Pattern classification method

Abstract

For assigning a test pattern to a class chosen from a predefined set of classes, the class membership probability for the test pattern is calculated as well as the confidence interval for the class membership probability based upon a number of training patterns in a neighbourhood of the test pattern in the feature space. The number of training patterns in the neighbourhood of the test pattern is obtained from computing a convolution of a density function of the training patterns with a Gaussian smoothing function centred on the test pattern, where the density function of the training patterns is represented as a mixture of Gaussian functions. The convolution of the smoothing function and the mixture of Gaussian functions can be expressed analytically.

Description

technical field [0001] The present invention generally relates to a method of pattern classification and a system implementing the method. Background technique [0002] Pattern classification is well known in many real-world applications such as speech recognition, vehicle seat occupant classification, data mining, risk prediction, diagnostic classification, etc. The primary goal of a pattern classifier is to assign a test pattern to one or more classes within a predefined set of classes. The test pattern can be thought of as a feature vector, or more precisely, as a number quantifying the features. For a given input pattern, a statistical classifier computes the conditional probabilities of different classes (hereinafter also referred to as "class membership probabilities"). Deviations of these class membership probabilities from 1 are usually interpreted as the risk of misclassification. [0003] The challenge in pattern classification is to reduce misclassification. A...

AI Technical Summary

Problems solved by technology

Likewise, if the density of the training patterns around that test pattern is low, the uncertainty will be high

Images