Synthesized Face Obfuscation for Privacy and Aesthetic Quality
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for protecting facial identities in images, such as using Generative Adversarial Networks (GANs) or superimposing stock images, often result in unsatisfactory results, especially when faces are cropped or obscured, and do not maintain aesthetic quality.
Innovation Solution
A system that detects a base face in an input image, selects similar facial images, synthesizes a new facial image by aligning and blending them, and adds it to the input image, even when no face is detected, while reconstructing the background to ensure seamless integration.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If cropping, blurring, or covering faces is used to protect identity, then privacy protection is improved, but image aesthetic quality deteriorates
Solution Approach 1:
The patent creates a synthetic copy of a face that resembles the original person but is not identical, using GANs to generate a realistic alternative face that protects identity while maintaining image quality. This resolves the contradiction by providing a copy that serves the aesthetic function without revealing the true identity.
Solution Approach 2:
The patent introduces a synthetic face as an intermediary between the original face and the public view. This intermediary element protects the original identity while maintaining the visual composition and aesthetic quality of the image, avoiding the need for crude cropping or blurring.
2Reliability
If GANs are used to synthesize new faces, then privacy protection is improved, but image quality deteriorates due to blurred results
Solution Approach 1:
The patent performs preliminary training of specialized GAN models on diverse facial datasets before deployment. This preliminary action ensures that when the GAN generates synthetic faces, they produce high-quality, sharp results rather than blurred images, resolving the quality issue while maintaining privacy protection.
Solution Approach 2:
The patent modifies key parameters and architecture of the GAN model, including using identity-conditioned generation and adjusting discriminator criteria to prioritize sharpness and realism. These parameter changes enable the GAN to generate high-fidelity synthetic faces that maintain both privacy and visual quality.
3Reliability
If stock images are superimposed to protect identity, then privacy protection is improved, but image realism deteriorates
Solution Approach 1:
Instead of using unrelated stock images, the patent creates a custom synthetic copy of the specific individual's face using GANs. This copy maintains the person's distinctive features and likeness while being artificially generated, thus preserving image realism and coherence unlike generic stock photos.
Solution Approach 2:
The patent applies local quality customization by conditioning the GAN generation on the original person's facial characteristics, ensuring the synthetic face locally matches the individual's unique features while protecting identity. This creates a realistic result tailored to the specific person rather than a generic stock image.
4Reliability
If existing facial protection tools are used, then identity protection is improved, but functionality deteriorates when faces are completely cropped out
Solution Approach 1:
The patent performs preliminary detection of whether a face is present in the uploaded image. When no face is detected (completely cropped out), the system switches to an alternative workflow that selects a generic face template instead of attempting to process a non-existent face, thus maintaining functionality in all scenarios.
Solution Approach 2:
The patent implements dynamic adaptability by changing the processing workflow based on face detection results. When a face is present, it generates a synthetic version; when no face is present, it uses a generic template. This dynamic adjustment ensures the tool remains versatile and functional across different user scenarios.
Data Source
AI summary
Methods, apparatus, and systems are provided for obfuscating facial identity in images by synthesizing a new facial image for an input image. A base face is detected from or selected for an input image. Facial images that are similar to the base face are selected and combined to create a new facial image. The new facial image is added to the input image such that the input image includes a combination of the base face and the new facial image. Where no base face is detected in the input image, a base face is selected from reference facial images based at least on pose keypoints identified in the input image. After a new facial image is generated based on the selected base face, a combination of the new facial image and the base facial image are added to the input image by aligning one or more pose keypoints.