Secure Dictionary Learning for Sparse Representation

In this paper, we propose secure dictionary learning for sparse representation based on a random unitary transform. Edge cloud computing is now spreading to many application fields including services that use sparse coding. This situation raises many new privacy concerns. The proposed scheme provides practical MOD and K-SVD schemes that allow computation on encrypted signals. We prove, theoretically, that the proposal has exactly the same dictionary and sparse coefficient estimation performance as sparse dictionary learning for unencrypted signals. It can be directly carried out by using MOD and K-SVD algorithms. Moreover, we apply it to image modeling based on an image patch model. Finally, we demonstrate its excellent performance on synthetic data and natural images.