A method for the classification of 2-D partial shapes using Fourier descriptors is presented. The problem is formulated as one of estimating the Fourier descriptors of the unknown complete shape from the observations ...
详细信息
ISBN:
(纸本)0818607211
A method for the classification of 2-D partial shapes using Fourier descriptors is presented. The problem is formulated as one of estimating the Fourier descriptors of the unknown complete shape from the observations derived from an arbitrarily rotated and scaled shape with missing segments. The estimation method minimizes a sum of two terms: the first term is a least-square fit to the given data subject to the condition that the number of missing boundary points is not known;the second term is the perimeter**2 /area of the unknown shape. Experiments with synthetic and real boundaries show that estimates closer to the true values of Fourier descriptors of complete boundaries are obtained. Classification experiments performed using real boundaries indicate that reasonable classification accuracies are obtained even when 20-30% of the data are missing.
The authors present a set of rules for the grouping of edges into straight lines. These rules have been encoded into an algorithm for the extraction of straight lines from images. Zero-crossing points of the Laplacian...
详细信息
ISBN:
(纸本)0818607211
The authors present a set of rules for the grouping of edges into straight lines. These rules have been encoded into an algorithm for the extraction of straight lines from images. Zero-crossing points of the Laplacian of the intensity images and the gradients at those points are used to extract initial information about local edges. These edges are input to a hierarchical linking and merging algorithm. Edges are linked on the basis of both geometric and intrinsic properties, e. g. , if their endpoints are close and their contrasts are similar. In the merging process, a sequence of linked edges is replaced by a longer straight line if the approximation is sufficiently good. The hierarchy allows for the representation of lines at multiple scales without smoothing the image. The algorithm performs well by finding intensity boundaries in natural scenes in the presence of texture and noise. The results indicate that the use of geometric grouping in extracting long, straight lines produces a significant improvement over results obtained from standard edge detection algorithms.
The authors estimate rigid-body motion from two 3-D point sets with and without correspondences. By utilizing properties of Fourier transformation, a frequency-domain approach for determining the motion without point ...
详细信息
ISBN:
(纸本)0818607211
The authors estimate rigid-body motion from two 3-D point sets with and without correspondences. By utilizing properties of Fourier transformation, a frequency-domain approach for determining the motion without point correspondences is obtained. In the algorithm, the problem of estimating the translation is first separated by subtracting the coordinates of the centroids from the 3-D point coordinates in the point sets. The rotation parameters are estimated by correlating the Fourier transforms of two functions defined on the two sets of the 3-D points. With the point correspondence given, an algorithm for finding the least-squares solution to the motion parameters is applied. It is proved that the least-squares solution is such that the centroids of the two point sets coincide. The determination of rotation and translation can therefore be decoupled. An iterative procedure for obtaining the rotation parameters is described.
Texture provides an important source of information about the local orientation of visible surfaces. To recover 3-D structure, the distorting effects of the perspective projection must be distinguished from properties...
详细信息
ISBN:
(纸本)0818607211
Texture provides an important source of information about the local orientation of visible surfaces. To recover 3-D structure, the distorting effects of the perspective projection must be distinguished from properties of the texture on which the distortion acts. This requires that assumptions must be made about the texture. Methods are reported that exploit the uniform density assumption, first introduced by J. J. Gibson (1950), as well as a generalization of Gibson's assumption that involves the sum of the lengths of the edges per unit area. The conditions under which either the number of texels can be counted or the boundaries of the texels can be located, are discussed. The strategy has been implemented using statistical methods for the case of textured planes. The algorithms, which are conceptually simple and computationally effective, have been tested with both synthetic and natural images and the results were very satisfactory.
The discrete and noisy nature of digitized contours limits the quality of the extracted features, especially when feature extraction involves the estimation of the derivatives of the contour. Smoothing should be appli...
详细信息
ISBN:
(纸本)0818607211
The discrete and noisy nature of digitized contours limits the quality of the extracted features, especially when feature extraction involves the estimation of the derivatives of the contour. Smoothing should be applied to remove the noise from the signal. The use of a smoothing spline approximation for smoothing the digitized contours and estimating their derivatives is proposed. Effective use of the smoothing spline approximation requires the determination of the optimal degree of smoothing (optimal value of regularization parameter). Predictive sample reuse methodology is used to devise an automatic algorithm for the estimation of the optimal amount of smoothing from the data. The optimally smoothed contour can be used to extract the features. For closed contours a periodic smoothing spline can be used which guarantees that the resulting smooth curve is closed. Experimental results demonstrate the robustness of the method for estimating the boundary curvature of digitized contours.
Constructing the 3-D model of an object involves integrating data or descriptions of an object obtained from multiple views and representing this integrated data or descriptions in a coherent manner. The authors prese...
详细信息
ISBN:
(纸本)0818607211
Constructing the 3-D model of an object involves integrating data or descriptions of an object obtained from multiple views and representing this integrated data or descriptions in a coherent manner. The authors present a new technique for automatic construction of 3-D models of arbitrarily shaped objects, given range and intensity data acquired from multiple views. The technique does not require the correspondence relationship between views to be determined. The object for which the model is to be constructed is assumed to rest on a plane (base plane). A pattern consisting of a single straight line is drawn on the base plane. By observing the orientation of the base plane pattern in the intensity images from multiple views, the interframe transformation required to register any two views in a common reference coordinate system is derived. Once the interframe transformation for every view has been computed, the range data from different views are expressed in a common reference coordinate system and merged. The merging of data is achieved in cylindrical coordinates to yield the model of the object. The technique also demonstrates a way to combine multiple sources of information, namely, range and intensity.
It was shown in two demonstrations that the boundaries between texture regions are more salient than the broad texture fields which are not adjacent to the boundaries. In the first demonstration, observers found bound...
详细信息
ISBN:
(纸本)0818607211
It was shown in two demonstrations that the boundaries between texture regions are more salient than the broad texture fields which are not adjacent to the boundaries. In the first demonstration, observers found boundary-based texture discriminations (i. e. , displays in which the only discrimination information occurred at the boundary between texture regions) easier to perform than field-based texture discriminations (i. e. , displays in which the only discrimination information occurred away from the boundary between texture regions). The second demonstration was of a texture illusion, which was analogous to the Craik-O'Brien-Cornsweet brightness illusion. These effects argue for an extension of previous accounts of local linkage between texture features, elements, or attributes to include linkages based on maximal dissimilarity, which produce boundaries between texture regions. The effects also suggest that texture perception and brightness perception may occur analogously, or perhaps through common mechanisms.
To make passive stereo work, calibration is needed for at least two reasons: to establish the epipolar geometry of the system and cut down the time taken by the correspondence process;and to reconstruct the 3-D data a...
详细信息
ISBN:
(纸本)0818607211
To make passive stereo work, calibration is needed for at least two reasons: to establish the epipolar geometry of the system and cut down the time taken by the correspondence process;and to reconstruct the 3-D data after matching between retinas has been completed. The mathematics required to solve for the camera parameters is analyzed for two cases. In the first case the 3-D coordinates of a number of reference points are assumed and an efficient optimization technique is derived making use of the physical constraints available and of the invariance with respect to the group of rigid displacements of some functions of the parameters. In the second case, the calibration problem is solved (up to a scale factor of course) without the knowledge of any 3-D coordinates, just by matching features between retinas. It is shown that the parameters estimation can be done recursively by extended Kalman filtering, yielding both the parameters and a measure of their uncertainty. It is believed that this approach can provide a mobile stereo system with the capability of self-calibration, which is badly needed to make passive stereo really useful in applications.
The following topics are dealt with: edge and boundary analysis;vision systems;motion;shape and 2-D description;stereo and 3-D description;patternrecognition;3-D models;architectures;vision models and texture;image s...
详细信息
ISBN:
(纸本)0818606339
The following topics are dealt with: edge and boundary analysis;vision systems;motion;shape and 2-D description;stereo and 3-D description;patternrecognition;3-D models;architectures;vision models and texture;image segmentation;applications and parallel algorithms;3-D analysis;contour analysis;character recognition;3-D descriptions from multiple views;and parallel architectures for imageprocessing. 123 papers were presented, of which 121 are published in full in the present proceedings.
P. J. Burt (1981) proposed, on the basis of his experiments, the HDC (hierarchical discrete correlation) method for realizing a Gaussian filter. The authors present a theoretical analysis of HDC, answer some questions...
详细信息
ISBN:
(纸本)0818606339
P. J. Burt (1981) proposed, on the basis of his experiments, the HDC (hierarchical discrete correlation) method for realizing a Gaussian filter. The authors present a theoretical analysis of HDC, answer some questions posed by Burt's experiments, and indicate the rules for the design of HDC filters.
暂无评论