The "brain" of a machine-based object/feature extraction system is the algorithms. To convert a concept of a photo-interpreter that can extract an object with ease into a compilable program, two bottlenecked problems must be solved: (1) the computer language should be high enough for the photo-interpreters to convert their concepts to code without difficulty, and (2) a means for generating error-free code must exist. To resolve these two issues, we propose two enabling technologies using motor vehicles extraction with 1-meter resolution imagery as test cases: (a) pseudo English as a programming language, and (b) an intelligent graphic user interface (GUI) that does not require typing as a programming aid. to perform this feasibility study, the data sets proposed include one-meter resolution space-based imagery, real or simulated, covering 10 metropolitan areas of the US produced by Space Imaging EOSAT. For modeling transportation planning, 3-meter resolution multispectral imagery is proposed for land cover and landaus extraction. The benefits of this project include increased productivity and reduced labor costs resulting from rapid prototyping of a solution algorithm for which programming with low-level languages like C is totally eliminated, and coding errors due to manual typing are reduced to a minimum.Commercial Applications:If successful, we will produce a commercial intelligent software system for generic remote sensing, pattern/object recognition and GIS applications. Federal applications include automatic vectorization of map features in scanned topographic maps, digital orthophotoquads, and NOAA nautical charts. Military applications include automatic target recognition.Key Words:pseudo English programming language, intelligent GUI, object extraction, segementation, motor vehicle extraction, landuse/landcover mapping, one-meter resolution imagery
