Segmentation of three-dimensional objects from background in digital holograms

Abstract

We present a technique for performing segmentation of three-dimensional, objects encoded using in-line digital holography from the scenes background. We create a volume of reconstructions through numerically reconstructing a digital hologram at a range of depths. For each reconstruction a variance map is created through calculating variance about a neighbourhood for each of the reconstructions pixels. We can then classify a pixel as object or background by thresholding the maximum variance of every pixel over all depths. We present segmentation results for objects of low and high contrast.