate separation pattern using a set of trial candidates. A genetic algorithm type philosophy could be used as a guiding technique. Each low-level algorithm is typically successful within a certain image region and it fails for some other regions. A smart split-and-merge approach, consistent with some set of common sense rules, could yield a much better low-level separation result than each individual low-level technique itself. For example, Canny edge detector would offer brown isoclines as good candidates and RGB clustering would offer the green patch as a good candidate for a region. Both propositions would be cross-checked and accepted as reasonable by both algorithms and the final result would contain both types of regions, separated with high fidelity. This type of medium-level geometrical reasoning could then be augmented and enforced by the high-level contextual reasoning within the full map understanding program.