Título: Critérios de Divisão Espacial para a Rede Neural T-CombNET aplicada ao Reconhecimento de Gestos
Autores: Lamar, Marcus V.; Bhuiyan, Md. Shoaib; Iwata, Akira
Resumo: A neural network structure dedicated to the time series recognition, T-CombNET, is presented. The model is developed from the large scale Neural Network CombNet-II, designed to deal with very large vocabulary for Chinese and Japanese character recognition. Our specific modifications of the original CombNet-II model allows it to do temporal analysis and to be used in large set of human movements recognition systems. In T-CombNET structure, one important setting is to define the input spaces that optimize the network performance. In this work a Biologically Motivated Criterion and an Interclass Distance Measurement based criterion are analyzed and applied to select the input spaces in the T-CombNET. The obtained results show that the use of JIDM based criterion improves the classification capability of the network. The T-CombNET is applied in a Japanese finger spelling recognition system, obtaining 99.4% in a user dependent hand gesture recognition system and 96.5% for a user independent.
Código DOI: 10.21528/CBRN2003-067
Artigo em PDF: 6CBRN_067.PDF
Arquivo BibTex: 6CBRN_067.bib