Linear feature extraction based on an AR model edge detector

Extraction of linear features is an important task in many image understanding and computer vision systems. In this paper, we describe a linear feature extracor which operates on the edge pixels detected by a space varying 2 - D autoregressive (AR) model. The sequence of steps for extraction of linear segments consists of tracing the edge points, linking and fitting with lines. The heuristics used in tracing and linking steps are specific to the characteristics of the AR model based edge detector. The performance of the linear feature extractor is illustrated using real images.