Vision AI-Based Automatic Selective De-Identification

Abstract

Recently, with personal information being increasingly used in several applications, its leakage and resulting damage have become a relevant concern. In video data, only people consenting to the collection of personal data are to be identified, and other people should be de-identified. In current automatic de-identification, not all objects are identified, and de-identification of only recognized objects is performed manually using video editor tools. Therefore, in this study, face/object-detection-based automatic selective de-identification was conducted through artificial intelligence technology to identify approved objects and de-identify remaining objects. Thus, selective protection can be automatically performed while ensuring the usability and stability of personal information.