Abstract
We propose a classification-based image watermarking algorithm that takes the features of the human visual system into account. An image is divided into blocks, and each block is watermarked in the discrete wavelet transform (DWT) domain, with a strength that depends on a classification of the block, which is obtained by spatial masking in both luminance and texture. We embed watermarks in LL band of DWT coefficients so that they cannot be easily damaged. To embed strong watermarks in images under the constraint of being invisible, classification based on spatial masking is applied to the image blocks before watermark embedding. The watermarks are embedded adaptively according to the classification. These watermarks are invisible, and we show experimentally that they are robust against several image processing operations.