An Efficient Refocusing Scheme for Camera-Array Captured Light Field Video for Improved Visual Immersiveness

Abstract

Light field video technology attempts to acquire human-like visual data, offering unprecedented immersiveness and a viable path for producing high-quality VR content. Refocusing that is one of the key properties of light field and a must for mixed reality applications has shown to work well for microlens based cameras, but as light field videos acquired by camera arrays have a low angular resolution, the refocused quality suffers. In this paper, we present an approach to improve the visual quality of refocused content captured by a camera array-based setup. Increasing the angular resolution using existing deep learning-based view synthesis method and refocusing the video using shift and sum refocusing algorithm produces over blurring of the in-focus region. Our enhancement method targets these blurry pixels and improves their quality by similarity detection and blending. Experimental results show that the proposed approach achieves better refocusing quality compared to traditional methods.