Role of Redundancy in Algorithms of Digital Film Restoration

In our presentation we focus on new technology developed for digital restoration of degraded movies. A low budget (but high quality) restoration project requires reliable automatic solutions for content improvement and enhancement. Restoration of audio tracks is now fairly well resolved as a problem, but effective automated solutions for the moving image remain elusive. In the case of image restoration, film damage is modeled mathematically and may be identified as incommensurate data in the context of a highly redundant pattern system. The problem of the damaged image is very similar to the problems related to Error Correcting Codes (ECC). You can find ECCs in any cell phone, and in many other digital devices such as satellite and cable boxes. The data in ECCs undergo encoding with the known (usually linear) model of redundancy, and any error in the encoded data violates the model. This means that in the most cases the error can be found and corrected.  However, in the case of films the model of redundancy is unknown; it has to be identified and described mathematically. Since the quality of restoration is directly related to the precision of the model, we have focused our efforts on developing the most accurate modeling of the characteristic and inherent redundancy of motion image material (cinema and video).

We will explain the basic principles of the modeling system we have developed, and demonstrate automated solutions for elimination of common, obtrusive defects such as dust, dirt, scratches, flicker, and video defects. These mathematical models take human perception into account into reconstruction process, and permit a wide range of human intervention. Because of the importance of factors including perception and the historical context in the practice of archival restoration, our process is designed to afford a maximum of control and aesthetic choice to restorers and curators. In other words, the automated function of the system allows for efficient and economical processing of damaged images, but remains sensitive to the requirements of human users.

During the presentation, several unique clips from different film archives highlighting automatic defect removal will be screened.

Joint work with I. Kozlov (Algosoft Tech USA)