Dismiss
InnovationQ will be updated on Sunday, Oct. 22, from 10am ET - noon. You may experience brief service interruptions during that time.
Browse Prior Art Database

Weight Quantification Model in the Feature Space for Biometric Evidence

IP.com Disclosure Number: IPCOM000244031D
Publication Date: 2015-Nov-05

Publishing Venue

The IP.com Prior Art Database

Abstract

This publication describes methods to quantify the weight of biometric e g fingerprint or facial evidence to aid examination by improving upon the extraction and analysis of features Three types of biometric data 1 a sample trace or mark 2 multiple representations of a target sample i e control and 3 a large collection of random samples contained in a reference database are analyzed Due to the complexity of an individual fingerprint pattern it is a challenge to characterize it mathematically Therefore methods have been developed to reduce the complexity of the characterization of a pattern in order to study the structure of the likelihoods of such patterns The methods rely on the measurement of similarities and differences e g score or metric between biometric patterns to convert a complex pattern into a one dimensional variable The methods identify and leverage one or more reference or origin spatial point s to transform the three types of fingerprint data utilized and to give proper statistically coherent properties to the calculated weight of the biometric evidence

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 7% of the total text.

Page 01 of 16

Weight Quantification Model in the Feature Space for Biometric Evidence

Abstract

This publication describes methods to quantify the weight of biometric (e.g., fingerprint or facial) evidence to aid examination by improving upon the extraction and analysis of features. Three types of biometric data: 1) a sample, trace, or mark 2) multiple representations of a target sample (i.e., control) and 3) a large collection of random samples contained in a reference database are analyzed. Due to the complexity of an individual fingerprint pattern, it is a challenge to characterize it mathematically. Therefore, methods have been developed to reduce the complexity of the characterization of a pattern in order to study the structure of the likelihoods of such patterns. The methods rely on the measurement of similarities and differences (e.g., score or metric) between biometric patterns to convert a complex pattern into a one-dimensional variable. The methods identify and leverage one or more reference or "origin" spatial point(s) to transform the three types of fingerprint data utilized, and to give proper statistically coherent properties to the calculated weight of the biometric evidence.

Introduction

Biometric data, with considerable justification, has come to be regarded as the zenith of forensic identification. Over the last century, millions of cases have been resolved worldwide because of biometric data discovered at crime scenes or other locations. Comparison methodologies have not evolved greatly during this time period and it is universal practice to present biometric data with a categorical opinion of identification or exclusion, or to classify the data as inconclusive and not to report it.

There is tremendous value in supplementing a biometric data examination process with one that includes a statistical model and is supported by appropriate databases for calculating numerical measures of the weight of data. Such a solution calls for the establishment of a logical framework for informing conclusions based on explicit assumptions and data that is open to revision and improvement.

The weight of biometric data may be quantified by computing the ratio between (a) the probability of observing a particular trace sample if a suspect left it and (b) the probability of observing the same trace sample if some random person left it. Three types of biometric data: a trace sample, multiple representations from a target sample, and a reference database of samples from random individuals may be used to assign statistical probability density functions for the numerator and denominator of a model (the "likelihood ratio") and to calculate the weight of the data.

Statistical probability density function assignment is very complicated as biometric evidence comprises complex pattern that cannot easily be summarized mathematically. Even when the pattern is mathematically summarized into multi- dimensional variables (e.g., N-dimensional space), assignin...