One Button Restoration – Big Data Meets Media Archives

The new technology developed by Algosoft was presented at The Reel Thing 2015 by
Mike Inchalik, Alexander Petukhov and Inna Kozlov.

In our presentation we describe a new image processing technology developed for the digital restoration of deteriorated motion picture film and display our resulting images. After years of research, we have found a way to apply and extend state-of-the-art algorithms used in “big data” analysis into a unique solution for the automatic digital restoration of motion pictures.

This new technology is very important to motion picture archives because up until now, digital restoration has only been practical for a relatively small number of films. There are massive quantities of content in archives around the world, and without high levels of automation and efficient processing, restoration is beyond the practical timescales and budgets of many.

We will show how new “adaptive” mathematical algorithms avoid the time-consuming cycles of manual fine-tuning and re-processing required by the digital restoration tools we all use today. These new algorithms do the fine-tuning themselves, and completely automatically. The algorithms do this very accurately, measuring the levels of grain, flicker, and imperfections for every frame and every color channel before automatically calculating nearly 50 parameters precisely for each shot. In most cases, this reduces most of today’s complicated digital film restoration processes to a simple click of a single button.

No digital restoration technology is practical if the final levels of grain and other aspects of the “look” are not under the direct control of the archivist who knows the aesthetics that must guide each restoration. These new algorithms allow just such control, and “after the fact” so that the final appearance of the project can be fine-tuned quickly and easily, after the unwanted artifacts have been removed automatically.

The bulk of our tests to date show that this new technology makes highly automated, high-quality image improvement possible for a very wide spectrum of motion picture footage – from the earliest cinematographic paper prints to footage shot with the latest digital professional and consumer cameras. These new algorithms are fast too- with processing speeds approaching real-time at 2K when running on powerful NVidia GPUs.

We will present an overview of the technology, the workflow and share before-after results.