Attribute Tri-factorization Model with Regularization of Relation Digraph
Aiming at the problems of incomplete mapping between attributes and features,as well as the insufficient mining of the attribute space structure in zero-shot image classification,an attribute tri-factori-zation model with regularization of relation digraph was proposed.Firstly,the mapping between attribute space and feature space was achieved by matrix tri-factorization of attributes.Secondly,the attribute rela-tion digraph was constructed using the weight matrix.Finally,the similarity between the testing sample and each testing class was calculated in either the attribute space or the feature space to finish image clas-sification.Experimental results on aPY and SUN datasets showed that the proposed model was capable of efficiently improving the accuracy of zero-shot image classification.
万方数据
维普
