We consider the problem of separating simultaneous source blended data in applied seismology. Cross-source interference that masks the desired signal in the common source domain can be translated into incoherent noise by rearranging the data in the common receiver domain. We show that applying a virtual blending/deblending process to the data in the common receiver domain enables obtaining an additional noisy version of the data. By measuring the local similarities and dissimilarities between the two noisy versions of the data, it is possible to discriminate between corrupt and non-corrupt data points. Corrupt data points can be replaced by a weighted sum (e.g., averaging) of neighboring non-corrupt data points. The proposed method is applied directly in the time-space domain, i.e., no computationally expensive data transformation is needed. Moreover, it can be straightforwardly extended to higher-dimensional data scenarios. Simulation results are given to validate the effectiveness of the proposed method.