In this paper, we propose a blind watermarking algorithm based on block discrete wavelet transform. The watermark is scrambled by Arnold and two pseudo random sequences are generated first, and then according to the wavelet tree structure, the significant wavelet coefficients in a block are found and quantized by pseudo random sequence. During the extraction, the quantized coefficients can be found without the original image, then the correlation between them and the pseudo random sequences are calculated to estimate the watermark bits. Moreover, in this method all high-frequency subimages are used to provide better robustness. The performance of the proposed watermarking is robust to several attacks, especially quite effective against image cropping attack.