Face Anonymization via Latent Vector Blending
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge lies in anonymizing faces in digital images, particularly when capturing recognizable individuals in stock images, as obtaining releases from all subjects can be difficult, limiting the accessibility and usage of such images in content creation.
Innovation Solution
A digital object editing system employs machine learning to generate a mixed face by combining a target face with a reference face using linear interpolation of latent vectors, replacing the target face with a mixed face that does not correspond to an actual human, thus avoiding licensing issues.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional face anonymization techniques are used, then licensing issues are avoided, but the anonymized faces appear artificial and unrealistic
Solution Approach 1:
The patent uses a third reference face as an intermediary to blend with the target face. This intermediary face serves as a mediator that transforms the target face into a new identity while preserving realistic facial characteristics. The blending process combines features from both the target and reference faces to create an anonymized face that appears natural and realistic, avoiding the artificial appearance of conventional techniques.
2Reliability
If releases are obtained from all subjects in the image, then licensing compliance is achieved, but the process becomes time-consuming and complex
Solution Approach 1:
The patent extracts and removes the identifying characteristics of the target face by blending it with a reference face. This extraction process eliminates the connection to the original subject's identity, effectively removing the need to obtain releases from the target face subject. The system extracts only the necessary facial features needed for realism while discarding the identity-specific information.
3Object-generated harmful factors
If simple face blurring is applied, then licensing issues are resolved, but the image quality and realism are degraded
Solution Approach 1:
The patent creates a composite face by combining features from the target face and a reference face. This composite approach merges multiple facial characteristics to generate a new face that maintains high image quality and realism. The blending process integrates skin texture, facial contours, and feature details from both sources, producing an anonymized face that appears natural rather than blurred or degraded.
Data Source
AI summary
Face anonymization techniques are described that overcome conventional challenges to generate an anonymized face. In one example, a digital object editing system is configured to generate an anonymized face based on a target face and a reference face. As part of this, the digital object editing system employs an encoder as part of machine learning to extract a target encoding of the target face image and a reference encoding of the reference face. The digital object editing system then generates a mixed encoding from the target and reference encodings. The mixed encoding is employed by a machine-learning model of the digital object editing system to generate a mixed face. An object replacement module is used by the digital object editing system to replace the target face in the target digital image with the mixed face.