Technical SectionExtracting a polyhedron from a single-view sketch: Topological construction of a wireframe sketch with minimal hidden elements
Introduction
The subject of this paper is related to 3D geometric modeling and to sketch-based CAD, with an emphasis on natural (or partial-view) sketch (Fig. 1(a)), i.e., a sketch without any hidden lines. More specifically, this work deals with the problem of automatically constructing a “polyhedron from a single natural sketch (PSS)”. The principal sub-problem of PSS is topological construction of a “wireframe sketch from a single natural sketch (WSS)”, for which a solution is presented here.
Regarding published research on the PSS/WSS problem, one observes the following: numerous papers have appeared from 1980 until today on this problem, offering various solutions; see, e.g., [1], [2], [3], [4], [5], [6], [7], [8], [9] and references therein. Despite that, it is fair to say that existing solutions are far from satisfactory, as, even in 2008, new methods are appearing (see, e.g., [10] and references therein) for the plainest case of the PSS/WSS problem, where the polyhedron (to be constructed) has only planar faces and is trihedral. Current methods view PSS/WSS as “image-processing” or “computer vision” problems, where emphasis is placed at analyzing the given input (sketch) using tools and techniques lacking a robust mathematical foundation. More specifically, many published methods are based on the “line-labeling (LL) methodology” (initiated in [11], [12], [13]), which tries to associate each sketch line to a “label” (with possible values=“convex”, “concave”, “occluding”) so that the whole set of labels is compatible with a predetermined set of “labeling rules”. Numerous authors have explored this idea (see [5], [14], [15], [16], [17], [18] and references therein). After presenting, in a series of papers (see references in [17]) improved versions of the methods in [5], Varley et al. conclude [17] that LL has very little to offer towards solution of the PSS/WSS problem.
It must be emphasized that the predominant “image-processing”/“computer vision” methodologies not only place focus on poor heuristics like LL, but also they clearly place PSS/WSS out of the correct context, which obviously is “graph theory” (for analyzing the sketches) and “3D solid modeling” (for constructing the corresponding polyhedron). Recently, Cao et al. [10] published a paper on the PSS problem that indeed moves away from the classical “image-processing” methodology and adopts “graph theory” to solve the WSS problem. The proposed method includes two main steps: (a) construction of an initial hidden structure, and (b) reduction of this structure to the “most plausible one” according to “human visual perception” (these two main steps correspond to Steps 3 and 4 in the description given in Section 5.1. of [10]). Unfortunately, this method is far from being complete as both steps (a) and (b) are based on heuristic criteria/processes lacking a solid justification, e.g., realization of step (b) is solely based on the heuristic criterion “the human visual system … tends to interpret a figure in such a way as to produce an object that is as symmetrical as possible” (see Section 5.4 in [10]). This approach is problematic as it cannot handle objects that are far from symmetric; indeed the test sketches used in [10] are either symmetric or “almost symmetric”.
The present research aims at producing a PSS algorithm that is free from the shortcomings of the “image-processing” methodology as well as of a technique like [10] based solely on “graph theory”. Indeed, an improved WSS algorithm is presented here employing robust tools from graph theory, 3D solid modeling and Euclidean geometry. A detailed topological analysis of sketches is given, followed by an efficient technique to complement a given “visible sketch” with appropriate hidden parts. The scope of the present research covers manifold polyhedra without holes, which are also trihedral. The employed sketch is a natural sketch, where:
- 1.
No element of the natural sketch causes two or more visible regions of it to correspond to one visible region in the wireframe sketch to be constructed.
- 2.
At most one T-junction (: this is defined in Section 2) exists in each region of the sketch.
The proposed methodology aims at producing a polyhedron, which is CAD-usable, i.e., a valid 3D solid model. This objective, combined with the fact that no information is available regarding the hidden part of the polyhedron, leads to the conclusion that this hidden part should be minimal (e.g., a single planar face) and at the same time sufficient to define a valid solid model. This “minimal-completion strategy” implies that from the given natural sketch a wireframe sketch should be derived where the number of hidden lines/junctions/regions is as small as possible.
Section snippets
Geometric modeling of sketches and solids with an emphasis on topological description
A sketch is a set of straight lines on a plane that intersect at junctions (i.e., points). In current research [5], [8], [10], [19] a sketch is considered to depict an orthographic projection of a manifold trihedral solid (each vertex of the solid belongs to exactly three faces) with planar faces. The solid (polyhedron) is considered to be in “general position” with respect to the given projection plane, i.e., no face or edge of the solid is perpendicular to that plane. Adjacent faces (edges)
Algorithm PSS: an outline
The algorithm proposed below solves the problem “polyhedron from a single natural sketch” (PSS Algorithm) and is divided into three parts (Fig. 3):
Part A: Topological construction of a wireframe sketch.
Part B: Geometric definition of the wireframe sketch.
Part C: Construction of the polyhedron.
In Part A, we develop a method deriving a “minimal wireframe sketch”, i.e., a valid wireframe sketch that is guaranteed to contain exactly the minimum number of hidden elements. This part is based on (a)
Determination of hidden regions
Hidden regions in a minimal wireframe sketch will consist of both hidden and visible lines and junctions. We define hidden regions with the use of “L–L paths” and “L–T paths” (see Fig. 4(a–b)). Definition 2 An L–L path in a natural sketch is a path comprised of boundary lines and boundary junctions, where the terminal junctions are L-junctions and the interior junctions (if there are any) all have degree 3. Definition 3 An L–T path in a natural sketch is a path comprised of boundary lines and boundary junctions, where the
Estimating the number of hidden sketch-elements
In this section, lower bounds for the number of hidden lines, hidden junctions and hidden regions in a wireframe sketch are calculated, on the basis of the following properties related to the topological validity of a polyhedron and of its corresponding wireframe sketch:
- (A)
The number of junctions of odd degree in a graph is always even [20].
- (B)
Every junction of the wireframe sketch must be of degree three.
- (C)
The number of lines L in the wireframe sketch is
where J is the number of junctions in the
Construction of the minimal wireframe sketch
Step S4 of the PSS algorithm constructs a wireframe sketch by analyzing L–L and L–T paths of the natural sketch. A similar methodology has already been used by Grimstead and Martin [3], however, these authors attempt to directly construct the solid corresponding to the given natural sketch, while here we first construct a wireframe sketch from which the corresponding polyhedron is derived.
Construction of the minimal wireframe sketch proceeds in two steps, outlined in the following subsections.
Results and discussion
In this section we present results produced by an implementation of Part A of the PSS Algorithm, confirming the main contributions of this research, which are robust algorithms to solve the following problems:
- (a)
determination of hidden regions in a wireframe sketch (Section 4),
- (b)
estimation of minimum number of hidden lines/junctions/regions in the wireframe sketch (Section 5), and
- (c)
construction of a minimal wireframe sketch (Section 6).
Tests were performed on the 24 natural sketches of Fig. 14 that
Conclusions and current research
This paper and the Thesis [19] propose a new method for the topological construction of a wireframe sketch from a single natural sketch. The method uses a host of results and algorithmic tools from graph theory, solid modeling and Euclidean geometry that were not employed by published methods. Indeed, current “sketch interpretation/processing” methods rely heavily on concepts and tools from image-processing and computer vision, and in particular on the “line-labeling (LL)” technique. This is a
References (26)
- et al.
Digital clay: deriving digital models from freehand sketches
Automation in Construction
(2000) - et al.
A survey on geometrical reconstruction as a core technology to sketch-based modeling
Computer and Graphics
(2005) On seeing things
Artificial Intelligence
(1971)- et al.
Genetic algorithms for ambiguous labelling problems
Pattern Recognition
(2000) - et al.
Frontal geometry from sketches of engineering objects: is line labelling necessary
Computer-Aided Design
(2005) Weavings, sections and projections of spherical polyhedra
Discrete Applied Mathematics
(1991)- Lamb D, Bandopadhay A. Interpreting a 3D object from a rough 2D line drawing. In: Proceedings of the 1st IEEE...
- et al.
A survey of 3D solid reconstruction from 2D projection line drawings
Computer Graphics Forum
(1993) - Grimstead J, Martin R. Creating solid models from single 2D sketches. In: Proceedings of the third ACM symposium on...
- Varley P. Automatic creation of boundary-representation models from single line drawings. Doctoral thesis, Cardiff...