METHOD FOR MULTICLASS CLASSIFICATION IN OPEN-SET SCENARIOS AND USES THEREOF

摘要

The proposed method is used for classification in open-set scenarios, wherein often it is not possible to first obtain the training data for all possible classes that may arise during the testing stage. During the test phase, test samples belonging to one of the classes used in the training phase are classified based on a ratio between similarity scores, as known correct class and test samples belonging to any other class are to be rejected and classified as unknown.

主权项

1. Method for multi-class classification applied to open-set scenarios, divided into training phase (1, 2 and 3) and testing phase (4, 5, 6, 7), characterized in that based on a ratio R=δ1/δ2 between similarity scores and comprising the following steps:
dividing (1) a set of n training samples into n classes of interest; training a multiclass classifier (2) using the samples divided into n classes of interest, so as to obtain the intrinsic parameters; acquiring a parameter of rejection (3) T through parameter optimization; receiving a test sample (4) s; finding the two classes (5) c1, c2 more similar to the test sample S between n trained classes; calculating the ratio R=δ1/δ2 of the two best similarity scores (6), being δ1 the similarity/distance score from the test sample S to the best class c1 and δ2 the similarity/distance score from the test sample S for the second best class c2; checking (7) if the ratio R of the two best similarity scores (6) is greater than the rejection parameter T (3).