Visual cluster analysis and pattern recognition methods

• US 6,304,675 B1
• Filed: 12/28/1993
• Issued: 10/16/2001
• Est. Priority Date: 12/28/1993
• Status: Expired due to Term

• Current Assignee: National Technology & Engineering Solutions Of Sandia LLC (Honeywell International Inc.)
• Original Assignee: Sandia Corporation (Martin Marietta Corporation)
• Inventors: Martinez, Rubel Francisco, Osbourn, Gordon Cecil
• Primary Examiner(s): CHANG, JON CARLTON
• Application Number:
  US08/174,548
• Time in Patent Office: 2,849 Days
• Field of Search: 382/181, 382/209, 382/224, 382/225, 382/228, 382/159, 382/170
• US Class Current: 382/225
• CPC Class Codes: G06F 18/23 Clustering techniques
• CPC Class Codes: G06F 18/256 of results relating to different input data, e.g. multimodal recognition

Abstract

A method of clustering using a novel template to define a region of influence. Using neighboring approximation methods, computation times can be significantly reduced. The template and method are applicable and improve pattern recognition techniques.

Claims

1. A pattern recognition method, comprising:

(a) assigning vector coordinates to each training data point of a training data set having class assignments, and assigning vector coordinates to each test data point of a test data set to be assigned classes;

(b) determining class assignments of each test data point of said test data set by clustering those points individually with said training data set, said clustering accomplished by means of;

i) selecting each of said test data points of said test data set and selecting each of said training data points of said training data set, and

ii) placing a test data point and a training data point on each of two specified positions of a region of influence that is oblong having a central portion narrowed by two opposing concave depressions with two wider larger end portions on both sides of said central portion;

(c) assigning said test data point to a class associated with said training data point if no other training data point of said training data set lies within said region of influence;

(d) assigning said test data point to an ambiguous class if said test data point groups with no class or more than one class in said training set; and

(e) repeating steps (b), (c), and (d) above for said training data set and said test data set.

2. A method to verify quality of a training data set used to make class assignments of data in pattern recognition, comprising:

(a) assigning vector coordinates to each data point of a data set having predetermined class assignments;

(b) clustering said data points of said data set by placing two of said data point on each of two specified positions of a region of influence that is oblong having a central portion narrowed by two opposing concave depressions with two wider larger end portions on both sides of said central portion;

(c) assigning a same class to said clustering data points if no other training data point of said data set lies within said region of influence;

(d) assigning said test data point to an ambiguous class if said test data point groups with no class or more than one class in said training set;

(e) assigning an ambiguous class if any other data point besides said clustering data points lies within said region of influence;

(f) repeating steps (b), (c), and (d) above for said data set.