A cockpit of multiple measures for assessing film restoration quality

Abstract In machine vision, the idea of expressing the quality of a films by a single value is very popular. Usually this value is computed by processing a set of image features with the aim of resembling as much as possible a kind of human judgment of the film quality. Since human quality assessment is a complex mechanism involving many different perceptual aspects, we believe that such approach may scarcely provide a comprehensive analysis. Especially in the field of digital movie restoration, a single score can hardly provide reliable information about the effects of the various restoring operations. For this reason we introduce an alternative approach, where a set of measures, describing over time basic global and local visual properties of the film frames, is computed in an unsupervised way and delivered to expert evaluators for checking the restoration pipeline and results. The proposed framework can be viewed as a car or airplane cockpit, whose parameters (i.e. the computed measures) are necessary to control the machine status and performance. This cockpit, which is publicly available online, would like to support the digital restoration process and its assessment.