I. Introduction
With the rapid development of emerging immersive media, the applications of Light Field Image (LFI) are becoming increasingly widespread, such as view rendering, post-capture refocusing, occlusion rendering, and free viewpoint video services [1], [2], [3], [4]. However, LFIs still face numerous challenges in practical applications. For example, the volume of data in an LFI is quite huge, which brings significant difficulties for storage and transmission. Typically, performing sparse viewpoint sampling and compression operations on LFI is essential, but LFI compression and reconstruction processing inevitably lead to its quality degradation. To address these issues, several high-performance quality metrics have been developed [5], [6], [7], [8], [9]. Another issue that needs to be noted is that light field cameras can only capture the scene content with narrow Field Of View (FOV) due to the limitations of hardware devices, for example, the imaging FOV of some light field camera is only with 60°×43°. In other words, during the process of LFI acquisition, light field cameras are unable to capture wide FOV scene content in one shot. Presently, there exist two common schemes to expand the FOV: 1) The hardware system of the light field camera should be modified and upgraded, which is relatively costly, technically challenging, and may potentially result in increased camera size; and 2) From the perspective of computational photography, LFI stitching technologies are utilized to expand the FOV. Relatively speaking, the latter choice is relatively simple, efficient, and easy to implement. Consequently, LFI stitching is more widely used in actual applications.