Enhancement of Low-Contrast Curvilinear Features in Imagery

A new method is described for enhancing low-contrast curvilinear features in imagery that combines directional filtering with Fischler, Tenenbaum and Wolfʹs F* algorithm for computing minimum cost paths. The method exploits a phenomenon called "the stability of lines over angle." The idea is that when a directionally filtered image contains a line plus noise, minimum cost paths tend to be aligned in the direction of the line with random jumps between parallel paths. When the input image contains noise only, the direction of minimum cost paths resemble random walks with drift. As the direction of the filter changes, minimum cost paths that follow true features persist and are more stable over angle than those that follow noise. Adding them up in an accumulator array over angle produces a larger number of votes along signal paths than along noise paths. This provides a means for enhancing trajectories of low-contrast features. Several examples illustrate the enhancement of forest trails in USGS aerial imagery, linear features on Mars, and roads in synthetic aperture radar imagery