The disadvantages of wavelet transform and contourlet transform are studied. Color image digital watermarking is implemented based on the multi-scale, multi-direction, and shift-invariance of the nonsubsampled transform. The coefficients in different scales and directions are obtained by image decomposition using the nonsubsampled contourlet transform. Embedding intensities of the low and high frequency coefficients are different invisibility and robustness of watermark requirements for according to the color perception characteristics of human visual system, watermark embedded in blue channel is more than the others, which only choose the local of maximal energy to embed. Compared with the algorithms of stationary wavelet transform and contourlet transform under the same embedding rule, this algorithm is more robust, with higher similarity which can be improved by 0.5 at most.