Datasets for the Quality Assessment of Light Field Imaging: Comparison and Future Directions

Abstract

With the increasing research focus on light field imaging in recent years, it has become essential for researchers in this field to either benefit from access to equipped laboratories with light field acquisition devices and displays or to have access to publicly available light field imaging datasets. Some datasets are indeed available, each with a different nature. For instance, some contain real world images or sources while others are based on purely synthetic images or sources generated by computer graphic tools; others are a combination of both. Datasets for the quality assessment of light field content include pristine light field content as well as sources affected by different levels of impairments. The latter are tested subjectively by a panel of viewers and often objective metrics are also calculated. This paper presents a comprehensive comparative review of 33 publicly available datasets that span from content-only datasets to specific task based datasets and quality assessment datasets. While our aim is to review and investigate what each dataset has to offer and which tests had been considered by their proposers, we also take the opportunity to leverage the results of previous studies to identify and discuss the challenges ahead and identify the areas with potential for improvement.

Keywords

• Light field displays
• quality assessment
• dataset
• subjective tests
• objective quality metrics