Based on the improved Watson's model, an adaptive spread-transform dither modulation (ASTDM) algorithm is proposed for efficient color image watermarking. The luminance masking threshold is modified to be consistent with amplitude without affecting the original parameters. Then, by incorporating the improved model with STDM, we achieve a maximum labeling strength and an appropriate quantization step-size shrinking with the local values of a host image. Finally, the Normalized SI Correction method is implemented to correct the possible color distortion. Experiments demonstrate that with high embedding rate and transparency, ASTDM can greatly increase the robustness against amplitude scaling while retaining anti-re-quantization property, which means the two main drawbacks of QIM algorithm are solved simultaneously.