Dirt detection for archive film restoration using an adaptive spatio-temporal approach

restoration We propose an adaptive spatio-temporal approach for the detection of dirt in archive film sequences. By combining useful features from conventional stand-alone spatial and temporal approaches our method achieves better performance in terms of both correct dirt detection rate and false alarm rate. Furthermore the incorporation of confidence weighting offers a useful control mechanism for semi-automatic restoration scenarios involving manual operator intervention. Overall our scheme offers a practical solution to the dirt detection problem without additional computational complexity overheads.