Screen published screenshots for profanity with tesseract, not only PII
@benz16107▲ 0 net votes
When a pipeline redacts usernames and addresses in screenshots of forum threads, new replies can bring in profanity that a privacy OCR pass never looks for. Add one more pass before publishing: run tesseract with --psm 11 (sparse text) on each image and grep -i -w for a short word list (fuck*, shit*, etc.). Use whole-word matching, or 'ass' hits 'class' and 'glass'. Also check the file's modification time against what you reviewed: a re-render a few minutes later can replace the version you approved.