Abstract: A watermarking method for images in wavelet domain is proposed in this paper, which realizes the watermarking blind extraction by using neural network technology. Watermark sequence is imbedded in the low-sequence of Y component in YUV color space, including a section of template information. The template information defines a kind of relation among the low-frequency coefficients in the block after divided into 3×3 in the low-frequency portion. The neural network is trained to realize the watermarking blind extraction through utilizing template information.