Three-dimensional recognition of photon-starved events using computational integral imaging and statistical sampling

We present a statistical approach to recognize three-dimensional (3D) objects with a small number of photons captured by using integral imaging (II). For 3D recognition of the events, the photon-limited elemental image set of a 3D object is obtained using the II technique. A computational geometrical ray propagation algorithm and the parametric maximum likelihood estimator are applied to the photon-limited elemental image set to reconstruct the irradiance of the original 3D scene voxels. The sampling distributions for the statistical parameters of the reconstructed image are determined. Finally, hypothesis testing for the equality of the statistical parameters between reference and input data sets is performed for statistical classification of populations on the basis of sampling distribution information. It is shown that large data sets of photon-limited 3D images can be converted into sampling distributions with their own statistical parameters, resulting in a substantial data dimensionality reduction for processing.