TY - CHAP
T1 - Space-variant and adaptive transform domain image restoration methods
AU - Yaroslavsky, L.
PY - 2007
Y1 - 2007
N2 - A family of space-time variant local adaptive transform domain methods for signal, image, and video processing (denoising, deblurring, enhancement) is described. The methods work in a space-time moving window in the domain of an orthogonal transform, and in each position of the window, nonlinearly modify the signal transform coefficients to obtain an estimate of the window central pixel. For the design of the transform coefficient nonlinear processing algorithm, minimum RMS restoration error approach is applied. This leads to transformdomain adaptive empirical Wiener filtering in form of the coefficient "soft" or "hard" thresholding. It is shown that among different possible transforms, discrete cosine transform proved to be the primer candidate for using in image and video processing. Very good edge-preserving noise-suppression capability of sliding window DCT domain (SWDCT) image denoising algorithms was confirmed experimentally on test and real-life images and was compared to that of the local "ideal" Wiener filtering, which they approximate. Applications of the algorithms for image denoising, deblurring, and enhancement are illustrated by numerous examples. Theoretically, sliding window transform-domain filtering methods can be treated as an implementation of image subband decomposition and nonlinear pointwise transformation of the subband components. Being considered in this way, they parallel wavelet denoising methods that exploit multiresolution property of wavelet transforms for enabling local adaptivity of the filtering. Image denoising capabilities of both families of image denoising methods are compared in experiments with test and real-life images and ID signals, which revealed the superiority of the SWDCT filtering in this respect. As a way to efficiently implement parallel SWDCT processing in windows of different sizes and to solve in this way the problem of selecting appropriate window size in SWDCT filtering, combining wavelet image multiresolution decomposition and SWDCT filtering in a hybrid processing is considered, and further improvement of filtering performance is demonstrated in extensive simulation experiments.
AB - A family of space-time variant local adaptive transform domain methods for signal, image, and video processing (denoising, deblurring, enhancement) is described. The methods work in a space-time moving window in the domain of an orthogonal transform, and in each position of the window, nonlinearly modify the signal transform coefficients to obtain an estimate of the window central pixel. For the design of the transform coefficient nonlinear processing algorithm, minimum RMS restoration error approach is applied. This leads to transformdomain adaptive empirical Wiener filtering in form of the coefficient "soft" or "hard" thresholding. It is shown that among different possible transforms, discrete cosine transform proved to be the primer candidate for using in image and video processing. Very good edge-preserving noise-suppression capability of sliding window DCT domain (SWDCT) image denoising algorithms was confirmed experimentally on test and real-life images and was compared to that of the local "ideal" Wiener filtering, which they approximate. Applications of the algorithms for image denoising, deblurring, and enhancement are illustrated by numerous examples. Theoretically, sliding window transform-domain filtering methods can be treated as an implementation of image subband decomposition and nonlinear pointwise transformation of the subband components. Being considered in this way, they parallel wavelet denoising methods that exploit multiresolution property of wavelet transforms for enabling local adaptivity of the filtering. Image denoising capabilities of both families of image denoising methods are compared in experiments with test and real-life images and ID signals, which revealed the superiority of the SWDCT filtering in this respect. As a way to efficiently implement parallel SWDCT processing in windows of different sizes and to solve in this way the problem of selecting appropriate window size in SWDCT filtering, combining wavelet image multiresolution decomposition and SWDCT filtering in a hybrid processing is considered, and further improvement of filtering performance is demonstrated in extensive simulation experiments.
UR - http://www.scopus.com/inward/record.url?scp=34548223309&partnerID=8YFLogxK
M3 - ???researchoutput.researchoutputtypes.contributiontobookanthology.chapter???
AN - SCOPUS:34548223309
SN - 9775945550
SN - 9789775945556
T3 - Eurasip Book Series on Signal Processing and Communications
SP - 201
EP - 240
BT - Advances in Nonlinear Signal and Image Processing
PB - Hindawi Publishing Corporation
ER -