This thesis describes our work towards a unified framework for automatic restoration of dirt and blotches in archive film. The framework we present here is composed of three stages, i.e. defect detection, false alarm elimination, and defect removal. First, we propose a novel probabilistic approach to detect defects in digitized archive film. An HMM is characterized and trained to model the statistical changes of temporal pixel transitions over several frames before and after the current frame. The trained HMM is then applied to measure how the likelihood of an unseen observation sequence being normal varies if each single observation within it was missing, one at a time in a leave-one-out fashion. The centre pixel of the observation sequence will be marked in the defect map, if the likelihood of the observation sequence without the centre pixel is larger, by a certain degree, than the average of all likelihoods. The resulting defect maps from our proposed defect detector encapsulates the true defects very well, but can suer from many false alarms. Therefore, we extend the defect detection method to add a two-stage false alarm elimination process, which is developed based on investigating the characteristics and causes of false alarms. The proposed false alarm approach first applies MRF modelling on the defect map to propagate neighbouring normal pixels into the false alarm region using spatial continuity constraints. Then, the pyramidal Lucas-Kanade feature tracker is adopted to impose temporal correlation constraints on spatially isolated false alarm regions. This helps increase the accuracy of the proposed method signicantly. Finally, we present a novel restoration method for defects and missing regions in archive films. Our statistical framework is based on Random Walks to examine the spatiotemporal path of a degraded pixel, and uses texture features in addition to intensity and motion information traditionally used in previous restoration works. The degraded pixels within a frame are restored in a multiscale framework by up- dating their features (intensity, motion and texture) at each level with reference to the attributes of normal pixels and other defective pixels in the previous scale as long as they fall within the defective pixel's random walk-based spatiotemporal neighbourhood. The proposed algorithms are compared against state-of-the-art and industry-standard methods to demonstrate their improved detection and restoration performance using our archive film restoration dataset.