If you need a fool-proof process flatten everything to a 'monochrome' (black and white) image and draw over the redacted areas with a total fill.
This looses text search, but any process that has to understand which objects are being redacted is vastly more complex and thus ripe for errors. Particularly if complex formats like word processor, spreadsheet, or PDF files are involved.
Traditional redaction implies a retention of length unless an entire block is considered protected / classified / secret in some way.
To destroy even length, or possibly stenographic content a transcription and maybe rephrasing might be required. However at that point it's better to go for a summary that avoids the content and know for sure that there's loss of meaning.
If you need a fool-proof process flatten everything to a 'monochrome' (black and white) image and draw over the redacted areas with a total fill.
This looses text search, but any process that has to understand which objects are being redacted is vastly more complex and thus ripe for errors. Particularly if complex formats like word processor, spreadsheet, or PDF files are involved.