Recognising characteristic elements through the analysis of curves on 3D models

We propose a new method for recognising characteristic curves on 3D shapes, identified by a set of characteristic points. We approximate these curves with known curves, e.g., spirals, their patterns or aggregations, to provide a localisation and quantitative measurement of style features like decorations, visual motifs or anatomical features on the digital models of 3D objects. To solve this problem, we adopt a generalisation of the Hough Transform (HT) which is able to deal with curves represented either in implicit and parametric form and extends the set of curves so far adopted for curve recognition with HT. In addition, we introduce new rules of composition and aggregation of characteristic curves into patterns or decorations, not limiting the recognition to a single curve at a time. Besides planar curves, our method yields the recognition of spatial curves (see Figure 1): to the best of our knowledge, this is the first attempt to apply the HT to the recognition of spatial curves without any projection onto a fitting plane.