Artistic line-drawings retrieval based on the pictorial content

Abstract

In this article, a general framework for the retrieval of artistic line-drawings is introduced. It relies on the pictorial content, defined as a combination of the stylistic content and the visual features of the represented subject. First, we propose an automatic method for the extraction of stroke contours in line drawings, relying on a filtering of the level lines of images. Next, the radius of the drawing tool is estimated from these segmented strokes. This information then efficiently tunes the extraction of several geometric features, including the distribution of curvature, endpoints, junctions and corners of strokes. The efficiency of the proposed method is illustrated with several experiments on two classified databases of artistic line-drawings, and compared with an approach based on the curvature scale space (CSS). Retrieval experiments suggest that the proposed framework is able to handle the pictorial effect delivered by line drawings to a human observer.