WebPurify launches profanity filter for memes and images
WebPurify has launched a limited beta release of Optical Character Recognition Profanity Filter Service. It combines WebPurify’s Optical Character Recognition technology with its profanity filter solution into a single API.
The software can detect profane text in memes and other images in blogs, forums, social media apps, children’s sites, in-game chats, interactive agency campaigns and more. According to WebPurify, the proliferation of memes and images have made it more difficult to filter out profanity, as many optical character recognition programs can only extract text from documents.
According to WebPurify, its text extraction technology is trained on user-generated content, making it an effective tool for scanning content such as images and memes for offensive language. Clients can also customize a “block and allow” list to address any additional concerns.
“We initially created this as an assistant to our live image moderation teams,” said Jonathan Freger, founder and CTO of WebPurify. “It was such a successful addition to our moderations platform that we decided to make it publicly available.”
While there are many systems that identify profanity to filter it out, Deep Analysis founder Alan Pelz-Sharpe said doing it at the OCR level is not common.
“There are many other vendors that analyze and extract text from images and memes, though they don’t necessarily utilize OCR,” he said. “Today, newer vendors tend to also leverage machine learning, computer vision and artificial intelligence.”
Pelz-Sharpe added that products like WebPurify or Microsoft Azure Content Moderator are good first steps in stopping profanity before it is published or distributed.
“The bigger problem is in capturing the context of a message and the words used within that context,” he said. “That’s a bigger challenge to resolve.”
Additionally, Pelz-Sharpe said identifying text in images is not an exact science but continues to improve.
“In truth, though, analyzing text in images remains tricky and inconsistent,” he said. “The increased use of deep learning and advanced AI techniques is improving the accuracy rate.”
According to Freger, the OCR Profanity Filter Service will be generally available in early January.