A Copula Based Unsupervised Domain Adaptation for Image Classification

A Copula Based Unsupervised Domain Adaptation for Image Classification

In this paper, we propose an unsupervised domain adaptation algorithm for image classification using principal component analysis (PCA) and Gaussian copula function alignment. The method is similar to the CORAL algorithm which aligns the correlation structure between source and target domains, but different in that it applies correlation alignment in copula spaces instead of the original variable spaces. Since a copula function enables us to analyze separately the dependency structure from the marginal distributions, the proposed algorithm is considered to be robust to a severely skewed characteristic of the marginals that can distort the correlation structure among the variables. We compared several feature level domain adaptation algorithms for image classification using office-caltech10 data set, and verified the proposed method showed better classification accuracy in an unsupervised domain adaptation framework.