We present a method for photon-counting sensing with three-dimensional (3-D) integral imaging for object recognition using independent component analysis (ICA). A lenslet array is used to capture multiple perspective images of a 3-D scene projected onto an image sensor. Photon-counting images of the captured elemental images are generated using a Poisson distribution. A kurtosis-maximization-based algorithm is used as a non-Gaussian maximization method to extract independent features from the photon-counting training data set. The photon-counting image data are preprocessed using principal component analysis to reduce the number of dimensions, increase the speed of the ICA step, and improve the classification performance. A photon-counting image of unknown input scene is classified using k-nearest neighbor and cosine angle metrics. Experimental results are presented, and the probability of classification errors is measured.