3D Surface Reconstruction Based On Plenoptic Image

Abstract

Light- field cameras, also known as plenoptic cameras, have recently become available to the consumer market. Plenoptic cameras can collect the 4D light field information and represent it in a single photograph. This thesis mainly concerns distance estimation based on the photographs taken by light-fi eld camera. A light-fi eld camera is implemented by placing a micro-lens array the main lens and micro-sensors. Each micro-lens record the light information arriving along the rays, not only the amount of the light also the direction. The light- field picture contains more than 2D information, which includes not only the spatial data, but also the angle data. Light-fi eld images record a 4D matrix which contains intensity, position and direction of incoming rays. With this information one can refocus the image with different focal depths to generate a focal stack plane. Each refocused image is a simple 2D image, and the focal stack becomes 3D based on a stack of 2D refocused images. This information can be used to estimate the distance of different objects in the original image. We also combined a segmentation algorithm into deconvolution processing to improve the performance of deconvolution. Various images with different characteristics were used to test the performance and to compare to other depth estimation techniques.