shape-adaptive DCT denoise

Pointwise Shape-Adaptive DCT Demobox - Public release v1.40 (5 october 2006) ------------------------------------------------------------------------------------------------------------ (c) 2005-2006 Alessandro Foi Kostadin Dabov Tampere University of Technology, Tampere, FINLAND http://www.cs.tut.fi/~foi/SA-DCT ------------------------------------------------------------------------------------------------------------ The package comprises five demos demo_SADCT_denoising.m (denoising of grayscale images), demo_SADCT_color_denoising.m (denoising of color images), demo_SADCT_deblocking.m (deblocking and deringing of grayscale and color B-DCT compressed images), demo_SADCT_deblurring.m (deblurring of grayscale images), demo_SADCT_inverse_halftoning.m (reconstruction of grayscale images from binary halftones) and a function which shows the adaptive-shape neighborhoods which are used in the demos as supports of the Shape-Adaptive DCT transform function_shape_explorer.m (visualization of the adaptive-shape neighborhoods). The demos are open-source, and may be modified and tuned to be exploited with other data. Each demo calls a pair of functions in order to perform hard-thresholding and Wiener filtering in SA-DCT domain. In particular, the denoising and deblocking demos use function_SADCT_thresholding_fast.dll (hard-thresholding in SA-DCT domain), function_SADCT_wiener_fast.dll (Wiener-filtering in SA-DCT domain), while the deblurring demo uses the pair function_SADCT_RI_thresholding.dll (hard-thresholding in SA-DCT domain for regularized inverse), function_SADCT_RW_wiener.dll (Wiener-filtering in SA-DCT domain for regularized-Wiener inverse). Two more functions are called by the color denoising demo in order to perform color-space transformations: function_rgb2LumChrom.m (inverse transformation), function_LumChrom2rgb.m (forward transformation). Starting from v1.10, a new function function_AnisLPAICI8.p is introduced to perform the Anisotropic LPA-ICI adaptive-scale selection. It provides four to five times faster processing for the LPA-ICI part and requires much less memory than the previous general code based on conv2 and function_ICI code. Requirements: This demobox is developed for Matlab version 7 and the dll's are compiled for x86 (Pentium3, Athlon, or better)+Windows. This demobox requires a few functions from the Anisotropic Nonparametric Image Restoration Demobox of LASIP Project (available for download at http://www.cs.tut.fi/~lasip/). It is suggested that the Pointwise Shape-Adaptive DCT Demobox and the Anisotropic Nonparametric Image Restoration Demobox are installed in the same folder. The deblocking demo demo_SADCT_deblocking.m uses the function jpeg_read.dll from the Matlab JPEG Toolbox by Phil Sallee. A complete release of this toolbox is included in our distribution in the file JPEG_Toolbox_MATLAB_1.4.zip. What's new in this release: changes from version 1.40 to 1.31: + added: demo_SADCT_inverse_halftoning.m (demo for reconstruction of grayscale images from binary halftones) * updated: function_rgb2LumChrom.m (cleaner code) * updated: function_LumChrom2rgb.m (cleaner code) * updated: demo_SADCT_deblurring.m (cleaner code) * updated: demo_SADCT_deblocking.m (added option to disable coefficient alignment) * updated: demo_SADCT_denoising.m (added option to disable coefficient alignment) * updated: demo_SADCT_color_denoising.m (added option to disable coefficient alignment) * updated: function_SADCT_RW_wiener.dll (minor fix to regularization term) changes from version 1.31 to 1.30: ! bugfix: demo_SADCT_deblurring.m (fixed: calculation of RW variance) changes from version 1.30 to 1.20: * updated: demo_SADCT_deblocking.m (added option to disable coefficient alignment) * updated: demo_SADCT_denoising.m (added option to disable coefficient alignment) * updated: demo_SADCT_color_denoising.m (added option to disable coefficient alignment) * updated: function_SADCT_thresholding_fast.dll (now works with Pentium3/Athlon CPU, coeff. alignment can be disabled) * updated: function_SADCT_wiener_fast.dll (now works with Pentium3/Athlon CPU, coeff. alignment can be disabled) * updated: function_SADCT_RI_thresholding.dll (now works with Pentium3/Athlon CPU) * updated: function_SADCT_RW_wiener.dll (now works with Pentium3/Athlon CPU) changes from version 1.10 to 1.20: + added: demo_SADCT_deblocking.m (demo for deblocking and deringing of grayscale and color images) * updated: demo_SADCT_denoising.m (added blocksize parameter to constrain the maximum size of the filtering block) * updated: demo_SADCT_color_denoising.m (added blocksize parameter to constrain the maximum size of the filtering block) * updated: function_shape_explorer.m (added blocksize parameter to constrain the maximum size of the filtering block) changes from version 1.04 to 1.10: + added: function_AnisLPAICI8.p (enables much faster calculation of h_opt_Q, lower memory requirements) * updated: demo_SADCT_denoising.m (uses function_AnisLPAICI8, added speed-up factor parameter) * updated: demo_SADCT_color_denoising.m (uses function_AnisLPAICI8, added speed-up factor parameter) changes from version 1.03 to 1.04: ! bugfix: demo_SADCT_deblurring.m (fixed: boundary handling for calculation of variance of LPA estimates) changes from version 1.02 to 1.03: * updated: demo_SADCT_denoising.m (improved comments, cleaner code, new image cropping, and more visualization options) * updated: demo_SADCT_color_denoising.m (improved comments, cleaner code, new image cropping, and more visualization options) * updated: function_rgb2LumChrom.m (improved comments and cleaner code) * updated: function_LumChrom2rgb.m (improved comments and cleaner code) * updated: function_shape_explorer.m (improved comments) * updated: function_SADCT_thresholding_fast.dll (faster processing of constant blocks) changes from version 1.01 to 1.02: ! bugfix: function_SADCT_thresholding_fast.dll (fixed: number of harmonics for square blocks) changes from version 1.00 to 1.01: ! bugfix: function_SADCT_thresholding_fast.dll (fixed: preservation of mean for square blocks) + added: function_shape_explorer.m (shows adaptive-shape supports of SA-DCT) Future additions: A Linux version of the SA-DCT functions might be released in the future. Updates, examples, publications, presentations, etc. can be found at http://www.cs.tut.fi/~foi/SA-DCT Disclaimer: Any unauthorized use of these routines for industrial or profit-oriented activities is expressively prohibited. By downloading and/or using any of these files, you implicitly agree to all the terms of the TUT limited license (included in the file Legal_Notice.txt). Feedback: The authors of the Pointwise Shape-Adaptive DCT Demobox can be contacted at their respective email addresses firstname.lastname@tut.fi