Objective Quality Assessment of 2D Synthesized Views for Light-Field Visualization

摘要

Super multi-view light-field displays provide glasses-free immersive 3D experience via continuous motion parallax. Such displays, particularly large ones, require input views acquired at high angular resolution, covering a large field of view. Light-field content can be captured or generated by several different tools and methods, such as the turntable arrangement for a single camera, arc, linear and grid arrangements of multiple cameras, light-field cameras and also virtual cameras. However, acquiring content with a large number of real cameras is often economically and technologically prohibitive. Interpolating intermediate views from those captured using sparse camera arrays has been suggested to achieve the required high view density. Such synthesized views suffer from several artifacts, which can severely affect the perceived quality. Objective quality assessment of synthesized images has been studied extensively on a per-image basis. Yet, when such content is visualized on a light-field display, the objective quality of the content may not be derived from the quality of the individual images. In this paper, we present a pilot study where we investigated degradations in the objective quality of 2D synthesized views while increasing the proportion of synthesized views in the multi-view content.