A New Raw Holoscopic Image Simulator and Data Generation

A holoscopic imaging system is a method for capturing three-dimensional information with a single camera and aperture, using a micro-lens array between the camera's main lens and image sensor. Due to the lack of holoscopic cameras on the market, researchers need a holoscopic simulator and a dataset to work with. Most of the available holoscopic simulators and datasets generate up-sampled viewpoint images, which are extracted from the holoscopic raw image. This paper proposes a new holoscopic simulator that generates a raw holoscopic image with variable resolution more efficiently. The proposed holoscopic simulator captures a 3D model, generates viewpoint images with a proper baseline separation using a Blender add-on function, and uses a multi-view stereo algorithm to generate ground-truth depth maps. To generate a raw holoscopic image, viewpoint image pixels are remapped to their associated micro-lenses, producing two forms of holoscopic images, a grid of viewpoint images, and a grid of elemental images. The holoscopic simulator is fast and efficient compared to existing holoscopic simulators. A comprehensive holoscopic dataset is also presented, including raw holoscopic 1.0 images and associated viewpoint images at multiple resolutions, as well as a holoscopic toolbox for viewpoint images and elemental images. The proposed holoscopic simulator and dataset offer a valuable resource for researchers interested in holoscopic imaging and provide a new approach to generating holoscopic images and associated data, potentially facilitating research in related areas and advancing the field of holoscopic imaging.