Gaussian scale-space features for semi-fragile image authentication

In this paper, we propose a new robust image hashing approach to be used for semi-fragile image authentication. We use a texture-based feature, the so-called gray-level blob in Gaussian scale-space, which is invariant to scaling, rotation, and translation. It is associated with the image content and thus independent of image geometry. Furthermore, we show that our new approach is robust against a variety of signal processing operations, such as lossy compression, contrast and luminance enhancement, blurring, sharpening as well as noise adding.