Structure-preserving light field reconstruction via spatial and implicit angular cues

To meet the requirement of densely sampled light fields (LFs) in various applications, LF angular super-resolution (LFASR) has been extensively studied. Current LFASR methods rarely preserve the inherent structure of the LF, often resulting in the loss of texture details and violating consistency in the parallax structure. To this end, we propose a method that leverages spatial and implicit angular cues to comprehensively capture the inherent structure of the LF. Specifically, we extract texture details using spatial cues and exploit implicit angular cues to guide the learning of the parallax structure. Then, we further mine these properties through the integration of spatial and implicit angular cues. Experimental results on both real-world and synthetic LFs demonstrate that our method surpasses state-of-the-art methods.