Machine learning approach for face image retrieval

Authors:
Dianhui H. Wang;Paul Conilione
Affiliations:
La Trobe University, Department of Computer Science and Computer Engineering, 3086, Melbourne, VIC, Australia;La Trobe University, Department of Computer Science and Computer Engineering, 3086, Melbourne, VIC, Australia
Venue:
Neural Computing and Applications - Special Issue on ICONIP2010
Year:
2012

Citing 0
Cited 1

Fault detection and diagnosis of non-linear non-Gaussian dynamic processes using kernel dynamic independent component analysis

Information Sciences: an International Journal

Quantified Score

Hi-index	0.00

Visualization

Abstract

Face image retrieval (FIR) is useful to many domain applications, such as helping police to catch criminals or managing householders. However, little research has been done that uses individual face features for image comparison and retrieval. This paper aims to develop a machine learning approach for face image retrieval based on the local face features of the eyes, nose, and mouth. Neural networks are used to localise facial features, and to implement a learning pseudo metric (LPM) to filter out irrelevant images for retrieval efficiently based on semantic information. Our FIR system performs below average given traditional performance measures, but inspecting actual retrieved images it shows strong promise. It is observed that the LPM semantic filtering method was found to reduce the database size by up to 50% without a significant reduction in retrieval performance.