Why block artifacts occur? EE5965 Advanced Image Processing Copyright Xin Li x f(x) original x f(x) ^ B2B3B JPEG decoded at low bit rate Only DC component is preserved
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
EE5965 Advanced Image Processing Copyright Xin Li 2009-
20121
Post-processing: Fighting Against Coding Artifacts
• Deblocking of DCT coded images– Image smoothing based approach– Wavelet-thresholding based approach
• Deringing of wavelet coded images– Re-compression approach– PDE-based approach
• Post-processing by alternating projections: a unified approach
EE5965 Advanced Image Processing Copyright Xin Li 2009-
20122
Block Artifacts
Smooth areas become blocky JPEG decoded image at 0.23bpp
zoom
Why block artifacts occur?
EE5965 Advanced Image Processing Copyright Xin Li 2009-
20123
x
f(x)
original
x
f(x)^
B 2B 3B
JPEGdecodedat lowbit rate
Only DC component is preserved
EE5965 Advanced Image Processing Copyright Xin Li 2009-
20124
Deblocking as Denoising
Standard image
denoisingalgorithm
JPEG compressed
imagepostprocessed
image
th 1000
Manually tune the threshold parameter!
EE5965 Advanced Image Processing Copyright Xin Li 2009-
20125
Experiment Results
before deblocking after deblocking
EE5965 Advanced Image Processing Copyright Xin Li 2009-
20126
Wavelet-based Deblocking
Behavior of block artifacts in wavelet domain
Fundamental Issues behind Deblocking
• Motivation - modeling uncertainty– Location of block artifacts is known (block
boundaries)– How to distinguish significant coefficients
generated by artifacts from those associated with true edges
• Strategy– Recall how JPEG2000 is free from block artifacts
EE5965 Advanced Image Processing Copyright Xin Li 2009-
20127
EE5965 Advanced Image Processing Copyright Xin Li 2009-
20128
Deblocking via Wavelet Thresholding
WaveletTransform
BlockWavelet
Transform reorder
X
X
Y1
Y2
X: JPEG decoded image
• Apply both WT and block-based WT to X to get Y1,Y2;
• Locate the coefficients at block boundaries;• If |Y1(i,j)|>T and |Y2(i,j)|<T, apply soft
thresholding to Y1(i,j);• Apply IWT to processed Y1 to obtain
deblocked image
EE5965 Advanced Image Processing Copyright Xin Li 2009-
20129
Deblocking Algorithm
EE5965 Advanced Image Processing Copyright Xin Li 2009-
201210
Example
Before deblocking (PSNR=27.39dB)After deblocking (PSNR=28.07dB)
EE5965 Advanced Image Processing Copyright Xin Li 2009-
201211
Ringing Artifacts
Sharp edges become unnatural JPEG2000 decoded image at 0.125bpp
zoom
Why ringing artifacts occur?
EE5965 Advanced Image Processing Copyright Xin Li 2009-
EE5965 Advanced Image Processing Copyright Xin Li 2009-
201213
Deringing by Re-compression
JPEG: JPEG2000 encoder JPEG-1: JPEG2000 decoder
Example
EE5965 Advanced Image Processing Copyright Xin Li 2009-
201214
before processing after processing
PDE-based Deringing
• The power of anisotropic diffusion– Nonlinear diffusion can handle a variety of
noise– Which PDE is suitable for deringing?– Implication into wavelet coding
EE5965 Advanced Image Processing Copyright Xin Li 2009-
201215
EE5965 Advanced Image Processing Copyright Xin Li 2009-
201216
Perona-Malik Filtering
PSNR=30.86dBPSNR=31.09dB
EE5965 Advanced Image Processing Copyright Xin Li 2009-
201217
Mean Curvature Filtering
PSNR=31.09dB PSNR=30.27dB
EE5965 Advanced Image Processing Copyright Xin Li 2009-
201218
Post-processing: Fighting Against Coding Artifacts
• Deblocking of DCT coded images– Image smoothing based approach– Wavelet-thresholding based approach
• Deringing of wavelet coded images– Re-compression approach– PDE-based approach
• Post-processing by alternating projections: a unified approach
Recall: Alternating Projection
EE5965 Advanced Image Processing Copyright Xin Li 2009-
201219
X0
X1
X2
X∞
Projection-Onto-Convex-Set (POCS) Theorem: If C1,…,Ck areconvex sets, then alternating projection P1,…,Pk will convergeto the intersection of C1,…,Ck if it is not empty Alternating projection does not always converge in the caseof non-convex set. Can you think of any counter-example?
C1
C2
EE5965 Advanced Image Processing Copyright Xin Li 2009-
201220
Projection Operators
● Constraint set QTC
otherwisexTyxTyTyxTy
xPTfPTfP QQQT 2/2/2/2/
)()},({1
y y+T/2y-T/2
● Constraint set sC fffPs )(
xB xB
f Ps(f)
EE5965 Advanced Image Processing Copyright Xin Li 2009-
201221
Projection-based Deblocking● DCT quantization set
}][|{},|{, gfQfCgTffCCCC QTQTQT
DCT Quantization● Smoothness constraint set
Cs={f|f is smooth in the block boundaries}
}||||{ 2 DffCs at block boundaries
Linear edge detection operatorYongyi Yang; Galatsanos, N.P.; Katsaggelos, A.K.; , "Projection-based spatially adaptive reconstruction of block-transform compressed images,“ IEEE Trans. on Image Proc., vol.4, no.7, pp.896-908, Jul 1995
EE5965 Advanced Image Processing Copyright Xin Li 2009-
201222
Projection-based Deringing
● WT quantization set
}][|{},|{, gfQfCgTffCCCC QTQTQT
WT Quantization● Smoothness constraint set
][,1, IcIcIcIcII WWEESSNN
tji
tji
Perona-Malik diffusion as a nonlinear projection operator
Xin Li; , "Improved wavelet decoding via set theoretic estimation," IEEE Trans. on CSVT, vol.15, no.1, pp. 108- 112, Jan. 2005
Algorithm Flowchart
EE5965 Advanced Image Processing Copyright Xin Li 2009-
201223
C1 : observation constraint set
C2 : regularization constraint set
Summary
• Connection with models (PDE-based, wavelet-based, patch-based)– They serve as image prior/regularization
constraint set– Jointly work with quantization (observation data)
constraint set• Convergence is NOT always guaranteed but
can be terminated strategically.
EE5965 Advanced Image Processing Copyright Xin Li 2009-