A Pattern Recognition Feature Optimization Tool Using the Visual Empirical Region of Influence Algorithm

MARTINEZ, RUBEL F.

Sandia National Laboratories

关键词: Computer Graphics;    Data Analysis;    Pattern Recognition;    Programming Languages;    99 General And Miscellaneous//Mathematics, Computing, And Information Science;   

DOI  :  10.2172/800997
RP-ID  :  SAND2002-1881
RP-ID  :  AC04-94AL85000
RP-ID  :  800997

美国|英语

来源: UNT Digital Library

【 摘 要 】

This document is the second in a series that describe graphical user interface tools developed to control the Visual Empirical Region of Influence (VERI) algorithm. In this paper we describe a user interface designed to optimize the VERI algorithm results. The optimization mode uses a brute force method of searching through the combinations of features in a data set for features that produce the best pattern recognition results. With a small number of features in a data set an exact solution can be determined. However, the number of possible combinations increases exponentially with the number of features and an alternate means of finding a solution must be found. We developed and implemented a technique for finding solutions in data sets with both small and large numbers of features. This document illustrates step-by-step examples of how to use the interface and how to interpret the results. It is written in two parts, part I deals with using the interface to find the best combination from all possible sets of features, part II describes how to use the tool to find a good solution in data sets with a large number of features. The VERI Optimization Interface Tool was written using the Tcl/Tk Graphical User Interface (GUI) programming language, version 8.1. Although the Tcl/Tk packages are designed to run on multiple computer platforms, we have concentrated our efforts to develop a user interface for the ubiquitous DOS environment. The VERI algorithms are compiled, executable programs. The optimization interface executes the VERI algorithm in Leave-One-Out mode using the Euclidean metric. For a thorough description of the type of data analysis we perform, and for a general Pattern Recognition tutorial, refer to our website at: http://www.sandia.gov/imrl/XVisionScience/Xusers.htm.