Automated File-Based Quality Control: A Machine-Learning Approach

摘要

In recent years, broadcasters successfully introduced file-based workflows to improve production efficiency. However, they are increasingly dealing with a proliferation of file formats, and many of them still have large archives that need to be digitized for reuse. To guarantee trouble-free workflows and long-term preservation in this quickly evolving digital domain, it is essential that media files adhere to well-described, established standards. Furthermore, their audiovisual quality should be up to broadcast level. A variety of content analysis tools checking container and encoding formats, as well as audiovisual quality, are available but often hard to configure, and frequently provide difficult-to-interpret results. In this research, a learning algorithm takes into account the results of several sources of content analysis to perform a reliable automatic interpretation, which is communicated as a traffic light decision to an operator who can then take further action if necessary. Thus, valuable time and money can be saved.