Abstract / Summary
Abstract Among the various reflective imaging modalities, eye reflections have emerged as a valuable source of visual information for person and scene analysis. Such imagery can capture contextual and identity-related cues from reflected environments, supporting applications in forensic analysis, biometric recognition, intelligent surveillance, and safety-critical domains. However, analysis of persons and scenes using eye-reflected images remains challenging due to their inherently small spatial extent, optical distortions, motion blur, illumination variability, and limited image resolution. Although machine learning techniques have demonstrated potential in addressing these challenges, existing research remains fragmented across different tasks, approaches, and application domains. While substantial research has investigated face identification and computer vision tasks using direct images, no systematic review has comprehensively synthesised machine learning-based approaches for person and scene analysis using eye-reflected images. This study addresses this gap by critically analysing machine learning techniques for person and scene analysis using eye-reflected images. A total of 12 primary studies, published between January 2006 and August 2026, were systematically identified and selected from Q1 and Q2 journals. The systematic review was conducted in accordance with the guidelines proposed by Kitchenham and Charters and the PRISMA 2020 framework. The review examined eye-reflected image sources, datasets, preprocessing techniques, machine learning models and architectures, analysis tasks, evaluation metrics, and application domains. A taxonomy of machine learning-based person and scene analysis is proposed to organise and synthesise this research area. The review identifies key research gaps and limitations and outlines future research directions to advance person and scene analysis using eye-reflected images.