In order to take into account the transparency and antigeometric transformation ability of the watermarking system, a robust image watermarking algorithm based on visual saliency and quantization index modulation is proposed in this paper. Firstly, the Ripplet transform is used to process the host image for getting the feature map. Then, the Gaussian probability density model is used to calculate the visual saliency mapping corresponding to the feature map, and divide it into a series of nonoverlapping subblocks for calculating the saliency mean of each subblock. The nondown sampling Contourlet transform is introduced to decompose the host image for outputing the corresponding lowpass subbands and bandpass directional subbands. Subsequently, the low pass subband is divided into smaller nonoverlapping subbands and the energy of each subband is calculated. The saliency mean and energy are jointed to calculate the quantization step corresponding to the embedded subblock, which treat it as a key. The singular value decomposition is used to process each subblock of low pass subband for obtaining the corresponding diagonal matrix, and the maximum singular value is found out. A watermarking embedding method is designed based on the mean of the maximum singular value corresponding to all subblocks, and according to the quantization step corresponding to each subblock, the watermarking data is hidden into the carrier to get the watermarking image. Finally, according to the received key, the watermarking extraction mechanism is defined to detect the watermarking data in the watermarking image. The experimental data show that this algorithm has higher transparency than the current blockbased watermarking technology, and under the conventional geometric content operation, it shows stronger robustness, and the restoration of watermarking distortion is the smallest.