A wavelet based intra- and inter-scale hybrid denoising scheme is presented in this paper. With overcomplete wavelet expansion, the wavelet coefficients with the same spatial orientation at several scales are combined and represented as a vector. The linear minimum mean squared-error estimation (LMMSE) is then imposed on each vector to incorporate the interscale dependency information cross scales to update the estimation. To exploit the wavelet interscale dependency, the covariance matrix of each vector is estimated locally by a centered square-shaped window. Experiments show that the proposed scheme outperforms the interscale-based or interscale-based methods. It is showed that the performance also depends on wavelet filters. In our experiments a biorthogonal wavelet, which well characterizes the interscale dependency, achieves the best results.
展开▼
