Premium
Classifiers sensitive to external context – theory and applications to video sequences
Author(s) -
Rafajłowicz Ewaryst
Publication year - 2012
Publication title -
expert systems
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.365
H-Index - 38
eISSN - 1468-0394
pISSN - 0266-4720
DOI - 10.1111/j.1468-0394.2010.00564.x
Subject(s) - computer science , context (archaeology) , curse of dimensionality , class (philosophy) , artificial intelligence , position (finance) , machine learning , bayesian probability , artificial neural network , pattern recognition (psychology) , hidden variable theory , bayesian network , paleontology , physics , finance , quantum mechanics , economics , quantum , biology
An external context like weather conditions, lighting, etc. influences classification results, but it is frequently omitted in a mathematical model of the problem at hand. Our aim is to propose a mathematical model, which extends the Bayesian problem of pattern recognition by incorporating external context variables. They are implanted as functions, which influence parameters of class distributions. We prove that context variables influence a shape or a position of the optimal class separating surface, without enlarging the dimensionality of a pattern space. Thus, one can treat the proposed extended Bayesian model as a fusion of patterns and external context variables, embedded into the same pattern space. Then, learning algorithms for neural network classifiers are proposed, which take context variables into account.