A key element in patternrecognition is the description of shape. The description of shape is facilitated by segmenting the boundary line at so-called critical points - corners (discontinuities in curvature), points o...
详细信息
A key element in patternrecognition is the description of shape. The description of shape is facilitated by segmenting the boundary line at so-called critical points - corners (discontinuities in curvature), points of inflection, and curvature maxima. Additional critical points are intersections and points of tangency. Algorithms are described for extracting such critical points in the presence of noise. An illustration is given showing how the critical points may be used in the development of a shape description system.
This paper explores several approaches to outlining point clusters in the plane in a visually acceptable way consistent with capturing shape attributes which may have application in patternrecognition and scene analy...
详细信息
This paper explores several approaches to outlining point clusters in the plane in a visually acceptable way consistent with capturing shape attributes which may have application in patternrecognition and scene analysis. Both line and region oriented methods are discussed and the strengths and weaknesses of each explored. General comments are made on the nature of the problem and the duality relationship between line and region representations. A variety of examples are offered for subjective evaluation.
Several methods for estimating a sample-based discriminant's probability of correct classification are compared with respect to bias, variance, and robustness. ″Smooth″ modification of the counting method, or sa...
详细信息
Several methods for estimating a sample-based discriminant's probability of correct classification are compared with respect to bias, variance, and robustness. ″Smooth″ modification of the counting method, or sample success proportion, is recommended to reduce bias while retaining stability and robustness. In contrast, the popular ″leave-one-out″ technique is a counting method whose bias reduction is vitiated by corresponding increase in variance.
New and useful results on multistage multiclass statistical classification and on model-driven, data-directed structural pattern analysis have been obtained by developing and exploiting connections between state-space...
详细信息
New and useful results on multistage multiclass statistical classification and on model-driven, data-directed structural pattern analysis have been obtained by developing and exploiting connections between state-space and AND/OR graph models, formal grammars, heuristic search and probabilistic decision making. This paper highlights the concepts and connections which have led to these results.
Performance of Hough-like transforms in the presence of inaccurate feature point estimates is examined comparing predicted and simulated results. Cases where there are several curve families are treated when correct a...
详细信息
Performance of Hough-like transforms in the presence of inaccurate feature point estimates is examined comparing predicted and simulated results. Cases where there are several curve families are treated when correct and incorrect transforms are applied. Results are given for texture inputs to the transforms. Scatter diagrams in the transformed space are given.
This paper presents a simple algorithm to detect and label homogeneous areas in an image without sequential region growing or edge boundary tracing. The scheme constructs directed trees with the image points as nodes,...
详细信息
This paper presents a simple algorithm to detect and label homogeneous areas in an image without sequential region growing or edge boundary tracing. The scheme constructs directed trees with the image points as nodes, guided by an edge value computed at every point. These directed trees segment the image into disjoint regions. The properties of the resultant segments are stated in terms of the edge image. The algorithm is shown to be simple, efficient and effective for detecting homogeneous segments in the presence of noise. Results of application of the algorithm to segment a LANDSAT multispectral scene of an agricultural area are included.
The approach to contour extraction described here formulates the contour extraction problem as one of minimum cost tree searching. Branch costs or metrics are defined which are indicative of the likelihood that a part...
详细信息
The approach to contour extraction described here formulates the contour extraction problem as one of minimum cost tree searching. Branch costs or metrics are defined which are indicative of the likelihood that a particular branch lies on the true contour. The branch metrics incorporate both local and global or contextual information. The most likely path or contour is then extracted by application of a heuristic tree searching algorithm. The approach has been successfully applied to a number of biomedical imageprocessing problems.
A new application for clustering is introduced. Clustering is used here for region selection in problems such as piecewise linear classification and density estimation. This method may prove fruitful in a wide class o...
详细信息
A new application for clustering is introduced. Clustering is used here for region selection in problems such as piecewise linear classification and density estimation. This method may prove fruitful in a wide class of problems in which one needs to divide the probability space into local regions and thus attack the problem at a localized level. Another advantage to this technique is its easy adaptability to a tree structure for region devision. The technique is discussed and results given.
Analysis of the LANDSAT multispectral scanner data undertaken in the context of identification of wheat, reveals that the data from all agricultural sites essentially lie on parallel two-dimensional planes in the four...
详细信息
Analysis of the LANDSAT multispectral scanner data undertaken in the context of identification of wheat, reveals that the data from all agricultural sites essentially lie on parallel two-dimensional planes in the four-dimensional feature space; and that even though the spectral response of wheat at each stage of the crop varies considerably from field to field, the pattern of temporal changes in the response is similar for each field. This temporal pattern, then, constitutes a signature of wheat and provides a basis for both supervised and unsupervised classification of data.
Stochastic representation of discrete images by partial differential equations proposed earlier is used for filtering of noisy images. Comparisons among various representations and their relative effectiveness in mode...
详细信息
Stochastic representation of discrete images by partial differential equations proposed earlier is used for filtering of noisy images. Comparisons among various representations and their relative effectiveness in modeling the actual statistics of the images is shown by examples. Certain image decompositions resulting in fast filtering algorithms are given. The results are also compared with spatial averaging and Fourier domain Wiener filters.
暂无评论