Abstract: With the development of multimedia application, digital images are easy to be manipulated
or tampered illegally. A novel image fingerprint algorithm is proposed which can be used for image
authentication and identification. The image is firstly normalized to make it robust against any affine
transforms. Then the normalized image is divided into 8×8 blocks and discrete cosine transform is
applied to each block. The DCT coefficients are weighted by the contrast masking matrix of Watson
visual model. Then weighted DCT coefficients is quantified to be bit sequences to form the image
fingerprint. Since a key is used in the algorithm, the fingerprint is hard to be forged. The experimental
results indicate that the proposed method is robust against modification including geometric
transformation such as rotation and scaling.

Key words: image fingerprint, image normalization, visual model, image authentication