A framework for junction detection using local Hough transforms

A novel method of junction detection in images is investigated. Junctions are modeled using locally applied Hough transforms. Two types of the Hough transform are proposed: (a) 1D transforms to characterize orientations of the junction segments and (b) 2D transforms to additionally estimate lengths of the segments. For any location of the circular window, the best-fit junction can be found and its strength (visual prominence) can be calculated. Then, actual junctions are detected as the local maxima of the strength function. To reduce computational complexity, junction detection and localization is based on the 1D model, while the 2D model can be used to further characterize detected junctions. The paper presents both theoretical issues of the method and its practical limitations. Relations to alternative techniques of junction detection are highlighted. Keywords-component; local features, line junctions, Hough transform, image gradient, interest points