Skip to Main Content
When a picture is worth 1,000 harmful words.

Most content moderation solutions can easily identify, classify, and filter disruptive text content. However, one subversion that is challenging to moderate is text embedded in images. Images, like memes, can be embedded with toxic language and phrases that artfully dodge even the most sophisticated content moderation tools. So just how prevalent are text-heavy images? Let’s take a look. 

The Prevalence of Text-embedded Images in Communities 

While memes have been around for a long time, their prevalence catapulted to a new height during the pandemic of 2020. According to Instagram, nearly 1 million posts mentioning “meme” were shared every day in 2020. If you’re following the math, that’s nearly 365 million memes posted on Instagram in 2020 alone! 

Gaming platforms also fall prey to the creative work of malicious offenders looking to secretly upload toxic content. With the ability to create skins and clothing for their avatars, players can upload harmful content that can sneak through many platforms where text detection is not present. What’s even worse are the images that get through in player-to-player messaging. A content moderation system with Visual-AI elements, like object detection will extract or eliminate obviously offensive items, but it will likely miss images with harmful text content hidden in or on them. 

Community Sift’s OCR Text Classification 

Community Sift can now read and classify text detected within an image.  Whether the text is overlayed on the image like the classic meme example, or naturally occurring such as words on a sign or billboard, Community Sift’s OCR feature can now identify, extract, classify the text and return a pass or fail rating based on the Community Guidelines as defined in your text policy guide. 

The ability to extract and classify text found within images is important to ensure harmful content and hate speech can be identified within this content before making its way onto your platform. Technology can often detect text that is not easy to spot by a human moderator such as very small text or text visually like the background. 

Additionally, Community Sift’s OCR capability can extract the text from sources such as passports, driver’s licenses, photo IDs, social security cards and bank cards, which can help your community protect itself from fraud and other illegal activity. 

This revolutionary technology can help ensure your community is safer than ever before.  

Schedule a free demo today
We’d love to walk you through the new Community Sift OCR feature in more detail to see how it might help make your life easier and boost your team’s performance. Contact our team today. 

Request Demo