Forbes: Face Obfuscation Rendering via Backpropagation Refinement Scheme
Abstract
We propose a new face obfuscation algorithm. This study trains a network model to restore identity information after removing it through various image transformations. The trained model creates an image that is unrecognizable to humans but identifiable only to the model. It shows excellent performance in various face obfuscation benchmark tests, and the effective de-identification performance is confirmed through the results of real random user studies.
- ConferenceECCV
- TopicFace obfuscation
- Year Published2024
Abstract
We propose a new face obfuscation algorithm. This study trains a network model to restore identity information after removing it through various image transformations. The trained model creates an image that is unrecognizable to humans but identifiable only to the model. It shows excellent performance in various face obfuscation benchmark tests, and the effective de-identification performance is confirmed through the results of real random user studies.