(Can’t) Picture This 2: An Analysis of WeChat’s Realtime Image Filtering in Chats By Jeffrey Knockel and Ruohan Xiong We found...
(Can’t) Picture This 2:
An Analysis of WeChat’s Realtime Image Filtering in Chats
We found that Tencent implements realtime, automatic censorship of chat images on WeChat based on text contained in images and on an image’s visual similarity to those on a blacklist. Tencent facilitates realtime filtering by maintaining a hash index populated by MD5 hashes of images sent by users of the chat platform. If the MD5 hash of an image sent over the chat platform is not in the hash index, then the image is not filtered. Instead, it is queued for automatic analysis. If it is found to be sensitive, then its MD5 hash is added to the hash index, and it will be filtered the next time a user attempts to send an image with the same hash.
This finding indicates that censorship measurement—like the kind conducted in this report—not only evaluates censorship but can also influence and modify the behaviour of a realtime, automatic censorship system by introducing novel items that can be flagged as sensitive and subsequently censored. This helps us understand previous measurements and has implications for future censorship measurement research.