Facial Image Anonymization via Latent Vector Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for facial image anonymization on edge devices are limited by their computational and memory resources, making it impractical to execute high-quality neural networks, and uploading images to remote servers poses data security risks.
Innovation Solution
Implementing a lightweight encoder and anonymization transformation on edge devices to encode facial images into low-dimensional latent vectors, which are then transmitted for decoding on a remote device, preserving key attributes like gender and ethnicity while modifying identity features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If high-quality neural networks are executed on edge devices for facial image anonymization, then anonymization quality is improved, but computational resource requirements and memory usage increase beyond edge device capabilities
Solution Approach 1:
The system segments the anonymization process into two distinct phases: encoding performed on the edge device and decoding performed on a remote server. This division allows the computationally intensive decoding operation to be offloaded, enabling high-quality anonymization without overwhelming edge device resources.
Solution Approach 2:
The patent introduces a remote server as an intermediary to handle the decoding of latent vectors. This intermediary absorbs the computational burden of high-quality decoding, allowing edge devices to maintain simple, resource-efficient encoding operations while still achieving high anonymization quality through the server's decoding capability.
2Power
If facial images are uploaded to remote servers for anonymization processing, then processing capability is improved, but data security risks increase
Solution Approach 1:
The system extracts and transmits only the essential latent vector representation to the remote server, leaving the actual facial image data on the edge device. This extraction approach enables the server to perform anonymization processing without having access to or storing the original sensitive facial images, thereby maintaining data security while providing remote processing capability.
Solution Approach 2:
The edge device performs preliminary encoding of the facial image into a latent vector before transmission. This preliminary action transforms the sensitive image data into a compressed, non-reversible representation that can be processed remotely without exposing the original facial information, thus enabling secure remote processing.
3Productivity
If lightweight encoder and anonymization transformation are implemented on edge devices, then resource efficiency is improved, but anonymization quality may be reduced
Solution Approach 1:
The anonymization process is segmented into lightweight encoding on the edge device and quality-enhancing decoding on the remote server. This segmentation allows the edge device to operate efficiently with resource-friendly operations while the server handles the computationally intensive decoding that ensures high anonymization quality.
Solution Approach 2:
A remote server acts as an intermediary to perform high-quality decoding of the latent vectors generated by the edge device's lightweight encoder. This intermediary compensates for the limitations of edge device processing, ensuring that the final anonymized output achieves high quality despite the resource efficiency constraints at the edge.
Data Source
AI summary
A system and method of performing facial image anonymization may include: obtaining, by at least one processor of a first computing device, at least one first image data element, depicting a first face; using, by the at least one processor, a first neural network (NN), to encode the facial image into a first low-dimension, latent vector; using, by the at least one processor, a transformation module, adapted to modify the first latent vector, to produce an anonymized latent vector; and transmitting, by the at least one processor, the anonymized latent vector, via a communication network to a remote second computing device, wherein the second computing device is adapted to decode the anonymized latent vector into a second image data element, depicting an anonymized face.