Software Sifts Photos For Most Clickable

As we have seen repeatedly since the glory days of dotcom-era metrics like 'stickiness,' attention does not necessarily translate into sales. But in a crowded visual marketplace, differentiation is a benefit.

Presumably until everyone is using the same tools and producing similar graphics which then cancel each other out. JL

Elizabeth Dwoskin reports in the Wall Street Journal:

New software promises to enhance the power of online imagery by indexing and prioritizing pictures that grab attention. These tools are for online businesses eager to boost measures of engagement such as clicks, comments and time spent on a page. Users were viewing, liking, and commenting on photos 30% more frequently than before.
Hundreds of millions of photos and videos flood the Internet each day, a potential gold mine for online businesses that can get people to click on them. But which pictures will draw the biggest crowd and generate the most revenue? How can publishers spot the next viral sensation amid a sea of dross?
Companies doing business on the Internet know that adding images to text fuels engagement. Adding a picture to a tweet, for example, makes it five times more likely that readers will click an embedded link, according to marketing firm Shift. But the rising tide of photos and videos is ever more challenging to sift through, and figuring out which ones will draw the most attention has been more art than science.
Now a new generation of software may help websites grab a bigger piece of the $149 billion annual trade in online advertising and $1.6 trillion in e-commerce. Companies like EyeEm Mobile GmbH and Neon Labs Inc. have developed software that rifles through images to zero in on subtle qualities that draw viewers.
“This technology will reshape the visual industry,” said Stephen Mayes, a former creative director of the Getty Images photo archive who has experimented with the software. Just as Google made the Web more useful by indexing online pages, the new software promises to enhance the power of online imagery by indexing and prioritizing pictures that grab attention, he said.
These tools are the latest hope for online businesses eager to boost measures of engagement such as clicks, comments and time spent on a page. Services such as Outbrain Inc. and Taboola Inc. automatically populate pages with content that has proved popular online, but they often deliver so-called clickbait that caters to lowest-common-denominator taste: “You’ll be Surprised Which Celebrities are Still Smoking.”
The new software focuses squarely on the mining of visual data. Building on image-recognition techniques that identify objects in photos, EyeEm’s program looks for patterns common to images selected by professional photographers. Neon Labs’ software spots image characteristics shown to trigger brain activity in neuroscience experiments.

Still, Neon Labs is updating its software to recognize photos of celebrities—or, more precisely, anyone whose face appears with extraordinary frequency. “Since political season is heating up, we are working on Trump and Clinton,” said co-founder Sophie Lebrecht.
Luca Paderni, a media analyst with Forrester Research Inc., predicted that most publishers eventually would use some form of algorithmic automation to curate their collections. Automation tools trained in how the mind sees beauty could provide more interesting visual experiences than those offered by stock photography companies, he said. But like zippy web-driven headlines, the images could also become commonplace over time.
Founded in 2011 in Berlin, EyeEm has raised $24 million from investors including venture capitalist Peter Thiel. The company, which operates an online image marketplace, programmed computers to study 50,000 photos ranked by a group of experienced photographers. This so-called deep learning approach found millions of characteristics distinguishing compelling photos from less interesting ones, said Ramzi Rizk, EyeEm’s chief technology officer.
“People say beauty is a subjective thing,” Mr. Rizk said. “But a lot of it can be quantified.”
Two months after rolling out the software, users of the startup’s photo marketplace—aiming to compete with stock-photo providers like Getty Images—were viewing, liking, and commenting on photos 30% more frequently than before, Mr. Rizk said. He declined to say whether sales had improved. The marketplace, which had been open for eight weeks, hadn’t gathered enough data to make a comparison, he said.
The profusion of online video opens further opportunities. Publishers or sites like YouTube often display a single video frame, or thumbnail, to entice audience members to view the entire clip. Software built by Neon Labs combs through clips in search of the most compelling frame.
Neon Labs’ software is based on principles Ms. Lebrecht gleaned during a decade spent studying neuroscience at Carnegie Mellon University and the National Science Foundation. The brains of volunteers tracked by magnetic resonance imaging responded to photos with missing information such as a face looking off to the side rather than straight ahead, she said. They were also attracted to bright lights and to evidence of emotional connections between people—among tens of thousands of other data points research found.
“There’s this innate need to auto-complete,” she said.
Neon Labs said its customers and partners, which include the Sundance Film Festival and U.S. news companies it declined to identify, made 16% to 40% more revenue on their videos due to increased clicks. It declined to specify its pricing.While the technology could encourage Web publishers to cater to base instincts, it also could raise the quality of images in interesting ways, said Joshua Benton, director of the Nieman Journalism Lab at Harvard University. “Maybe the algorithm is as subtle as Jony Ive’s eye,” he said, referring to Apple Inc.’s legendary industrial designer. “Perhaps it’s better to have lots of Jony Ives than one.”