Portfólio de Tecnologias – Inova Unicamp

Method for multiclass classification in open-set scenarios and uses thereof

NOVELTY – The method involves dividing a set of n training samples into n classes of interest. A multiclass classifier is trained (2) using the samples divided into n classes of interest to obtain the intrinsic parameters. A parameter of rejection through parameter optimization is acquired and a test sample is received (4). The two classes more similar to the test sample between n trained classes are found (5) and the ratio of the two best similarity scores are calculated (6). The ratio of the two best similarity scores greater than the rejection parameter is checked (7). USE – Method for multi-class classification applied to open-set scenarios such as fingerprint recognition, face recognition, speech recognition, object recognition, scene recognition, character recognition, actions recognition, remote sensing image classification and others general pattern recognition application. ADVANTAGE – The method extends multiclass classifiers to properly and efficiently handle the open-set scenario that is to search for solutions in which the risk of the unknown is minimized. Using method for classification of the views, the system of the machine is able to use the available information of the views to construct the model automatically, without bothering the operator about the type of the view. The training data error and the open space of risk are minimized. DESCRIPTION OF DRAWING(S) – The drawing shows a flowchart illustrating the steps of the method for multi-class classification applied to open-set scenarios. Step for training multiclass classifier (2) Step for receiving test sample (4) Step for finding two classes (5) Step for calculating the ratio of the two best scores (6) Step for checking whether ratio is greater than the rejection parameter (7)

Main Application Field

T01 (Digital Computers); T04 (Computer Peripheral Equipment)

INVENTORS:

RIBEIRO MENDES JÚNIOR PEDRO
MEDEIROS DE SOUZA ROBERTO
DE OLIVEIRA WERNECK RAFAEL
VECCHIA STEIN BERNARDO
WATANABE DANIEL
RODRIGUES DE ALMEIDA WALDIR
AUGUSTO BIZETTO PENATTI OTÁVIO
DA SILVA TORRES RICARDO
ROCHA ANDERSON

794_OPENSET

Patent number: US2016092790-A1;BR102014023780-A2

PATENT STATUS:

For information contact Inova Unicamp

FOR ADDITIONAL INFORMATION:

parcerias@inova.unicamp.br

+55 (19) 3521-5207 / 2607

This technology profile has been automatically generated.

COMPARTILHE:

TECNOLOGIAS RELACIONADAS:

[qodef_portfolio_list type=”gallery-space” image_size=”full” show_load_more=”no” order_by=”date” order=”DESC” category=”engenharia-eletrica” number=”3″ columns=”three” filter=”no”]