Approximate Image Authentication and Correction using Spatial and Frequency Domain Features

Abstract:
Images traversing from source to sink can undergo intentional and unintentional modifications. An image authentication scheme is presented in this paper which is capable of detecting modifications in the image, locating their positions and correcting some or all of them. It is also able to tolerate minor unintentional modifications. However, intentional forgeries are detected by the proposed scheme. The idea is to build a watermark using local frequency domain image features and protecting them with Reed Solomon codes. This helps in the localization of modifications and their correction to the extent possible with the used code. At the same time, the global image histogram features are protected using an approximate message authentication code. This provides tolerance in authentication in the presence of unintentional modifications such as channel noise, image transformation operations, compression etc. Security strength of the proposed image authentication scheme is analyzed in the paper. Simulation results in the presence of unintentional modifications show the noise tolerant authentication capability, whereas simulation results in the presence of intentional modifications demonstrate the forgery detection capability of the proposed scheme.