deep-doLCE: A New Machine Learning Approach for the Color Reconstruction of Digitized Lenticular Film

Dr. Giorgio Trumpy,giorgio.trumpy@fiwi.uzh.ch

Dr. David Pfluger, senior researcher

Some of the first home movies in color were shot on 16 mm lenticular film during the 1920s to 1940s. This very special film is embossed with a vertical array of hundreds of tiny cylindrical lenses that allowed to record color scenes on a black&white silver emulsion. The most efficient approach to obtain digital color images from these historical motion pictures is to scan the silver emulsion in high-resolution and let a software extract the encoded color information. The localization of the lenticular screen is the first and most complicated step of this process. A 'classic' signal processing method proved to deliver successful results in some cases, but more often adverse factors—damaged or warped film, scanning problems—hinder the successful localization of the lenticular screen.

The deep-doLCE project explores a more advanced and robust method, using an already available big dataset of digitized lenticular films to train a new deep learning software. The aim is to create an easy-to-use software that revives awareness of the lenticular color processes thus making these precious historical color movies available again to a public and securing them for posterity. 

Hasler Foundation