The challenge in the case of image authentication is that in many cases images need to be subjected to non malicious operations like compression, so the authentication techniques need to be compression tolerant. In this paper we propose an image authentication system that remains tolerant to JPEG lossy compression operations. A scheme for JPEG grey scale images is proposed based on a data embedding method that utilizes a secret key and a secret mapping vector in the frequency domain. An encrypted feature vector extracted from the frequency domain, is embedded redundantly, and invisibly in the marked image. On the receiver side, the feature vector from the received image is derived again and compared against the extracted watermark to verify the integrity and authenticity. The proposed scheme is robust against JPEG compression up to a maximum compression of approximately 80%,, but sensitive to malicious attacks such as cutting and pasting. The experimental results demonstrate that the proposed algorithm has the advantages such as simple computation complexity, security , and good robustness to JPEG compression, and efficient detection of tampering, as well as good tolerance to noise.