Reliable image straight-line extraction based on the integration of intensity discontinuity and edge continuity

Abstract

This paper describes general framework of extraction of image straight-line segments based on the integration of intensity discontinuity and edge continuity in digital image. First, multiple "edge contours" are traced out from digital image by implementing classic LOG algorithm and contour tracing principle. In this step, Zero-Cross points are detected without implementing traditional 'edge labeling' procedure and thus, the continuity between adjacent edge points are maintained for the convenience of edge linking and can take advantage of contour tracing principle; After that, Houghbased heuristic linking algorithm is developed to extract initial straight-line segments from "edge contour" respectively. In this step, important prior knowledge, e.g., "seed" point and direction of straight-line segments, is deduced from Hough transformation and based on the knowledge, possible straight-line segments contained in each "edge contour" are extracted through heuristic linking via adjacent continuity again; Finally, initial straight-line segments are testified and improved with Least Squares Template Matching (LSTM) technique by estimating the similarity of intensity distribution between initial straight-line segment and its ideal intensity template. The experiment of extracting straight-line segments from aerial or ground image respectively, for building 3D reconstruction demonstrates that the proposed approach is robust and effective. Contributions are concluded as well.