Geometry-Aware Light Field Angular Super Resolution Using Multiple Receptive Field Network

Since a light field camera acquires a set of images with multiple views, it has a fundamental limitation of spatial and angular resolution due to the limited sensor resolution and a number of lenses. This paper presents a light field angular super resolution method using a multi-receptive field network. The proposed network consists of two main modules, depth estimation and light field blending modules. Specifically, the proposed depth estimation module with hierarchical multiple receptive fields can accurately estimates flow maps for image warping. Experimental results demonstrate that the proposed method generates more improved results than existing methods in the sense of quantitative evaluation. For that reason, the proposed network can be applied to various light field applications.