Ask a Question

What Is PII Masking (Personally Identifiable Information)?

Table of Contents

PII Masking - Definition

PII masking in the context of images and video recordings is a controlled, technical process of permanently obscuring elements within a frame that enable the identification of an individual. In the European Union, the concept of PII largely corresponds in practice to the scope of personal data as defined in Article 4(1) of Regulation (EU) 2016/679 (GDPR), meaning any information relating to an identified or identifiable natural person. In visual materials, the most common identifiers include faces and vehicle license plates. PII masking is applied to reduce the risk of re-identification and to comply with the principles of data minimization and necessity of processing.

Effective anonymization, according to guidance from supervisory authorities (WP29, EDPB, CNIL), requires that the process be practically irreversible using reasonably available means. This involves selecting masking methods and parameters that prevent the identification of an individual, either directly or through data linkage, while preserving the usability of the material for its declared purpose.

The Role of PII Masking in Image and Video Anonymization

PII masking is a fundamental risk-reduction measure in video surveillance, media publishing, UX research archiving, and the sharing of AI training data. The EDPB emphasizes that processing data from video devices must comply with the principles of proportionality and purpose limitation, and that disclosing images of individuals or vehicle identifiers requires a valid legal basis and relevance to the purpose of publication.

  • Faces: A person’s image constitutes personal data. If it is used for unique identification based on biometric characteristics (e.g., facial recognition systems), the processed data may qualify as biometric data, a special category of personal data under Article 9 GDPR.
  • License plates: In many EU countries, supervisory authorities consider license plates to be personal data when they allow identification of a vehicle owner or user through accessible registers. In Poland, interpretations vary depending on the processing context; in practice, masking is recommended when content is made public.
  • Exceptions to consent requirements: Under copyright law (Article 81), consent is not required in specific cases, such as images of well-known public figures, individuals shown as part of a larger whole (e.g., a public gathering), or where consent was granted in exchange for payment.

PII Masking Technologies for Images and Video

Automated, effective PII masking in images and video relies on object detection and localization within frames, combined with spatio-temporal processing. Deep learning is commonly used to build face and license plate detection models, including convolutional neural networks, one-stage and two-stage detection architectures, as well as newer transformer-based models. Once deployed, the AI model is used operationally to identify regions that require masking.

  • Detection and tracking: Per-frame detection followed by temporal object tracking to ensure mask consistency (e.g., SORT, DeepSORT, or equivalent algorithms).
  • Region expansion: Adding margins around detected objects to compensate for localization errors and camera movement.
  • Masking methods: Gaussian blur, pixelation (mosaic), solid fills, or textured fills. Parameters should be selected to minimize the risk of reversal and re-identification.
  • On-premise processing: Reduces data transfer to third parties and simplifies control over the processing chain.

Gallio PRO-class software automatically masks only faces and vehicle license plates. It does not perform real-time anonymization or live video stream anonymization. Masking of logos, tattoos, or documents within recordings is possible manually using the editor. The system does not collect logs containing face or license plate detections, nor any other personal data.

Key Parameters and Quality Metrics for PII Masking

The assessment of PII masking quality combines detection accuracy, localization precision, computational performance metrics, and privacy robustness testing. The table below summarizes the most commonly used measures in image and video anonymization projects.

Metric

Definition

Relevance to Privacy

 

Precision

TP / (TP + FP)

A lower false-positive rate reduces unnecessary masking.

Recall

TP / (TP + FN)

Critical for minimizing the risk of leaving faces or plates unmasked.

F1 Score

2 × Precision × Recall / (Precision + Recall)

Balances detection effectiveness and accuracy.

IoU

Intersection / Union of detected and reference areas

Ensures the mask fully covers the identifying region.

mAP

Mean Average Precision across IoU thresholds

Indicative detector quality at varying matching levels.

Latency

ms/frame or ms/image

Resource planning; important for batch processing.

Throughput

frames/s or images/s

Production scalability and SLA compliance.

Reversibility testing

Assessment of susceptibility to deblurring and super-resolution

Validates the irreversibility required for anonymization.

Manual audits

Review of high-risk samples

Identifies edge cases missed by automated metrics.

Challenges and Limitations

PII masking in images and video involves both technical and legal challenges. Detection quality decreases under poor lighting, strong motion blur, occlusions, unfavorable angles, or low resolution. Inaccurate localization can lead to under-masking, while overly aggressive filters reduce the usability of the material. Pixelation and light blurring may be partially reversible using super-resolution algorithms, which is why parameters should be selected conservatively. Automated solutions typically cover only faces and license plates; other forms of PII require manual editing and quality control procedures. Additionally, differences in legal interpretation (e.g., license plates in Poland) necessitate compliance policies tailored to the jurisdiction and processing purpose.

Normative References and Sources

Below is a selection of key legal acts and standards shaping requirements and best practices for PII masking in visual materials. These documents define personal data concepts, criteria for effective anonymization, and privacy management frameworks.

See also

Back to Glossary