Generation of N-parametric appearance-based models through non-uniform sampling

Abstract

In this work, a generalization of non-uniform sampling technique to construct appearance-based models is proposed. This technique analyses the object appearance defined by several parameters of variability, determining how many and which images are required to model appearance, with a given precision ε. Throughout non-uniform sampling, we obtain a guideline to spend less time on model construction and to diminish storage, when pose estimation no matters. The proposed technique is based on a scheme of N-linear interpolation and SSD (Sum-of-Squared-Difference) distance, and it is used in conjunction with the eigenspaces method for object recognition. Experimental results showing the advantages are exposed. © Springer-Verlag 2004.