Meet Verified Pixel: A ‘visual spellcheck’ for verifying images in news
(This article was written by Alastair Reid, Managing Editor of First Draft News. First Draft News is a daily destination site for journalists who source and report stories from social media. The article was originally published on First Draft on January 15, 2016. Reblogged with permission.)
It seems every major news story now brings with it a minefield of fake pictures for newsrooms to sort through, a minefield that grows ever bigger each year.
Now every time a big story breaks many of these images are re-used by people in the wrong context, whether that is to paint Syrian refugees as terrorists, mislead reports from Eastern Ukraine or more generally mislead news organisations and the public.
Knowledge of verification skills is growing, but how can journalists keep up with the fakers and sort the wheat from the chaff quickly? A new prototype for newsgathering and verification aims to help them do just that.
“The base assumption was how could we combine many known verification tests into one simplified workflow to help automate the process quickly,” said Sam Stewart, project lead of Verified Pixel, fresh out of its funding period from the Knight Foundation.
Along with Sam Dubberley, co-founder of Eyewitness Media Hub, and Douglas Arellanes, co-founder of media technology organisation Sourcefabric, Stewart spent a large part of 2015 developing the prototype. They invited users to experiment with the dashboard in the hope it will speed up the verification process and let anyone, no matter their experience, check images automatically.
“The idea is for a reliability of verification every time it comes into the newsroom,” Dubberley said.
“I sometimes call it a visual spellcheck,” said Stewart. “Everyone is running spellcheck because it’s built into the platform, but there could be a day where everyone is performing verification because it’s automated and very easy to do. This is our foray into attempting to get there.”
Verified Pixel conducts a number of checks on any new image added to its database: checking Google Images and TinEye to see if and when the picture has appeared online before; scanning the image file for EXIF data to understand when, where and on what device it was captured; and running it through image forensics tool Izitru to see if the image has been altered.
The images are then displayed in a dashboard where users can review the checks and add tags or comments if necessary. For the uninitiated, there is a glossary of relevant terms, functions and concepts.
Images appear in the dashboard chronologically as standard, with tabs below each to show how they performed in particular tests. Screenshot: Verified Pixel
“[The current checks] form our minimum viable product,” said Arellanes, “although we’re looking to implement a lot more and there are more and more useful services coming up all the time.”
Cross referencing date and location metadata with historical weather information from Wolfram Alpha is one possibility, he said, as is adding any machine learning or optical character recognition tools to help identify any people, locations or text in an image.
But where much of the verification around eyewitness media focusses on social media sources, the Verified Pixel team streamline the workflow in newsrooms further by automatically extracting images emailed to a specific address. As such, news organisations can ask their audience for submissions which are collected in the dashboard with notes and checks already completed.
“Anyone’s test works best the closer you get to the original file,” said Stewart, stressing that integrating input from social media is still possible, but “the metadata is not stripped out [as happens with most social networks] and there’s more information in the file as it becomes the closer-to-camera original.”
During the initial tests, journalists used the prototype to verify images from the aftermath of the Nepal earthquake, said Stewart, or the refugee crisis on the border between Hungary and Serbia, and the conflict in Donetsk.
Users can go into detail on the status of checks on an image, and leave notes for colleagues. Screenshot: Verified Pixel
But Verified Pixel is not alone in this field, of course. Tools like SAMDesk are already carving out a space in newsrooms as a way to combine the newsgathering and verification process of social media images, while Verifeye Media launched in beta last year to simplify the verification process for newsrooms and pay photographers for their work.
Verified Pixel is fully open source, but there are associated costs with running the software. The team are happy to make “as many individual accounts as people are interested in” for their prototype, said Dubberley, but a next step is to speak to organisations about the possibility of testing in a newsroom with a high volume of submitted pictures. What journalists do with the results is up to them.
“What we really hope is there will never be a fully-automated solution,” said Stewart, “but there will definitely be deep in-roads to automating a solution and there will be someone at the end making the final decision.
“They’re making that decision with much more information provided to them at an exponentially quicker pace than there was before. That’s the hope.”
“Our whole goal with this thing is to take the drudgery out of verification,” said Arellanes, “so journalists can concentrate on the stuff, the journalism, that will make a difference.”