Class Specific Object Recognition using Kernel Gibbs Distributions

• Autores: Barbara Caputo
• Localización: ELCVIA. Electronic letters on computer vision and image analysis, ISSN-e 1577-5097, Vol. 7, Nº. Extra 2, 2008 (Ejemplar dedicado a: Special Issue on Computational Modelling of Objects Represented in Images), págs. 96-109
• Idioma: inglés
• Enlaces
  • Texto completo  1    2  
• Resumen

  • Feature selection is crucial for effective object recognition. The subject has been vastly investigated inthe literature, with approaches spanning from heuristic choices to statistical methods, to integration of multiplecues. For all these techniques the final result is a common feature representation for all the consideredobject classes. In this paper we take a completely different approach, using class specific features. Ourmethod consists of a probabilistic classifier that allows us to use separate feature vectors, selected specificallyfor each class. We obtain this result by extending previous work on Class Specific Classifiers andKernel Gibbs distributions. The resulting method, that we call Kernel-Class Specific Classifier, allows usto use a different kernel for each object class by learning it. We present experiments of increasing level ofdifficulty, showing the power of our approach.