Регулируемая селективность в многомодальном распознавании образов

The problem of multi-modal pattern recognition is considered under the assumption that the kernel-based approach is applied, which assumed that within each particular modality a kernel function can be specified. The danger of over-fitting makes it necessary to truncate the set of initially available madalities. Two known wrapper-based kernel fusion techniques, Relevance and Support Kernel Machines, offer a toolkit of combining patter recognition modalities. In this paper, we propose the modications of the fusion techniques equipped with the ability to preset the desired level of feature-selectivity.