This paper proposes a blind watermarking method which utilizes the Euclidean distance between adjacent coefficients in wavelet trees to improve robustness, and lower distortion. To decrease the probability of being detected or deleted, different combinations of significant or insignificant coefficients are proposed. We embed the watermark in either two significant or two insignificant coefficients. The value of detector is adaptable when a watermarked image is under different attacks. This can facilitate us to extract watermark more accurately. The experimental results show that the superiority of the proposed schemes and the advantages are demonstrated through comparison with conventional wavelet tree based watermarking.