A novel effective adaptive digital watermarking algorithm was proposed without using the traditional method of combining spatial and DWT domain based on the features of human visual system (HVS). The loaded image was divided into several blocks and calculated their variances. A pseudo random sequence was produced after comparing these variances with the variance of the whole loaded image. Then a spread spectrum watermark sequence was created after combining original watermark and pseudo random sequence

and the robustness was accordingly enhanced by utilizing the spread spectrum codes. The loaded image was decomposed by DWT (Discrete Wavelet Transform) into 2-level sub-band images. The watermark sequence was embedded into the high-frequency coefficients of the loaded image. An innovative quantization method was proposed. To improve the security of the watermark system

the cryptography technology was used in the embedding procedure. A long periodic pseudo random m-sequence was created to choose the location of watermark to be embedded in. Each bit watermark that used to embed in load image needs to mark choice every time. The traditional quantization methods usually use the constant quantized interval and the scheme proposed in this paper is called odd-even adjudegment rule. The scheme could adaptively adjust the quantized interval by calculating the adjacert symbols' mean value of the wavelet coefficients which is selected to embed the watermark. The top-left

top and left coefficients were selected as the adjacent symbols of embedded wavelet coefficient's neighboring symbols. The fidelity of the watermarked image and the invisibility of the watermark have been enhanced by this method. The main idea of odd-even adjudgement rule is to quantize the coefficients into the most adjacent odd quantization mark or even quantization mark according to the watermark's value. The blind extraction could be realized by using this rule. Experimental results show that the watermark proposed in this paper is invisible (the PSNR is higher than 41) and is robust in the case of against some normal attacks such as JPEG compression

additive noise and filtering

etc. (the similarity between the original watermark and extracted watermark is almost above 0.85).