Abstract: A new approach to adaptively select the local regularization parameters for regularized image restoration is proposed.Firstly, the local regularization parameter should be proportional to local noise variance in degradedi mage is proposed.Then, an algorithmto estimate noise variance in wavelet domainis given, and the local regularization parameters are selected adaptively according to the distribution of estimated noise variance in wavelet domain.The experimental results show that for many different types of noise the estimated noise variance is consistent with the real one in distribution and the proposed approach can achieve restoration than the Katsaggelos’approach does.