Two types of local adaptive filters are known:
* Filters based on a nonstationary mean, nonstationary variance (NMNV) image model.
* Filters implemented on a short space basis when an image is divided into
nonoverlapping subimages and each subimage is restored separately, the restoration being
performed in the domain of DFT.
In the suggested local adaptive filters:
* Signal local spectral density in DCT domain, rather than only signal local mean and
variance, is used as an adaptation parameter; the use of DCT improves spectral estimation
needed for the local filter design.
* The filter window is continuously sliding over the image with the step of one pixel and,
in each position of the window, only its central pixel is estimated.
Fig. 1. Flow diagram of local adaptive filtering in DCT domain