Discriminative transfer feature learning based on robust-centers

•We illustrate the influences of outliers and sample-based distances in UDA.•Robust class-centers and domain-centers are used to reduce the influence of outliers.•We minimize MMD with robust centers to steadily align the source and target domains.•Three distances using robust centers are created to reduce the number of distances.•DTFLRC efficiently optimizes domain, intra-, and inter-class discrepancies.•Sufficient experiments demonstrate that DTFLRC outperforms the advanced methods.