Advances in Light Field Salient Object Detection: A Comprehensive Survey

Abstract Salient object detection (SOD) has garnered significant attention in computer vision research, particularly in recent years. Light field (LF) data, which capture both spatial and angular information, provides a richer context for detecting salient objects than traditional RGB images. This survey provides a systematic review of LF SOD, starting with an explanation of the fundamentals of LFs, covering their theoretical foundations and various acquisition techniques. We conduct an in-depth analysis of LF SOD model architectures, highlighting advancements in both classical and deep learning-based approaches. Additionally, we present a comprehensive overview of existing LF SOD datasets, discussing their characteristics, limitations, and suitability for evaluation. Additionally, we present a detailed overview of current LF SOD datasets, discussing their features, limitations, and how appropriate they are for evaluation. To ensure robust assessment, we examine the commonly used evaluation metrics and loss functions. Through quantitative and qualitative analyses, we assess the strengths and weaknesses of current LF SOD methods, identifying key research gaps. Based on these insights, we outline critical limitations and propose future research directions to address existing challenges and advance the field.