A Multi-scale Nonparametric/Parametric Hybrid Recognition Strategy with Multi-category Posterior Probability Estimation

The synthesis of an effective multi-category nonlinear classifier with the capability to output calibrated posterior probabilities to enable post-processing is of great significance in practical recognition situations in that the posterior probability reflects the assessment uncertainty. In this paper, a multi-scale nonparametric and parametric hybrid recognition strategy is developed for this purpose. Based on the binary tree representation for nested structure, a new nonlinear polychotomous classification algorithm with the capability of estimating posterior probability is developed on the strength of kernel learning and Bayesian decision theory. In particular, by capitalizing on the intrinsic conexus between hierarchical structure and multi-scale analysis, the polychotomous multi-scale Bayesian kernel Fisher discriminant is implemented for building the classifier at different scales for different levels. Finally, the performance of the proposed classification and posterior probability estimation algorithm is validated by designing a multicategory Bayesian kernel Fisher discriminant classifier for a satellite images dataset.