Bivariate shrinkage was a denoising method based on the interscale dependency of wavelet coefficients which using bivariate model as the distribution of the wavelet coefficient and its parent. Although the joint coefficient-parent distributions are different for coefficients in different scales and sub bands, bivariate shrinkage uses the same model for all the coefficients. In order to improve the performance of the bivariate shrinkage method, variable parameter bivariate model was proposed for the joint coefficient-parent distribution of wavelet coefficients in this paper. Based on the new model, a sub band adaptive denoising method was proposed using Bayesian maximum a posteriori estimation theory. In the experiments, the dual tree complex wavelet transform which is shift-invariant and directional selectivity was used for both the new method and bivariate shrinkage method. The results show that the PSNR values of the new method were improved.