Nonlinear pyramids for object identification

Image pyramids constructed via nonlinear filtering and subsampling are investigated for object identification and tracking task. Two nonlinear structures, the morphological pyramid and the anisotropic diffusion pyramid are used in coarse-to-fine target recognition algorithms. The background and theoretical development of the pyramidal strategies are presented, and important implementation decisions are discussed. Particularly, the analysis focuses on the sampling schemes and the selection of the pyramid root level for target identification. Experimental results are provided that demonstrate the performance of both nonlinear pyrimidal techniques on noisy infrared image sequences. The results show that the morphological and anisotropic diffusion pyramids allow reliable, efficient extraction of features for rapid object identification.