The University of Southampton
University of Southampton Institutional Repository

Discovering human descriptions for ubiquitous visual identification

Discovering human descriptions for ubiquitous visual identification
Discovering human descriptions for ubiquitous visual identification
Identifying suspects in surveillance footage is paramount in ensuring public safety, preventing crime, policing and forensic investigation. At present, finding an individual in real-world CCTV footage given only an eye-witness description is near impossible. The vast majority of contemporary research assumes coarse, expertly-defined categories to describe subjects, ineffective in dealing with unconstrained, low quality and obscured images. Such brittle representations hamper semantic image discrimination and the ability to learn robust predictors from challenging subject matter. This thesis explores human and machine centric techniques for representing and learning semantic human descriptions for suspect identification. By investigating the duality of human-machine communication, we enhance the capabilities of traditional attributes and soft biometric descriptors, expanding their versatility and applicability towards challenging images and large-scale surveillance datasets. We experiment with crowdsourcing human annotations using ordered and similarity comparisons, and estimating attributes from images employing a variety of state-of-the-art machine learning techniques. Our focus is on utilising a lean lexicon of global and body characteristics that are most pertinent when estimated from stand-alone surveillance footage. Significant improvements in suspect retrieval and identification performance are achieved by discovering enhanced soft biometric descriptions which represent visual trait characteristics with more precision and relevance. This work evolves the areas of soft biometrics and identity science, drawing ideas from contemporary image attribute recognition, semantic attribute discovery, pedestrian reidentification and perceptual psychology. Our findings indicate that increasing not only the volume, but the complexity of information conveyed between humans and machines is key in deploying soft biometrics ubiquitously.
University of Southampton
Martinho-Corbishley, Daniel
6dd73e5c-9a7e-41bd-b896-fb1ea9852abb
Martinho-Corbishley, Daniel
6dd73e5c-9a7e-41bd-b896-fb1ea9852abb
Nixon, Mark
2b5b9804-5a81-462a-82e6-92ee5fa74e12

Martinho-Corbishley, Daniel (2018) Discovering human descriptions for ubiquitous visual identification. University of Southampton, Doctoral Thesis, 137pp.

Record type: Thesis (Doctoral)

Abstract

Identifying suspects in surveillance footage is paramount in ensuring public safety, preventing crime, policing and forensic investigation. At present, finding an individual in real-world CCTV footage given only an eye-witness description is near impossible. The vast majority of contemporary research assumes coarse, expertly-defined categories to describe subjects, ineffective in dealing with unconstrained, low quality and obscured images. Such brittle representations hamper semantic image discrimination and the ability to learn robust predictors from challenging subject matter. This thesis explores human and machine centric techniques for representing and learning semantic human descriptions for suspect identification. By investigating the duality of human-machine communication, we enhance the capabilities of traditional attributes and soft biometric descriptors, expanding their versatility and applicability towards challenging images and large-scale surveillance datasets. We experiment with crowdsourcing human annotations using ordered and similarity comparisons, and estimating attributes from images employing a variety of state-of-the-art machine learning techniques. Our focus is on utilising a lean lexicon of global and body characteristics that are most pertinent when estimated from stand-alone surveillance footage. Significant improvements in suspect retrieval and identification performance are achieved by discovering enhanced soft biometric descriptions which represent visual trait characteristics with more precision and relevance. This work evolves the areas of soft biometrics and identity science, drawing ideas from contemporary image attribute recognition, semantic attribute discovery, pedestrian reidentification and perceptual psychology. Our findings indicate that increasing not only the volume, but the complexity of information conveyed between humans and machines is key in deploying soft biometrics ubiquitously.

Text
Final thesis - Version of Record
Available under License University of Southampton Thesis Licence.
Download (59MB)

More information

Published date: March 2018

Identifiers

Local EPrints ID: 420947
URI: http://eprints.soton.ac.uk/id/eprint/420947
PURE UUID: 111b348b-c16e-473f-9e8b-712b20227ae0
ORCID for Mark Nixon: ORCID iD orcid.org/0000-0002-9174-5934

Catalogue record

Date deposited: 18 May 2018 16:30
Last modified: 14 Mar 2019 01:56

Export record

Contributors

Author: Daniel Martinho-Corbishley
Thesis advisor: Mark Nixon ORCID iD

University divisions

Download statistics

Downloads from ePrints over the past year. Other digital versions may also be available to download e.g. from the publisher's website.

View more statistics

Atom RSS 1.0 RSS 2.0

Contact ePrints Soton: eprints@soton.ac.uk

ePrints Soton supports OAI 2.0 with a base URL of http://eprints.soton.ac.uk/cgi/oai2

This repository has been built using EPrints software, developed at the University of Southampton, but available to everyone to use.

We use cookies to ensure that we give you the best experience on our website. If you continue without changing your settings, we will assume that you are happy to receive cookies on the University of Southampton website.

×