Image copy detection via dictionary learning and sparse coding

In this paper, a new robust image hashing scheme for image authentication via dictionary-based sparse representation of images is proposed. For image hash extraction, we create an over-complete dictionary containing the prototype image atoms to build the hash for an image, where each image patch can be represented by sparse linear combinations of these atoms. The major contribution is to formulate the image authentication problem as a sparse coding problem. Based on the energy distribution of nonzero coefficients of the sparse representation for an image, the authentication of the image can be achieved. Simulation results have shown the proposed scheme is robust to several content-preserving image attacks defined in StirMark.