Boundary extraction of linear features using dual paths through gradient profiles

An algorithm for automated extraction of linear feature boundaries in 2D images is presented. From a marker set approximating the medial axis, we generate 1D gradient profiles orthogonal to linear features. The algorithm uses dual shortest paths through an image generated from gradient profiles to extract boundaries. The algorithm offers an alternative to a watershed type approach and performs well on images with noise and areas of low contrast. We present the results of our algorithm on microscopy images of neurite outgrowth and other images containing linear features.