Optimal subband wavelet thresholding using noisy and non-noisy data of images

Images in vision systems formed by microwave illumination are considered to be impaired by multiplicative noise. Nevertheless, there are both points corrupted by the noise and points with approximately correct values of intensity. Encoding of such images may apply a wavelet basis using thresholding of the wavelet coefficients. The optimal threshold was obtained for each subband of a multilevel wavelet transformation. It is iteratively reached by approaching the minimum distortion variance estimator, which holds the balance between errors added by distortion of non-noisy data and the residual noise of the de-noised data. Noisy and non-noisy data are found by applying the coefficient variance estimator.