Typical decision rules for classifying a Gaussian random vector require a simultaneous computation of quadratic forms. The paper discusses a fast algorithm for this problem which uses a Kalman filter and possibilities...
详细信息
Typical decision rules for classifying a Gaussian random vector require a simultaneous computation of quadratic forms. The paper discusses a fast algorithm for this problem which uses a Kalman filter and possibilities for a model identification.
A generalization to discrete constraint analysis based on a grammatical model is described and its application to shape recognition is discussed. The generalization is called a hierarchical relaxation process and it i...
详细信息
A generalization to discrete constraint analysis based on a grammatical model is described and its application to shape recognition is discussed. The generalization is called a hierarchical relaxation process and it is based on deriving local constraints at many levels of the description from a stratified, context-free system of relational models. These constraints are then applied to a network of hypotheses about low-level image descriptions. A hierarchical relaxation process for shape analysis is then constructed.
This symposium contains 96 papers, of which 5 are given in abstract form only. Eighty-four papers are indexed separately. The topics covered are: medical ultrasound images;shape analysis and edge detection;defense app...
详细信息
This symposium contains 96 papers, of which 5 are given in abstract form only. Eighty-four papers are indexed separately. The topics covered are: medical ultrasound images;shape analysis and edge detection;defense applications;image filtering and compression;statistical patternrecognition;computed tomography;map data processing;computational aspects PRIP methodologies;structural patternrecognition;image segmentation and structure;imageprocessing applications;cooperative methods in imageprocessing;pictorial data bases and data structure;algorithms in pattern classification;and analysis of cell images.
A method of mixture identification is presented employing an analog computer. The advantage of the technique is the fast and straightforward computation in a dynamic fashion. The disadvantages are the lack of precisio...
详细信息
A method of mixture identification is presented employing an analog computer. The advantage of the technique is the fast and straightforward computation in a dynamic fashion. The disadvantages are the lack of precision and the requirement of a large number of operational amplifiers especially if the interval to which the unknown parameters are known to belong is large.
Existing results on the development of minimum decision error algorithms are presented. Critical comments are presented and the emphasis is on the practical implementation of such algorithms. Some drawbacks of existin...
详细信息
Existing results on the development of minimum decision error algorithms are presented. Critical comments are presented and the emphasis is on the practical implementation of such algorithms. Some drawbacks of existing algorithms are pointed out and a more efficient algorithm is proposed and discussed.
A new two-dimensional noise filtering method is presented which depends only upon the statistical characteristics of the data and a one-dimensional filter synthesis algorithm.
A new two-dimensional noise filtering method is presented which depends only upon the statistical characteristics of the data and a one-dimensional filter synthesis algorithm.
In practical recognition problems the sample size is often limited and the actual performance may be quite different from that theoretically predicted. The design and evaluation of recognition systems must take into a...
详细信息
In practical recognition problems the sample size is often limited and the actual performance may be quite different from that theoretically predicted. The design and evaluation of recognition systems must take into account the finite sample constraint. Finite sample effects are considered here in three major problem areas: distance and information measures, classification rules, and contextual analysis.
Methods are presented, for block implementation of two-dimensional FIR and IIR filters. The advantages of such implementation parallel the known advantages of block implementation of digital one-dimensional filters.
Methods are presented, for block implementation of two-dimensional FIR and IIR filters. The advantages of such implementation parallel the known advantages of block implementation of digital one-dimensional filters.
Textures are constructed from textural primitives according to certain rules that determine their interrelationships. Some techniques are presented for extracting the primitives, measuring their attributes, and determ...
详细信息
Textures are constructed from textural primitives according to certain rules that determine their interrelationships. Some techniques are presented for extracting the primitives, measuring their attributes, and determining their structural relationships. The problem of locating changes either in attributes or in relationships is discussed, and early results are given.
The notions of capacity and distribution-free error estimation are extended to nonlinear Boolean classifiers on patterns with binary-valued features. Quantitative relationships are established between the dimensionali...
详细信息
The notions of capacity and distribution-free error estimation are extended to nonlinear Boolean classifiers on patterns with binary-valued features. Quantitative relationships are established between the dimensionality of the feature vectors, the combinational complexity of the decision rule, the number of samples in the training set and the classification performance of the resulting classifier.
暂无评论