NETLines: Recovering line-networks via gradient-based line segments refinement
Xiaohu Lu, Jian Yao, Kai Li, Li Li, Kao Zhang, Jinge Tu · 2015
In this paper, we propose a novel line segment detector, named as NETLines, which can produce a set of accurate line segments and a set of node-connected line-networks formed by connection of the line segments and the image boundary. Based on the line segments generated by other line segment detectors (e.g., EDLines [1]) on an edge map, the proposed algorithm efficiently makes use of the gradient map of the original image instead of its edge map to extend and refine the line segments. The line-networks are constructed with the extended and refined line segments and a set of nodes generated by connecting of the line segments. Furthermore, the line segments and line-networks are optimally supplemented and refined by linking and merging the line segments. Experimental results on a set of natural images illustrate the proposed NETLines produces more accurate and complete line segments compared with state-of-the-art line segments detectors.