Joint Subspace and Dictionary Learning with Dynamic Training Set for Cross Domain Image Classification

Abstract

The assumption that training samples and test samples obey the same distribution is grossly violated when images are from distinct domains, which can lead to degradation in classification performance. In this paper, we propose a sparse representation based domain adaptation approach to tackle the cross-domain image classification problem. Specifically, our method aims to alleviate the distribution discrepancy and preserve the discriminative information from source domain, which is achieved by jointly learning a common subspace and a discriminative dictionary. We update the subspace and dictionary with dynamic training samples by selecting the target samples with high prediction confidence and adding them to the source sample set. We evaluate the proposed method on two cross-domain image datasets under both the unsupervised and semi-supervised domain adaptation scenarios. The experimental results demonstrate our method is more effective than competing domain adaptation methods for cross-domain image classification.