Low-complexity privacy preserving scheme based on compressed sensing and non-negative matrix factorization for image data

Various sensors in Internet of things capture many images, and there is growing concern about their secure storing and sharing. Compressed sensing (CS) is a promising solution for this problem. However, traditional CS-based security frameworks only provide computational secrecy with high reconstruction complexities. In this paper, we propose a low-complexity privacy preserving scheme based on CS and non-negative matrix factorization (NMF) to protect image privacy while maintaining the utility of data. Specifically, CS is used to compress and encrypt the data, and then noise is added to improve security. At the same time, the basis matrix generated by NMF is used to construct a decoding matrix and a decryption matrix. For legitimate users with the decoding matrix, the low-dimensional data of the original signal can be obtained through simple matrix multiplication without complex reconstruction, which can be used for subsequent mining processing. For legitimate users with the decryption matrix, the approximation of the original signal can be obtained by only one-time matrix multiplication. Compared with other traditional schemes, the proposed one avoids the complex reconstruction and ensures the utility of data without revealing the privacy. Experiments and analyses verify the merits of our design in both privacy protection and computational complexity.