Subband noise estimation for adaptive wavelet shrinkage

In this article, we present an adaptive image denoising method based on subband noise modeling. For wavelet shrinkage, choosing the threshold depends on correctly estimating the noise variance. By modeling the inter-subband noise variance with a parameterized normalized exponential function, the problem becomes identifying the maximum noise variance. Such a maximum exists in the highest decomposition level and can be estimated by locating the extreme of the first derivative of the subband variance function. The experiments demonstrate that our method outperforms peers, especially in the cases of large noise variance.