Advanced integral imaging display system based on high-resolution simplified light field image acquisition method

Abstract

In this report, we proposed an advanced integral imaging 3D display system using a simplified high-resolution light field image acquisition method. A simplified light field image acquisition method consists of a minimized number of cameras (three cameras placed along the vertical axis) to acquire the high-resolution perspectives of a full-parallax light field image. Since the number of cameras is minimized, the number of perspectives (3×<i>N</i>) and the specifications of the 3D integral imaging display unit (<i>N</i>×<i>N</i> elemental lenses) cannot be matched. It is possible to utilize the additional intermediate-view elemental image generation method in the vertical axis; however, the generation of the vertical viewpoints as many as the number of elemental lenses is a quite complex process and requires huge computation/long processing time. Therefore, in this case, we use a pre-trained deep learning model, in order to generate the intermediate information between the vertical viewpoints. Here, the corrected perspectives are inputted into a custom-trained deep learning model, and a deep learning model analyzes and renders the remaining intermediate viewpoints along the vertical axis, 3×<i>N</i> → <i>N</i>×<i>N</i>. The elemental image array is generated from the newly generated N×N perspectives via the pixel rearrangement method; finally, the full-parallax and natural-view 3D visualization of the real-world object is displayed on the integral imaging 3D display unit.