CVGIP: Image Understanding最新文献

In this paper we consider a Hough transform line finding algorithm in which the voting kernel is a smooth function of differences in both line parameters. The shape of the voting kernel is decided in terms of a hypothesis testing approach, and the shape is adjusted to give optimal results. We show that this new kernel is robust to changes in the distribution of the underlying noise and the implementation is very fast, taking typically 2-3 s on a Sparc 2 workstation for a 256 × 256 image.

We present a method for semantic labelling of edges and reconstruction of range data by fusion of registered range and intensity images. An initial set of edge labels is derived using a physical model of object geometry and shading. A final edge classification and range reconstruction are obtained using Bayesian estimation within coupled Markov random fields employing constraints of surface smoothness and edge continuity. The approach is demonstrated on synthetic and real source data, obtained from an active laser rangefinder.

The construction of a three-dimensional object model from a set of images taken from different viewpoints is an important problem in computer vision. One of the simplest ways to do this is to use the silhouettes of the object (the binary classification of images into object and background) to construct a bounding volume for the object. To efficiently represent this volume, we use an octree, which represents the object as a tree of recursively subdivided cubes. We develop a new algorithm for computing the octree bounding volume from multiple silhouettes and apply it to an object rotating on a turntable in front of a stationary camera. The algorithm performs a limited amount of processing for each viewpoint and incrementally builds the volumetric model. The resulting algorithm requires less total computation than previous algorithms, runs in close to real-time, and builds a model whose resolution improves over time.

This paper presents a bibliography of nearly 1900 references related to computer vision and image analysis, arranged by subject matter. The topics covered include architectures; computational techniques; feature detection and segmentation; image analysis; two-dimensional shape; pattern; color and texture; matching and stereo; three-dimensional recovery and analysis; three-dimensional shape; and motion. A few references are also given on related topics, such as geometry, graphics, image input/output and coding, image processing, optical processing, visual perception, neural nets, pattern recognition, and artificial intelligence, as well as on applications.

Extracting geometric primitives is an important task in model-based computer vision. The Hough transform is the most common method of extracting geometric primitives. Recently, methods derived from the field of robust statistics have been used for this purpose. We show that extracting a single geometric primitive is equivalent to finding the optimum value of a cost function which has potentially many local minima. Besides providing a unifying way of understanding different primitive extraction algorithms, this model also shows that for efficient extraction the true global minimum must be found with as few evaluations of the cost function as possible. In order to extract a single geometric primitive we choose a number of minimal subsets randomly from the geometric data. The cost function is evaluated for each of these, and the primitive defined by the subset with the best value of the cost function is extracted from the geometric data. To extract multiple primitives, this process is repeated on the geometric data that do not belong to the primitive. The resulting extraction algorithm can be used with a wide variety of geometric primitives and geometric data. It is easily parallelized, and we describe some possible implementations on a variety of parallel architectures. We make a detailed comparison with the Hough transform and show that it has a number of advantages over this classic technique.

A planar shape distorted by a projective viewing transformation can be recognized under partial occlusion if an invariant description of its boundary is available. Invariant boundary descriptions should be based solely on the local properties of the boundary curve, perhaps relying on further information on the viewing transformation. Recent research in this area has provided a theory for invariant boundary descriptions based on an interplay of differential, local, and global invariants. Differential invariants require high-order derivatives. However, the use of global invariances and point match information on the distorting transformations enables the derivation of invariant signatures for planar shapes using lower order derivatives. Trade-offs between the highest order derivatives required and the quantity of additional information constraining the distorting viewing transformations are made explicit. Once an invariant is established, recognition of the equivalence of two objects requires only partial function matching. Uses of these invariants include the identification of planar surfaces in varying orientations and resolving the outline of a cluster for planar objects into individual components.

In this paper the relevance of perspective geometry for 3D scene analysis from a single view is asserted. Analytic procedures for perspective inversion of special primitive configurations are presented. Four configurations are treated: (1) four coplanar segments; (2) three orthogonal segments; (3) a circle arc; (4) a quadric of revolution. A complete and thorough illustration of the developed methodologies is given. The importance of the selected primitives is illustrated in different application contexts. Experimental results on real images are provided for configurations (3) and (4).

This paper addresses the problem of recognizing an object in a given scene using a three-dimensional model of the object. The scene may contain several overlapping objects, arbitrarily positioned and oriented. A laser range scanner is used to collect three-dimensional (3D) data points from the scene. The collected data is segmented into surface patches, and the segments are used to calculate various 3D surface properties. The CAD models are designed using commercially available CADKEY and accessed via the industry standard IGES. The models are analyzed off-line to derive various geometric features, their relationships, and their attributes. A strategy for identifying each model is then automatically generated and stored. The strategy is applied at run-time to complete the task of object recognition. The goal of the generated strategy is to select the model′s geometric features in the sequence which may best be used to identify and locate the model in the scene. The generated strategy is guided by several factors, such as the visibility, detectability, the frequency of occurrence, and the topology of the features. The paper concludes with examples of the generated strategies and their application to object recognition in several scenes containing multiple objects.

Plants such as trees can be modeled by three-dimensional hierarchial branching structures. If these structures are sufficiently sparse, so that self-occulation is relatively minor, their geometrical properties can be recovered from a single image.

A set of modules to extract partial descriptions of SHGC objects in an edge image is presented. It consists of modules to find end edges, to find meridian edges, to find cross-section edges, and to recover 3D shapes. The first goal of the system is to extract geometrical edges derived from an SHGC object. From an input edge image, pairs of end edges are detected first by verifying strong geometrical constraints for the ends of an SHGC. Then, meridian edges are detected by using the constraint for tangent intersections and the ones related to the end edges. The second goal is to recover 3D information of the object. The axis of SHGC and the axes of skewed symmetry in cross-section edges are detected. Then, original cross section and the sweeping rule are recovered by utilizing these three orthogonal axes. Extracted geometrical edges and 3D information from real images are shown.