Bernard Chazelle

Shape distributions

Measuring the similarity between 3D shapes is a fundamental problem, with applications in computer graphics, computer vision, molecular biology, and a variety of other fields. A challenging aspect of this problem is to find a suitable shape signature that can be constructed and compared quickly, while still discriminating between similar and dissimilar shapes.In this paper, we propose and analyze a method for computing shape signatures for arbitrary (possibly degenerate) 3D polygonal models. The key idea is to represent the signature of an object as a shape distribution sampled from a shape function measuring global geometric properties of an object. The primary motivation for this approach is to reduce the shape matching problem to the comparison of probability distributions, which is simpler than traditional shape matching methods that require pose registration, feature correspondence, or model fitting.We find that the dissimilarities between sampled distributions of simple shape functions (e.g., the distance between two random points on a surface) provide a robust method for discriminating between classes of objects (e.g., cars versus airplanes) in a moderately sized database, despite the presence of arbitrary translations, rotations, scales, mirrors, tessellations, simplifications, and model degeneracies. They can be evaluated quickly, and thus the proposed method could be applied as a pre-classifier in a complete shape-based retrieval or analysis system concerned with finding similar whole objects. The paper describes our early experiences using shape distributions for object classification and for interactive web-based retrieval of 3D models.

Matching 3D models with shape distributions

Measuring the similarity between 3D shapes is a fundamental problem, with applications in computer vision, molecular biology, computer graphics, and a variety of other fields. A challenging aspect of this problem is to find a suitable shape signature that can be constructed and compared quickly, while still discriminating between similar and dissimilar shapes. In this paper, we propose and analyze a method for computing shape signatures for arbitrary (possibly degenerate) 3D polygonal models. The key idea is to represent the signature of an object as a shape distribution sampled from a shape function measuring the global geometric properties of an object. The primary motivation for this approach is to reduce the shape matching problem to the comparison of probability distributions, which is simpler than traditional shape matching methods that require pose registration, feature correspondence or model fitting. We find that the dissimilarities between sampled distributions of simple shape functions (e.g. the distance between two random points on a surface) provide a robust method for discriminating between classes of objects (e.g. cars versus airplanes) in a moderately sized database, despite the presence of arbitrary translations, rotations, scales, reflections, tessellations, simplifications and model degeneracies. They can be evaluated quickly, and thus the proposed method could be applied as a pre-classifier in an object recognition system or in an interactive content-based retrieval application.

Fractional cascading: I. A data structuring technique

In computational geometry many search problems and range queries can be solved by performing an iterative search for the same key in separate ordered lists. In this paper we show that, if these ordered lists can be put in a one-to-one correspondence with the nodes of a graph of degreed so that the iterative search always proceeds along edges of that graph, then we can do much better than the obvious sequence of binary searches. Without expanding the storage by more than a constant factor, we can build a data-structure, called afractional cascading structure, in which all original searches after the first can be carried out at only logd extra cost per search. Several results related to the dynamization of this structure are also presented. A companion paper gives numerous applications of this technique to geometric problems.

Functional approach to data structures and its use in multidimensional searching

We establish new upper bounds on the complexity of multidimensional searching. Our results include, in particular, linear-size data structures for range and rectangle counting in two dimensions with logarithmic query time. More generally, we give improved data structures for rectangle problems in any dimension, in a static as well as a dynamic setting. Several of the algorithms we give are simple to implement and might be the solutions of choice in practice. Central to this paper is the nonstandard approach followed to achieve these results. At its root we find a redefinition of data structures in terms of functional specifications.

Filtering search: a new approach to query answering

We introduce a new technique for solving problems of the following form: preprocess a set of objects so that those satisfying a given property with respect to a query object can be listed very effectively. Among well-known problems to fall into this category we find range query, point enclosure, intersection, near-neighbor problems, etc. The approach which we take is very general and rests on a new concept called fitering search. We show on a number of examples how it can be used to improve the complexity of known algorithms and simplify their implementations as well. In particular, filtering search allows us to improve on the worst-case complexity of the best algorithms known so far for solving the problems mentioned above.

A theorem on polygon cutting with applications

Let P be a simple polygon with N vertices, each being assigned a weight ∈ {0,1}, and let C, the weight of P, be the added weight of all vertices. We prove that it is possible, in O(N) time, to find two vertices a,b in P, such that the segment ab lies entirely inside the polygon P and partitions it into two polygons, each with a weight not exceeding 2C/3. This computation assumes that all the vertices have been sorted along some axis, which can be done in O(Nlog N) time. We use this result to derive a number of efficient divide-and-conquer algorithms for: 1. Triangulating an N-gon in O(Nlog N) time. 2. Decomposing an N-gon into (few) convex pieces in O(Nlog N) time. 3. Given an O(Nlog N) preprocessing, computing the shortest distance between two arbitrary points inside an N-gon (i.e., the internal distance), in O(N) time. 4. Computing the longest internal path in an N-gon in O(N2) time. In all cases, the algorithms achieve significant improvements over previously known methods, either by displaying better performance or by gaining in simplicity. In particular, the best algorithms for Problems 2,3,4, known so far, performed respectively in O(N2), O(N2), and O(N4) time.

An optimal convex hull algorithm in any fixed dimension

We present a deterministic algorithm for computing the convex hull ofn points inEd in optimalO(n logn+n⌞d/2⌟) time. Optimal solutions were previously known only in even dimension and in dimension 3. A by-product of our result is an algorithm for computing the Voronoi diagram ofn points ind-space in optimalO(n logn+n⌜d/2⌝) time.

A minimum spanning tree algorithm with inverse-Ackermann type complexity

A deterministic algorithm for computing a minimum spanning tree of a connected graph is presented. Its running time is <italic>0</italic>(<italic>m</italic> α(<italic>m, n</italic>)), where α is the classical functional inverse of Ackermanns function and <italic>n</italic> (respectively, <italic>m</italic>) is the number of vertices (respectively, edges). The algorithm is comparison-based : it uses pointers, not arrays, and it makes no numeric assumptions on the edge costs.

Cutting hyperplanes for divide-and-conquer

Givenn hyperplanes inEd, a (1/r)-cutting is a collection of simplices with disjoint interiors, which together coverEd and such that the interior of each simplex intersects at mostn/r hyperplanes. We present a deterministic algorithm for computing a (1/r)-cutting ofO(rd) size inO(nrd−1) time. If we require the incidences between the hyperplanes and the simplices of the cutting to be provided, then the algorithm is optimal. Our method is based on a hierarchical construction of cuttings, which also provides a simple optimal data structure for locating a point in an arrangement of hyperplanes. We mention several other applications of our result, e.g., counting segment intersections, Hopcrofts line/point incidence problem, linear programming in fixed dimension.

A deterministic view of random sampling and its use in geometry

The combination of divide-and-conquer and random sampling has proven very effective in the design of fast geometric algorithms. A flurry of efficient probabilistic algorithms have been recently discovered, based on this happy marriage. We show that all those algorithms can be derandomized with only polynomial overhead. In the process we establish results of independent interest concerning the covering of hypergraphs and we improve on various probabilistic bounds in geometric complexity. For example, givenn hyperplanes ind-space and any integerr large enough, we show how to compute, in polynomial time, a simplicial packing of sizeO(rd) which coversd-space, each of whose simplices intersectsO(n/r) hyperplanes.