Quasar > Image Processing > Restoration

Restoration


An overview of the different topics:
Noise estimation
Wavelets

bregman_utils.q

Contains some general routines that are often used in Bregman/Primal Dual iterative algorithms. These algorithms solve non-linear inverse problems that are encountered within the field of image reconstruction and restoration.

See bregman_utils.q


An overview of the different functions and classes:

bregman_utils.qContains some general routines that are often used in Bregman/Primal Dual iterative algorithms. These algorithms solve non-linear inverse problems that are encountered within the field of image reconstruction and restoration.
cgn_randGenerate a colored gaussian random noise, with the given power spectral density.
choose_sparsity_transformSelection of a sparsity transform
combine_framesCombination of sparsity frames
composite_degradationComposition of degradation operations: A1 o A2 (A1 after A2)
correlated_gaussian_noise_degradationModeling of correlated Gaussian noise with specified power spectral density
covariance_4x4Computes an 4x4 covariance matrix (of a 4-component image, across color channels)
covariance_8x8Computes an 8x8 covariance matrix for an image with 8 components
denoise_nlmeansDenoising of images using non-local means (assuming additive white Gaussian input noise)
displacement_diffComputes a displacement difference. Useful for implementing discrete derivatives, such as used in Total Variation (TV) approaches. The function uses circular boundary extension.
gaussian_blur_degradationModeling of a Gaussian blur
generate_psdGenerates the power spectral density matrix, corresponding to the given point-wise function)
generic_blur_degradationGeneric blur degradation model
kernelshiftHiep's kernelshift function: pads a given filter kernel (mask) with zeros so that it has the dimensions rows x cols. This function is useful for implementing a convolution in the frequency domain
L2_CGSolves the Primal Dual L2-problem using the conjugate gradient method In particular, the problem: (lambda * A^H C_inv A + I) x = (lambda * C_inv A^H y + S^H (d + b)
L2_FFTSolves the Primal Dual L2-problem using the FFT method Note - A_fft and A_fft_H are here specified in FFT domain
L2_generic_solverObtain a generic solver for the L2 problem
L2_pointwiseSolves the Primal Dual L2-problem using the pointwise method using the point-wise method
Noise estimation
primaldual_generic_restorationGeneric image restoration using the primal dual algorithm
shrinkSoft-shrinkage of a subband (or cell array)
soft_stepA soft step function at x=1 (based on the cosine function)
Wavelets
white_gaussian_noise_degradationModeling of white Gaussian noise