Efficient Privacy Preserving Protocols for Visual Computation
Maneesh Upmanyu (homepage)
The rapid expansion of the Internet is receiving a great deal of attention world-wide. The technological developments and the increase in online communications have played a vital role in revolutionalizing the Information Age. There is a tremendous growth of online applications to manipulate the user's personal data, which has resulted in the widespread availability of the user's personal data in the digital form. This raises the issue of privacy protection against potential misuse of this data by legitimate service providers or intruders. Without proper countermeasures to thwart the attacks, security problems become a major threat and a serious impediment to further development of business applications on communication systems.
Many of the current solutions provides information security by assuming a level of trust among the parties. The leakage of the critical data to third parties is prevented by applying cryptographic primitives as a secure layer over the standard algorithm. On the other hand, privacy preservation computation is more closely related to Secure Multiparty Computation (SMC). SMC enables two parties; one with the function f() and the other with the input x; to compute f(x) without revealing them to each other. However, the solutions based on the general protocol of SMC requires enormous computational and communication overhead, thus limiting the practical deployment of the secure algorithms.
In this dissertation, we focus on development of `highly-secure', `comunication and computationally efficient' algorithms to problems with `immediate impact' in the domain of computer vision and related areas. Security issues in computer vision primarily originates from the storage, distribution and processing of the personal data, whereas privacy concerns with tracking down of the user's activity. The primary challenge is in providing the ability to perform generic computations on the visual data, while ensuring provable security. In this thesis, we propose lightweight encryption's for visual data, such that the server should be able to carry out the computations on the encrypted data and also store the stream if required, without being able to decipher the actual contents of the image. Moreover, the protocols are designed such that the interaction and the data communication among the servers is kept to a minimum.
It has been proven before that the best way to achieve secure computation on a remote server is by using the cryptographic protocol of SMC. Thus, a method that provides provable security, while allowing efficient computations without incurring either significant computation or communication overhead has remained elusive till now. We show that, for designing secure visual algorithms one can exploit certain properties such as scalability, limited range etc, inherent to visual data to break this impenetrable barrier. We study and propose secure solutions for applications such as Blind Authentication, i.e. blindly authenticating a remote-user using his biometric. Subsequently, we present a highly secure framework for carrying out visual surveillance on random looking video streams at remote servers. We then propose a simple and an efficient cloud-computing based solution using the paradigm of secret sharing to privately cluster an arbitrary partitioned data among N users. The solutions we propose are accurate, efficient and scalable and has potential to extend over to even more diverse applications.
In our first work, blind authentication, we propose private biometric authentication protocol which is extreamly secure under a variety of attacks and can be used with a wide variety of biometric traits. The protocol is blind in the sense that it reveals only the identity, and no additional information about the user or the biometric to the authenticating server or vice-versa. The primary advantage of the proposed approach is the ability to achieve classification of a strongly encrypted feature vector using generic classifiers such as Neural Networks and SVMs. Our proposed solution addresses the concerns of user's privacy, template protection, and trust issues. And captures the advantages of biometric authentication as well as the security of public key cryptography.
We then present an efficient, practical and highly secure framework for implementing visual surveillance on untrusted remote computers. To achieve this we demonstrate that the properties of visual data can be exploited to break the bottleneck of computational and communication overheads. The issues in practical implementation of certain algorithms including change detection, optical flow, and face detection are addressed. Our method enables distributed secure processing and storage, while retaining the ability to reconstruct the original data in case of a legal requirement. Such an architecture provides us both security as well as computation and communication efficiency.
We next extend our proposed paradigm to achieve the ability to do un-supervised learning using K-means in the encrypted domain. Traditional approaches uses primitives such as SMC or PKC, thus compromising the efficiency of the solutions and in return provide very high level of privacy which is usually an overkill in practice. We use the paradigm of secret sharing , which allows the data to be divided into multiple shares and processed separately at different servers. Our method shows that privacy need not be always at the cost of efficiency. Our proposed solution is not only computationally efficient but also secure independent of whether or not P ≠ NP.
|Year of completion:||June 2010|
|Advisor :||C. V. Jawahar, Anoop M. Namboodiri, Dr. Kannan Srinathan|
Maneesh Upmanyu, Anoop M. Namboodiri, Kannan Srinathan and C. V. Jawahar - Blind Authentication: A Secure Crypto-Biometric Verification Protocol IEEE Transactions on Information Forensics and Security, Vol. 5(2), pp. 255-268 (2010). [PDF]
Maneesh Upmanyu, Anoop M. Namboodiri, K. Srinathan and C. V. Jawahar - Efficient Biometric Verification in Encrypted Domain Proceedings of the 3rd International Conference on Biometrics (ICB 2009), pp. 899-908, June . 2-5, 2009, Alghero, Italy. [PDF]
Maneesh Upmanyu, Anoop M. Namboodiri, K. Srinathan and C.V. Jawahar - Efficient Privacy Preserving Video Surveillance Poceedings of the 12th International Conference on Computer Vision (ICCV), 2009, Kyoto, Japan [PDF]
- Maneesh Upmanyu, Anoop M. Namboodiri, K. Srinathan and C.V. Jawahar - Efficient Privacy Preserving K-Means Clustering: ; Pacific Asia Workshop on Intelligence and Security Informatics (PAISI) 2010 (Best Paper Award) (Download).