Automatic General Purpose Linear Feature Extractor for Digital MultispectralImagery

摘要

A parallel general purpose feature detection method is described based on signalstrength (bright thresholding) and connectedness.The method is based upon the article Automatic road identification and labeling in Landsat 4 TM images. Conceptually, it is useful to think in terms of road identification in an image, with the identifying other cultural features such as airport runways, docks, jetties, and railway lines. Although many road detection methods are available, few work on satellite imagery in which most roads are only one or two pixels wide; the same can be said about other cultural features in such imagery. The technique consists of several phases: image sharpening which determines a most likely road direction a magnitude for each pixel; seed selection determining those pixels most likely to be in a feature; a line following technique. A multispectral approach and a slightly improved line following technique are employed to obtain an improvement over the algorithm described.