EURASIP Journal on Applied Signal Processing
Volume 2002 (2002), Issue 4, Pages 410-417
doi:10.1155/S1110865702000719
Abstract
A watermarking scheme is presented in which the
characteristics of both spatial and frequency techniques are
combined to achieve robustness against image processing and
geometric transformations. The proposed approach consists of
three basic steps: estimation of the just noticeable image
distortion, watermark embedding by adaptive spreading of the
watermark signal in the frequency domain, and extraction of
relevant information relating to the spatial distribution of
pixels in the original image. The just noticeable image
distortion is used to insert a pseudo-random signal such that its
amplitude is maintained below the distortion sensitivity of the
pixel into which it is embedded. Embedding the watermark in the
frequency domain guarantees robustness against compression and
other common image processing transformations. In the spatial
domain most salient image points are characterized using the set
of Hilbert first-order differential invariants. This information
is used to detect geometrical attacks in a frequency-domain
watermarked image and to resynchronize the attacked image. The
presented schema has been evaluated experimentally. The obtained
results show that the technique is resilient to most common
attacks including rotation, translation, and scaling.