Perceptual Colour-based Geolocation of Human Trafficking Images for Digital Forensic Investigation

Authors: Herrmann, Jessica; Bamigbade, Opeyemi; Sheppard, John and Scanlon, Mark

Publication Date: November 2024

Publication Name: 2024 Cyber Research Conference - Ireland (Cyber-RCI),

Abstract:

This paper investigates the effectiveness of colour-based descriptors in Content-Based Image Retrieval (CBIR) and examines the impact of various parameters on image matching accuracy. The aim is to improve image retrieval methods to support digital forensic investigators in human trafficking cases. Colour values are used as key components to describe specific image characteristics and the technique is evaluated on the Hotels-50K dataset. The method achieved a Top-50 accuracy of over 95%, enabling efficient data triage and significantly reducing the volume of images to be examined. Using 2 and 10 colour descriptors is found to optimise the balance between information richness and dimensionality reduction. Performance is further improved by optimised image selection, reducing false-positive rates, and increasing robustness. The approach demonstrates potential in advancing image analysis tools in human trafficking investigations and other contexts, opening new avenues for using colour values in crime detection and image data analysis. Future research may refine the Euclidean distance method used in the image similarities measure by introducing weighted distance measurements to reduce the impact of common colour values, and investigate lighting and saturation effects.

Download:

Download Paper as PDF

BibTeX Entry:

@inproceedings{herrmann2024colourbasedgeolocation,
author={Herrmann, Jessica and Bamigbade, Opeyemi and Sheppard, John and Scanlon, Mark},
title="{Perceptual Colour-based Geolocation of Human Trafficking Images for Digital Forensic Investigation}",
booktitle={2024 Cyber Research Conference - Ireland (Cyber-RCI)},
year=2024,
pages = {},
month=11,
publisher={IEEE},
abstract={This paper investigates the effectiveness of colour-based descriptors in Content-Based Image Retrieval (CBIR) and examines the impact of various parameters on image matching accuracy. The aim is to improve image retrieval methods to support digital forensic investigators in human trafficking cases. Colour values are used as key components to describe specific image characteristics and the technique is evaluated on the Hotels-50K dataset. The method achieved a Top-50 accuracy of over 95%, enabling efficient data triage and significantly reducing the volume of images to be examined. Using 2 and 10 colour descriptors is found to optimise the balance between information richness and dimensionality reduction. Performance is further improved by optimised image selection, reducing false-positive rates, and increasing robustness. The approach demonstrates potential in advancing image analysis tools in human trafficking investigations and other contexts, opening new avenues for using colour values in crime detection and image data analysis. Future research may refine the Euclidean distance method used in the image similarities measure by introducing weighted distance measurements to reduce the impact of common colour values, and investigate lighting and saturation effects.}
}