A robust structure extraction method based on image edge

In this paper, we propose a novel approach with desirable universality and expansibility for structural feature extraction. Structural feature is a high-level 2D feature that delivers information including image components and spatial relations of them. In the proposed approach, we first trace edges to line segments, fit line segments to longer lines and extract parallel pairs. Then, we extract contours of image components by grouping lines and parallel pairs. Lastly, we reveal the salient spatial relations of components by theorems and judging rules. The theorems, which are concluded from the step of components extraction, facilitate the following analytical procedure. Experimental results tested on 1960 real images demonstrate that our approach can extract integrated contours of image components and analyze the spatial relations between them in desirable time consuming, and the approach is suitable for many application fields.