Amazon Nova automates PII redaction in images for compliance

Overview of Amazon Nova

Amazon Nova is a family of foundation models designed to handle complex image analysis workflows, particularly focusing on the identification and redaction of Personally Identifiable Information (PII). Capturing PII in images presents unique challenges due to the varied forms it can take, such as reflections, partial views, or embedded text.

Challenges in PII Redaction

Organizations face serious legal obligations to redact PII under regulations like GDPR and PCI DSS. Failure to do so can lead to significant fines and reputational damage. Traditional masking tools often struggle with real-world scenarios where PII appears unexpectedly within images.

How Nova Works

Amazon Nova employs advanced vision understanding capabilities to assess image content in context. It coordinates multiple tools to achieve pixel-level precision in redacting PII while maintaining the integrity of the image for its intended use.

Integration with Other Tools

The PII redaction solution leverages Meta's Segment Anything Model (SAM 3) for pixel-level segmentation and Amazon Textract for optical character recognition (OCR). This combination allows for meticulous redaction of challenging PII scenarios such as fingerprints and identification cards.