Ron Wein
A continuous curve C in ℝ^{2} is called xmonotone, if every vertical line intersects it at a single point at most. For example, the circle x^{2} + y^{2} = 1 is not xymonotone as the vertical line x = 0 intersects it at (0, 1) and at (0, 1); however, it is possible to split the circle into an upper part and a lower part, such that both of these parts are xmonotone. We consider vertical segments as weakly xmonotone, to properly handle inputs that contain such vertical curves.
An xmonotone curve can be represented as a univariate function y = C(x), defined over some continuous range R_{C} ⊆ ℝ. Given a set C = { C_{1}, C_{2}, … , C_{n} } of xmonotone curves, their lower envelope is defined as the pointwise minimum of all curves. Namely, the lower envelope of the set C can be defined as the following function:

Similarly, the upper envelope of C is the pointwise maximum of the xmonotone curves in the set:

Given a set of xmonotone curves C, the minimization diagram of C is a subdivision of the xaxis into cells, such that the identity of the curves that induce the lower envelope over a specific cell of the subdivision (an edge or a vertex) is the same. In nondegenerate situations, an edge  which represents a continuous interval on the xaxis  is induced by a single curve (or by no curves at all, if there are no xmonotone curves defined over the interval), and a vertex is induced by a single curve and corresponds to one of its endpoints, or by two curves and corresponds to their intersection point. The maximization diagram is symmetrically defined for upper envelopes. In the rest of this chapter, we refer to both these diagrams as envelope diagrams.
Lower and upper envelopes can be efficiently computed using a divideandconquer approach. First, note that the envelope diagram for a single xmonotone curve C_{k} is trivial to compute: we project the boundary of its range of definition R_{Ck} onto the xaxis and label the features it induces accordingly. Given a set D of (non necessarily xmonotone) curves in ℝ^{2}, we subdivide each curve into a finite number of weakly xmonotone curves, and obtain the set C. Then, we split the set into two disjoint subsets C_{1} and C_{2}, and we compute their envelope diagrams recursively. Finally, we merge the diagrams in linear time by traversing both diagrams in parallel.
The package basically contains two sets of free functions: lower_envelope_x_monotone_2 (begin, end, diag) (similarly upper_envelope_x_monotone_2()) construct the envelope diagram for a given range of xmonotone curves, while lower_envelope_2 (begin, end, diag) (similarly upper_envelope_2()) construct the envelope diagram for a range of arbitrary (not necessarily xmonotone) curves. In this section we explain more on the structure of the envelope diagram these functions output.
A minimization diagram or a maximization diagram is represented by a model of the concept EnvelopeDiagram_1. This concept defines the structure of the subdivision of the xaxis into 0dimensional cells called vertices, and 1dimensional cells called edges. The important property of this subdivision is that the identity of the curves that induce the lower envelope (or the upper envelope) over each cell is fixed.
Figure 34.2 shows the lower envelope of a set of eight line segments, and sketches the structure of their minimization diagram. Each diagram vertex v is associated with a point p_{v} on the envelope, which corresponds to either a curve endpoint or to an intersection point of two (or more) curves. The vertex is therefore associated with a set of xmonotone curves that induce the envelope over p_{v}. Each vertex is incident to two edges, one lying to its left and the other to its right.
An edge in the envelope diagram represents a continuous portion of the xaxis, and is associated with a set of xmonotone curves that induce the envelope over this interval. Note that this set may be empty if no xmonotone curves are defined over this interval. In degenerate situations where curves overlap, there may be more than a single curve that induces the envelope over the interval the edge represents. An envelop diagram of a set of curves either consists of a single unbounded edge (in case the curve set is empty or if the envelope contains a single unbounded curve that is below or above all other curves), or at least one vertex and two unbounded edges, while each additional vertex comes along with an additional edge. It is possible to directly access the leftmost edge, representing the unbounded interval that starts at ∞, and the rightmost edge, representing the unbounded interval that ends at ∞. (In the example depicted in Figure 34.2 we have only bounded curves, so the leftmost and rightmost edges represent empty intervals. This is not the case when we deal, for example, with envelopes of sets of lines.)
Any model of the concept must define a geometric traits class, which in turn defines the Point_2 and X_monotone_curve_2 types defined with the diagram features. The geometric traits class must be a model of the ArrangementXMonotoneTraits_2 concept in case we construct envelopes of xmonotone curves. If we are interested in handling arbitrary (not necessarily xmonotone) curves, the traits class must be a model of the ArrangementTraits_2 concept. This concepts refined the ArrangementXMonotoneTraits_2 concept; a traits class that models this concepts must also defines a Curve_2 type, representing an arbitrary planar curve, and provide a functor for subdividing such curves into xmonotone subcurves.
The following example demonstrates how to compute and traverse the minimization diagram of line segments, as illustrated in Figure 34.2. We use the curvedata traits instantiated by the Arr_segment_traits_2 class, in order to attach a label (a char in this case) to each input segment. We use these labels when we print the minimization diagram:
File: examples/Envelope_2/envelope_segments.cpp
#include <CGAL/Gmpq.h> #include <CGAL/Cartesian.h> #include <CGAL/Arr_segment_traits_2.h> #include <CGAL/Arr_curve_data_traits_2.h> #include <CGAL/Envelope_diagram_1.h> #include <CGAL/envelope_2.h> #include <list> #include <iostream> typedef CGAL::Gmpq Number_type; typedef CGAL::Cartesian<Number_type> Kernel; typedef CGAL::Arr_segment_traits_2<Kernel> Segment_traits_2; typedef Segment_traits_2::X_monotone_curve_2 Segment_2; typedef CGAL::Arr_curve_data_traits_2<Segment_traits_2, char> Traits_2; typedef Traits_2::Point_2 Point_2; typedef Traits_2::X_monotone_curve_2 Labeled_segment_2; typedef CGAL::Envelope_diagram_1<Traits_2> Diagram_1; int main () { // Consrtuct the input segments and label them 'A' ... 'H'. std::list<Labeled_segment_2> segments; segments.push_back (Labeled_segment_2 (Segment_2 (Point_2 (0, 1), Point_2 (2, 3)), 'A')); segments.push_back (Labeled_segment_2 (Segment_2 (Point_2 (1, 2), Point_2 (4, 5)), 'B')); segments.push_back (Labeled_segment_2 (Segment_2 (Point_2 (1, 5), Point_2 (7, 2)), 'C')); segments.push_back (Labeled_segment_2 (Segment_2 (Point_2 (4, 2), Point_2 (6, 4)), 'D')); segments.push_back (Labeled_segment_2 (Segment_2 (Point_2 (8, 3), Point_2 (8, 6)), 'E')); segments.push_back (Labeled_segment_2 (Segment_2 (Point_2 (9, 2), Point_2 (12, 4)), 'F')); segments.push_back (Labeled_segment_2 (Segment_2 (Point_2 (10, 2), Point_2 (12, 1)), 'G')); segments.push_back (Labeled_segment_2 (Segment_2 (Point_2 (11, 0), Point_2 (11, 5)), 'H')); // Compute the minimization diagram that represents their lower envelope. Diagram_1 min_diag; lower_envelope_x_monotone_2 (segments.begin(), segments.end(), min_diag); // Print the minimization diagram. Diagram_1::Edge_const_handle e = min_diag.leftmost(); Diagram_1::Vertex_const_handle v; Diagram_1::Curve_const_iterator cit; while (e != min_diag.rightmost()) { std::cout << "Edge:"; if (! e>is_empty()) { for (cit = e>curves_begin(); cit != e>curves_end(); ++cit) std::cout << ' ' << cit>data(); } else std::cout << " [empty]"; std::cout << std::endl; v = e>right(); std::cout << "Vertex (" << v>point() << "):"; for (cit = v>curves_begin(); cit != v>curves_end(); ++cit) std::cout << ' ' << cit>data(); std::cout << std::endl; e = v>right(); } CGAL_assertion (e>is_empty()); std::cout << "Edge: [empty]" << std::endl; return (0); }
The next example computes the convex hull of a set of input points by constructing envelopes of unbounded curves, in our case lines that are dual to the input points. Here use the Arr_linear_traits_2 class to compute the lower envelope of the set of dual lines. We read a set of points P = p_{1}, … , p_{n} from an input file, and construct the corresponding dual lines P^{*} = p^{*}_{1}, … , p^{*}_{n}, where the line p^{*} dual to a point p = (p_{x}, p_{y}) is given by y = p_{x} x  p_{y}. We then compute the convex hull of the pointset P, using the fact that the lines that form the lower envelope of P^{*} are dual to the points along the upper part of P's convex hull, and the lines that form the upper envelope of P^{*} are dual to the points along the lower part of the convex hull; see, e.g., [dBvKOS00, Section 11.4] for more details. Note that the leftmost edge of the minimization diagram is associated with the same line as the rightmost edge of the maximization diagram, and viceverse. We can therefore skip the rightmost edges of both diagrams:
File: examples/Envelope_2/convex_hull.cpp
#include <CGAL/Gmpq.h> #include <CGAL/Cartesian.h> #include <CGAL/Arr_linear_traits_2.h> #include <CGAL/Arr_curve_data_traits_2.h> #include <CGAL/Envelope_diagram_1.h> #include <CGAL/envelope_2.h> #include <vector> typedef CGAL::Gmpq Number_type; typedef CGAL::Cartesian<Number_type> Kernel; typedef CGAL::Arr_linear_traits_2<Kernel> Linear_traits_2; typedef Linear_traits_2::Point_2 Point_2; typedef Linear_traits_2::Line_2 Line_2; typedef CGAL::Arr_curve_data_traits_2<Linear_traits_2, unsigned int> Traits_2; typedef Traits_2::X_monotone_curve_2 Dual_line_2; typedef CGAL::Envelope_diagram_1<Traits_2> Diagram_1; int main (int argc, char* argv[]) { // Read the points from the input file. const char* filename = (argc > 1) ? argv[1] : "ch_points.dat"; std::ifstream in_file(filename); if (!in_file.is_open()) { std::cerr << "Failed to open " << filename << " ..." << std::endl; return 1; } // Read the points from the file, and construct their dual lines. std::list<Dual_line_2> dual_lines; unsigned int n; in_file >> n; std::vector<Point_2> points; points.resize(n); for (unsigned int k = 0; k < n; ++k) { int px, py; in_file >> px >> py; points[k] = Point_2 (px, py); // The line dual to the point (p_x, p_y) is y = p_x*x  p_y, // or: p_x*x  y  p_y = 0: Line_2 line = Line_2 (Number_type(px), Number_type(1), Number_type(py)); // Generate the xmonotone curve based on the line and the point index. dual_lines.push_back (Dual_line_2 (line, k)); } in_file.close(); // Compute the lower envelope of dual lines, which corresponds to the upper // part of the convex hull, and their upper envelope, which corresponds to // the lower part of the convex hull. Diagram_1 min_diag; Diagram_1 max_diag; lower_envelope_x_monotone_2(dual_lines.begin(), dual_lines.end(), min_diag); upper_envelope_x_monotone_2(dual_lines.begin(), dual_lines.end(), max_diag); // Output the points along the boundary convex hull in counterclockwise // order. We start by traversing the minimization diagram from left to // right, then the maximization diagram from right to left. std::cout << "The convex hull of " << points.size() << " input points:"; Diagram_1::Edge_const_handle e = min_diag.leftmost(); while (e != min_diag.rightmost()) { std::cout << " (" << points[e>curve().data()] << ')'; e = e>right()>right(); } // Handle the degenerate case of a vertical convex hull edge: if (e>curve().data() != max_diag.leftmost()>curve().data()) std::cout << " (" << points[e>curve().data()] << ')'; e = max_diag.leftmost(); while (e != max_diag.rightmost()) { std::cout << " (" << points[e>curve().data()] << ')'; e = e>right()>right(); } std::cout << std::endl; return 0; }
We conclude by an example of envelopes of nonlinear curves. We use the Arr_circle_segment_traits_2 class to construct the lower and the upper envelopes of a set of four circles, as depicted in Figure 34.3. Note that unlike the two previous examples, here our curves are not xmonotone, so we use the functions that compute envelopes of arbitrary curves:
File: examples/Envelope_2/envelope_circles.cpp
#include <CGAL/Gmpq.h> #include <CGAL/Cartesian.h> #include <CGAL/Arr_circle_segment_traits_2.h> #include <CGAL/Arrangement_2.h> #include <CGAL/Envelope_diagram_1.h> #include <CGAL/envelope_2.h> typedef CGAL::Gmpq Number_type; typedef CGAL::Cartesian<Number_type> Kernel; typedef Kernel::Point_2 Kernel_point_2; typedef Kernel::Circle_2 Circle_2; typedef CGAL::Arr_circle_segment_traits_2<Kernel> Traits_2; typedef Traits_2::Curve_2 Curve_2; typedef CGAL::Envelope_diagram_1<Traits_2> Diagram_1; /*! Print the given envelope diagram. */ void print_diagram (const Diagram_1& diag) { Diagram_1::Edge_const_handle e = diag.leftmost(); Diagram_1::Vertex_const_handle v; while (e != diag.rightmost()) { std::cout << "Edge: "; if (! e>is_empty()) { Circle_2 circ = e>curve().supporting_circle(); std::cout << " (x  " << CGAL::to_double(circ.center().x()) << ")^2 +" << " (y  " << CGAL::to_double(circ.center().y()) << ")^2 = " << CGAL::to_double(circ.squared_radius()) << std::endl; } else std::cout << " [empty]" << std::endl; v = e>right(); std::cout << "Vertex (" << CGAL::to_double(v>point().x()) << ' ' << CGAL::to_double(v>point().y()) << ')' << std::endl; e = v>right(); } CGAL_assertion (e>is_empty()); std::cout << "Edge: [empty]" << std::endl; return; } /*! The main program. */ int main () { // Create four input circles. Curve_2 circles[4]; circles[0] = Circle_2 (Kernel_point_2 (1, 3), CGAL::square(2)); circles[1] = Circle_2 (Kernel_point_2 (4, 5), CGAL::square(4)); circles[2] = Circle_2 (Kernel_point_2 (5, 1), CGAL::square(1)); circles[3] = Circle_2 (Kernel_point_2 (6, 7), CGAL::square(2)); // Compute the minimization diagram that represents their lower envelope. Diagram_1 min_diag; lower_envelope_2 (&(circles[0]), &(circles[4]), min_diag); print_diagram (min_diag); // Compute the maximization diagram that represents the upper envelope. Diagram_1 max_diag; upper_envelope_2 (&(circles[0]), &(circles[4]), max_diag); print_diagram (max_diag); return (0); }