System and method for feature set reduction
First Claim
1. A system for automatic pattern classification by ranking features within a feature set using a computer readable memory and processing capabilities, said system comprising:
- a feature extractor, connected to receive a plurality of exemplars, for producing an n-element feature vector for each one of said plurality of exemplars, said n-element feature vectors comprising a design set;
a training set compiler, connected to receive said design set, for creating a training set;
a projection space processor, connected to receive said training set created by said training set compiler, for generating a reduced feature space for said training set; and
a Procrustes feature ranking processor, connected to receive reduced feature space, for determining a Procrustes angle for each feature in said feature set, wherein said Procrustes angle is the angle between said feature and its projection onto said reduced feature space, and for linearly ranking said features in increasing numerical size of their respective Procrustes angle, wherein said linearly ranked features comprise a ranked feature set.
2 Assignments
0 Petitions
Accused Products
Abstract
A system and method for ranking features by exploiting their relationship the Fisher projection space. The system ranks n features in a feature set using a design set comprising exemplars from each of M possible event classes of an associated feature-based classification system. A training set is created by randomly selecting exemplars from each of the M classes in the design set. A "smoothed" Fisher projection space for the training set is created by replacing the sample means and the within-class sample covariance matrix normally used in deriving a Fisher projection space with expressions for the mean vectors and covariance matrices derived from event class probability density function estimates. The angle between a given feature and the smoothed Fisher projection space is calculated for each feature in the feature set, and the features are then ordered by increasing numerical size of this angle. The system produces a reduced feature set by eliminating those features which are not important for classification based on the linear ranking of the features.
46 Citations
19 Claims
-
1. A system for automatic pattern classification by ranking features within a feature set using a computer readable memory and processing capabilities, said system comprising:
-
a feature extractor, connected to receive a plurality of exemplars, for producing an n-element feature vector for each one of said plurality of exemplars, said n-element feature vectors comprising a design set; a training set compiler, connected to receive said design set, for creating a training set; a projection space processor, connected to receive said training set created by said training set compiler, for generating a reduced feature space for said training set; and a Procrustes feature ranking processor, connected to receive reduced feature space, for determining a Procrustes angle for each feature in said feature set, wherein said Procrustes angle is the angle between said feature and its projection onto said reduced feature space, and for linearly ranking said features in increasing numerical size of their respective Procrustes angle, wherein said linearly ranked features comprise a ranked feature set. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11)
-
-
12. A method of ranking features within a feature set used with an automatic pattern classification system having M exclusive event classes using a computer readable memory and processing capabilities, said method comprising the steps of:
-
using a measurement system of said automatic classification pattern system to acquire a plurality of exemplars; producing an n-element feature vector for each one of said exemplars, said n-element feature vectors comprising a design set; creating a training set by randomly sampling said design set, said training set comprising a subset of said design set; generating a reduced feature space using said n-element feature vectors comprising said training set; determining a Procrustes angle for each feature in said feature set, wherein said Procrustes angle is the angle between said feature and its projection onto said reduced feature space; and ranking said features by linearly ordering said features in increasing numerical size of their respective Procrustes angle, wherein said linearly ranked features comprise a ranked feature set. - View Dependent Claims (13, 14, 15, 16, 17, 18)
-
-
19. A method of ranking features within a feature set used with an automatic pattern classification system having M exclusive event classes using a computer readable memory and processing capabilities, said method comprising the steps of:
-
using a measurement system of said automatic classification system to acquire a plurality of exemplars; producing an n-element feature vector for each one of said exemplars, said n-element feature vectors comprising a design set; creating a training set by randomly sampling said design set, said training set comprising a subset of said design set; generating a smoothed Fisher projection space using said n-element feature vectors comprising said training set; determining a Procrustes angle for each feature in said feature set, wherein said Procrustes angle is the angle between said feature and its projection onto said smoothed Fisher projection space; ranking said features by linearly ordering said features in increasing numerical size of their respective Procrustes angle, wherein said linearly ranked features comprise a ranked feature set; retaining said ranked feature set; repeating said steps of creating a training set, generating a smoothed Fisher projection space, determining a Procrustes angle for each feature in said feature set, ranking said features, and retaining said ranked feature set a predetermined number of times to generate a plurality of ranked feature sets; determining a ranking count from said ranked feature sets; constructing a thresholded ranking count for said ranking count; determining a breakpoint in said thresholded ranking count and identifying a Procrustes number associated with said breakpoint; and compiling a reduced feature set, said reduced feature set comprising a number of features equal to said Procrustes number.
-
Specification