A method for discriminative dictionary learning with application to pattern recognition

RUFINER, HUGO LEONARDO; ROLÓN, ROMAN; SPIES, RUBEN; DI PERSIA, LEANDRO EZEQUIEL

Comodoro Rivadavia

Congreso; VI Congreso de Matemática Aplicada, Computacional e Industrial; 2017

Asociación Argentina de Matemática Aplicada, Computacional e Industrial

Pattern recognition is a scientific discipline whose purpose is the classification of objects into different categories or classes. Object categorization deals with the detection or recognition of ?generic? categories, reason for which it known as ?generic object recognition?. In this article, sparse representation of signals in terms of adiscriminative multi-class dictionary for image recognition is presented. A sparse representation approximates an input signal over a linear combination of a few atoms of the given dictionary. A balanced set of input signals selected from the Caltech 101 database is used for learning the discriminative dictionary. The sparse vectors are then used asinput of a multi-class classifier. The proposed method shows improvements over the standard KSVD method.