Invisible Image Watermarking using Frequency Domain Feature Points

Abstract

For robust invisible watermarking scheme for images, it is crucial to withstand both malicious and common image manipulations (e.g. JPEG compression, pixel jittering or cropping). In this paper, local image features are used. For this, a Maximally Stable Extremal Regions (MSER) detector, which is fast and invariant to affine transformations, is used. The watermark itself is embedded multiple times in circular areas across image. This is achieved by manipulation of magnitude coefficients in frequency domain. During detection, watermark is recovered using majority criterion.