Dynamic LVQ Models for Classification of Spatiotemporal Patterns

Título: Dynamic LVQ Models for Classification of Spatiotemporal Patterns

Autores: Monteiro, Isaque Q.; Barreto, Guilherme A.; Nascimento, Patrícia V.

Resumo: This paper proposes the combination of three short-term memory (STM) mechanisms with the Learning Vector Quantization (LVQ) model for classifying spatiotemporal patterns. The goal is to investigate the ability of these dynamic models to acquire neural representations of faces that are invariant to changes in images caused by the movement of the subjects. The proposed models are evaluated by their ability to recognize faces in sequences of images, as well as by their sensitivity to memory parameters and image noise. A simple theoretical analysis for understanding the discriminative power of the proposed spatiotemporal classifiers is also provided. Through simulations, it is shown that the dynamic variants of LVQ perform considerably better than the static LVQ model, achieving classification rates similar to those obtained by a dynamic MLP network.

Palavras-chave: Learning vector quantization; face recognition; spatiotemporal classifiers; short-term memory

Páginas: 6

Código DOI: 10.21528/CBRN2005-073

Artigo em PDF: CBRN2005_073.pdf

Arquivo BibTex: CBRN2005_073.bib