Externally indexed torrent

If you are the original uploader, contact staff to have it moved to your account

Textbook in PDF format

This book provides state-of-the-art Face De-Identification techniques and privacy protection methods, while highlighting the challenges faced in safeguarding personal information. It presents three innovative image privacy protection approaches, including differential private k-anonymity, identity differential privacy guarantee and personalized and invertible Face De-Identification. In addition, the authors propose a novel architecture for reversible Face Video De-Identification, which utilizes deep motion flow to ensure seamless privacy protection across video frames.
This book is a compelling exploration of the rapidly evolving field of Face De-Identification and priacy protection in the age of advanced AI-based face recognition technology and pervasive surveillance. This insightful book embarks readers on a journey through the intricate landscape of facial recognition, Artificial Intelligence (AI), social network and the challenges posed by the digital footprint left behind by individuals in their daily lives. The authors also explore emerging trends in privacy protection and discuss future research directions.
Face recognition is a biometric technology that automatically recognizes people’s facial features including statistics and geometric features, which is one of the most important applications of image analysis and understanding. Face recognition tasks can be further divided into binary classification and multiclassification. The binary classification task is also called face verification, which is used to compare whether two images have the same identity. The multiclassification task is also called face retrieval, such as searching for a face with a specific identity in a database of many faces. The widely known face recognition is the abbreviation for identity recognition and verification based on optical facial images. The face recognition process can be simply summarized as using a computer to analyze a face video or image. Firstly, it detects and possibly tracks (just for videos) the faces, so as to localize them. Secondly, it aligns the faces to normalized canonical coordinates. Thirdly, it extracts effective facial features. Finally, it determines the identity of the face object through a comparison of the above-mentioned features.
The research on face recognition can be traced back to the late 1960s. The main idea is to design feature extractors and then use Machine Learning algorithms for classification. Traditional methods rely on hand-made features, such as edge texture description, and combine with Machine Learning techniques such as principal component analysis, linear discriminant analysis, and support vector machines.
After introducing Deep Learning techniques into the field, the approaches have been transferred to extract features with neural networks, which has greatly improved the accuracy and robustness. The Deep Learning models can be trained by a large amount of data to learn the representation of various variability such as lighting conditions, postures, facial expressions, and so on.
Today, face recognition technology has been widely used in our daily life. Face verification can be treated as a new way of identity confirmation for fast face comparison, mobile payment authentication, security identity verification, etc. Face retrieval can be applied to investigate suspects, complete search of missing persons’ databases, and repeated investigation of multiple certificates for one person. At present, the face recognition model can achieve satisfactory accuracy on a specific dataset, but the influence of illumination and posture is still the main challenge. In addition, cross-racial and cross-age recognition problems are also worth studying.
Due to potential privacy issues, the application of face recognition technology is currently under controversy, and the face privacy protection task is receiving more and more attention. Face de-identification, the main content of this book, is an innovative technical idea to solve the dilemma. There is no consistent definition of de-identification in the existing literature. Ribaric et al. defined de-identification in multimedia content as “the process of concealing or removing personal identifiers, or replacing them with surrogate personal identifiers in multimedia content.” During this process, other facial features that are not related to identity should remain unchanged, such as expression, posture, and background. After this process, the de-identified face will be judged by the face recognition technology as no longer the same identity as the original face. At the same time, the identity-protected face is expected to retain as much similarity to the original image as possible for normal viewing and sharing and can still be analyzed and processed by other identity-agnostic computer vision methods, such as face detection, motion monitoring, and emotion recognition.
Researchers working in Computer Science, Artificial Intelligence, Machine Learning, data privacy and cybersecurity as well as advanced-level students majoring in Computers Science will find this book useful as reference or secondary text. Professionals working in the fields of biometrics, data security, software development and facial recognition technology as well as policymakers and government officials will also want to purchase this book