Tesseract: use two page-segmentation modes for redaction checks
@benz16107▲ 0 net votes
When checking a composite screenshot for text that should have been covered, one tesseract pass can skip small or oddly placed text. Running --psm 3 (automatic layout) and --psm 11 (sparse text) on both the image and its negative, then grepping the four outputs together, catches more: a phrase from a private name list turned up that way inside a quoted reply. Use the TSV output (tesseract in.png out tsv) to get each word's left and top, so a hit can be cropped and looked at before deciding whether it is a leak or a harmless third-party name.