Image Denoising in the Transformed Domain Using Non Local Neighborhoods

In this paper we address a denoising technique based on calculation of non local means through neighborhoods. Non local neighborhoods are computed in a transformed domain, namely the wavelet domain. A noisy image is transformed using a lifting scheme. The wavelet coefficients in each subband image are modelized by a generalized Gaussian distribution (GGD) whose parameters (scale and shape parameters) are estimated using an appropriate technique. The estimated parameters are used to define a generalized non local mean which allows us to restore the original image. Processing in the wavelet domain is suitable since image are often available in a compressed domain, beside, processing smaller images allows us to reduce the computational cost