In this paper, an adaptive spatio-temporal approach for the detection of dirt in archive film sequences is proposed. By combining useful features from conventional stand-alone spatial and temporal approaches this method achieves better performance in terms of both correct dirt detection rate and false alarm rate.
