Person de-Identification: A Comprehensive Review of Methods, Datasets, Applications, and Ethical Aspects Along-With New Dimensions

Abstract

Person de-identification has become a challenging problem that is receiving substantial attention because of the growing demand for privacy protection and related regulations. In this context, computer vision and Deep Learning (DL) algorithms offer automated solutions for Face de-identification (FDeID), commonly used to conceal personal identities in visual data. The existing survey studies addressing the FDeID topic lack comprehensive coverage of modern generative DL-based FDeID methods, limitations of data resources, proposing new applications, and potential technical and ethical research directions, which are covered for the first time in this survey. Throughout the manuscript, we offer critical analysis from various perspectives with a recurring theme of the growing impact that generative deep learning techniques are beginning to have on FDeID and related areas such as gait de-identification. In addition, we suggest 17 novel research dimensions and corresponding research questions in both technical and dataset perspectives, which will advance the research frontiers in this domain. The insights presented in this survey can benefit the research community and diverse stakeholders such as law enforcement, healthcare, industry, etc. It offers valuable insights into the performance analysis of existing methodologies, identifies research gaps, highlights application domains, and suggests precise possible avenues for future contributions.