The First Draft coalition have published a series of ‘download and print out’ guides to verifying photos or video footage. A one sheet photo verification guide can be found here and the video verification guide here. Further versions and a project description can be found on Medium here. The guides form part of First Drafts collection of articles and case studies on the theme of ‘handling eyewitness media’, all published together on the ‘About First Draft News’ home page.
Just added to the collection is a piece about an interesting project by MIT research student Cynthia Fang which attempts to automate the video verification process. Introducing her project at a recent TechRaking event at MIT, she explained why searching for uploads of video footage is different to searching for images;
‘…reverse image search tools from TinEye and Google both widely used by newsrooms to trace a picture’s online history. Upload the image in question, or just copy and paste a URL, and both tools will show where that or similar images may have appeared on the Internet before. Easy. But reverse video search would be too “computationally expensive” said Fang. Searching for the thumbnail of an image may help, but analysing all the frames in a video and all the frames in all potential matches is just not practical or efficient, so other methods are necessary’.
The title of the article; ‘Automating the video verification process is difficult but not impossible’, rather indicates that the project is not yet complete. However she outlines an innovative approach to verification, which uses the available URL to initiate a keyword search through her prototype programme which returns a list of likely entities which can then be re-searched for further digital clues.
Finding the earliest example of a video is just one step, the first step, in identifying whether it is legitimate and will never fully determine a content’s veracity by itself. Fang still wants to improve the search function, adding other video sites beyond YouTube and the capability to search in different languages, but speeding up the process and reducing the need to view horrific images over and over could go a long way in helping newsrooms in their verification workflow.