| 摘要 |
The method effects a deblurring of the image by performing a deconvolution of the digital signal and by carrying out a denoising of the image with respect to the compression noise and the instrumental noise by expressing the digital signal on a basis of wavelet packets on which the two types of noise are Gaussian. The determination of the basis on which the compression noise and the instrumental noise are Gaussian is effected in the course of learning the characteristics of the instrument chain associated with the image compression which includes the steps of constructing a table of coefficients as a function of the deblurring deconvolution effect on an average noise signal expressed in a generalized Walsh basis; constructing a decomposition basis which is the least decomposed by comparing the values of the coefficients to a fineness threshold; constructing a wavelets basis allowing the bets denoising with respect to the average noise decomposed in the generalized Walsh basis; and constructing the best basis by fusing the two basis and preserving the finest decomposition coefficients of each basis for the construction of the best basis. The decorruption stage comprises the steps of detecting the modules of the coefficients of the image composition in the best basis with respect to a threshold, and of processing the coefficietns. The threshold value is a function of the noise type difference at all points of the image, a thresholding parameter, the deconvolution filter value, the grey level of the image and the local value of the quantization interval. Small coefficients are multiplied by a noise factor, and large coefficients are deconvoluted. At least one step of supersampling of packets is effected before the denoising and/or the deconvolution in order to avoid the phenomenon of the folding distortion, that is aliasing.
|
