This action might not be possible to undo. Are you sure you want to continue?

D˜ ung T. V˜ o

∗

Digital Media Solutions Lab, SISA

(Samsung US R&D Center), Irvine, CA, USA

Truong Q. Nguyen

Video Processing Lab, ECE Department

UC San Diego, La Jolla, CA, USA

ABSTRACT

The paper proposes a novel method for coding artifact reduction

in compressed images. For removing blocking artifacts, a local-

ized DCT-based ﬁlter with condition on the similarity between

surrounding blocks is considered. To reduce ringing, a localized

fuzzy ﬁlter is utilized to avoid the blurry effect of linear ﬁlter and

painting-like effect of conventional fuzzy ﬁlter. To enhance chroma

components and reduce the color bleeding, the localized ﬁlter for

luma component are implemented for the chroma components. Sim-

ulations on a wide range of compressed images are performed to

verify the effectiveness of the algorithm.

Index Terms— coding artifact reduction, fuzzy ﬁlter, localized

ﬁltering.

1. INTRODUCTION

Visual quality of block-based compressed images is degraded by

blocking artifacts, ringing artifacts and color bleeding, especially

at low-bit-rate coding. Blocking artifacts occur at the border of

neighboring blocks when each block is processed independently

with coarse quantization of discrete cosine transform (DCT) coef-

ﬁcients. In natural images, high frequency components tend to have

smaller values than the low frequency components. But they are

quantized with higher quantization step size than the low frequency

components. This makes the high frequency components more

distorted or even rounded to zero value. Loss in high frequencies

will cause ringing artifacts, which affect the visual quality seriously

in detail areas. For chroma components, they are downsampled

and then compressed with even higher quantization matrix than the

luma components. Color becomes blocky and mismatchs to its luma

information. These lead to color bleeding.

Visual effects of coding artifacts vary fromone codecs to another

but these artifacts always have directional and data-dependent prop-

erties. Because of the block-based compression, blocking artifacts

in luma and chroma components occur at the horizontal and verti-

cal directions at pixels between 2 blocks. Furthermore, due to the

loss of high frequency components during the coarse quantization,

the ringing artifacts appear along the strong edges. The data depen-

dent property is from the usage of the same quantization step size

matrix for every blocks. Coding artifacts degrade the visual quality

more in detail areas than in ﬂat areas. These characteristics make the

coding artifacts different than the recording noise, where it is usu-

ally assumed as additive zero-mean white Gaussian noise [1]. Fig. 1

shows the differences between additive Gaussian noise and coding

artifacts in MJPEG compression. The noisy image in Fig. 1.(a) is

degraded by additive Gaussian noise with variance of 0.01 while the

compressed image in Fig. 1.(b) is encoded with scaling factor of 4

∗

D˜ ung T. V˜ o performed this work while at UC San Diego

(a) Noisy image (b) Compressed image

Fig. 1. Comparison on visual effect between coding artifacts and

Gaussian noise.

for the quantization step size matrix. As shown in this example,

Gaussian noise is uniformly distributed over pixels while the coding

artifacts is non-uniformly distributed due to their special character-

istics. This implies that these coding artifacts should be treated in a

different way other than methods for denoising in pre-processing.

Many pixel-domain and DCT-domain post-processing approa-

ches have been considered for blocking artifact reduction . These in-

clude low-pass ﬁltering [2] and adaptive median ﬁltering [3] which

were applied to remove the high frequencies caused by sharp edges

between adjacent blocks. In H.264/AVC, an adaptive deblocking ﬁl-

ter [4] was proposed to selectively ﬁlter the artifacts in the coded

block boundaries. With assumption on the small changes of neigh-

boring DCT coefﬁcients at the same frequency in a small region, a

ﬁxed low pass ﬁlter in [5] or adaptively weighted low pass ﬁlters in

[6] and [7] are applied to the transform coefﬁcients of the shifted

blocks. Although effectively reducing the blocking artifacts, these

mentioned methods also blur the output images.

To reduce ringing artifacts, edges or edgy areas which contain

ringing are detected and then are processed with an adaptive ﬁlter

as in [8] or a gray-level morphological nonlinear smoothing ﬁlter

as in [9]. These ﬁlters remove the ringing artifacts, but at the same

time also reduce the details of the images. To avoid the blurry effect,

isotropic fuzzy ﬁlters [10] ﬁlters are used, but their output sometime

have painting-like effects with very strong edges above smooth ar-

eas.

This paper proposes a novel method to reduce the blocking, ring-

ing and color bleeding artifacts using localized ﬁltering. For de-

blocking, a conditionally lowpass ﬁlter is applied for local shifted

blocks around the block of interest. To avoiding blurry effect, a novel

fuzzy ﬁlter is proposed for deringing. Filter coefﬁcients of this ﬁl-

ter are weightedly locally to reduce the painting-like effect of fuzzy

ﬁlters. The paper is organized as follows. Section 2 introduces the

conditionally localized DCT-based ﬁlters for deblocking. Section 3

describes the localized fuzzy ﬁlter for deringing. Simulations and

1269 978-1-4577-0539-7/11/$26.00 ©2011 IEEE ICASSP 2011

T

c

m

i

'E

n

i

1

2

3

4

xq

xq,i

Fig. 2. Translation between blocks of image xs and x.

comparison results are presented in Section 4. Finally, Section 5

gives the concluding remarks.

2. CONDITIONALLY LOCALIZED DCT-BASED

DEBLOCKING FILTERS

Assume the image Y (m, n) is compressed to Yq(m, n) by N × N

blockwise processing with a transform T followed by a quantization

matrix kQ where Q is the standard quantization matrix and k is the

scaling factor.

˜

Y and

˜

Yq are the transformed coefﬁcients of Y and its

quantized value, respectively. Because of the block-size processing,

the correlation between the N ×N block of pixels are destroyed. To

increase this cross-block correlation, a DCT-based domain low pass

ﬁlter [6] is applied through the surrounding N × N blocks Yq(m+

mi, n+ni) together with the block of interest Yq(m, n), as shown

in Fig.2. Assume that the DCT transform (type II) of block Yq(m+

mi, n+ni) is

˜

Yq,m

i

,n

i

˜

Yq,m

i

,n

i

(u, v) =

2

N

kukv

N−1

m=0

N−1

n=0

Yq(m+mi, n+ni)C

m

u

C

n

v

for m, n = 0, . . . , N−1 (1)

and

˜

Yq,0,0 =Xq where the shifted motion vector is considered equal

to 0. The localized DCT-based deblocking ﬁlter is applied to the

DCT coefﬁcients at the same coefﬁcient of these local DCT coef-

ﬁcients to form the enhanced DCT coefﬁcients

˜

Y

q

˜

Y

q

(u, v) =

(m

i

,n

i

)∈Ω

DB

WDB(mi, ni)

˜

Yq,m

i

,n

i

(u, v)

for u, v = 0, . . . , N−1. (2)

where WDB is the lowpass ﬁlter coefﬁcients for deblocking and ΩDB

is the local window centered at location (m, n) which consists all

(mi, ni). To avoid the reconstructed output

˜

Y

q

biasing from

˜

Yq, the

ﬁlter coefﬁcient must satisfy

(m

i

,n

i

)∈Ω

DB

WDB(mi, ni) = 1 ∀(mi, ni) ∈ ΩDB (3)

Normally

WDB(0, 0) ≥ WDB(mi, ni). (4)

so that the DCT coefﬁcients of the center blocks have highest con-

tribution to the output. The enhanced N × N block Y

q

is recon-

structed by inverse DCT transform. For highly compressed images,

the high DCT frequencies tend to have zero value. As can be seen

in (2), these coefﬁcients are recreated because of the non-zero val-

ues of

˜

Yq,m

i

,n

i

(u, v). In [11], a condition of quality enhancement

was considered for the linear temporal ﬁlter over many frames. It

is shown that the difference between the block of interest and the

temporally surrounding blocks must be small enough so that these

blocks are rather related to each other to achieve the quality improve-

ment. Similarly in this paper for image enhancement, a condition of

the small difference between the block of interest and the spatially

surrounding blocks is implemented as follows

N−1

m=0

N−1

n=0

¸

¸

¸Yq(m+mi, n +ni) −Yq(m, n)

¸

¸

¸ ≤ Th (5)

where Th is a threshold. Only surrounding blocks which satisfy this

condition will be locally ﬁltered using coefﬁcients WDB(mi, ni).

3. LOCALIZED FUZZY DERINGING FILTERS

Although the localized deblocking ﬁlters as described in Section 2

can reduce the blocking artifacts and ringing artifacts, it will blur

the images due to its lowpass ﬁltering. The deblocking ﬁlter should

be only applied to pixels near to block borders, not all pixels in the

blocks

YDB(m, n) =

_

Y

q

(m, n) if (m, n) near to block borders

Yq(m, n) otherwise.

(6)

To reduce the ringing artifact, this section will discuss a novel local-

ized fuzzy ﬁlters. Fuzzy ﬁlters are non-linear ﬁlters which are based

on the real-valued spatial-rank relation, as described in [10][12].

Assume that a normalized fuzzy ﬁlter hDR,n is applied to a set

ΩDR of neighboring samples YDB(m+mi, n+ni) around the input

YDB(m, n) to form the output

YDR(m, n) =

(m

i

,n

i

)∈Ω

DR

hDR,n(mi, ni)YDB(m+mi, n+ni) (7)

where hDR(mi, ni, n) is calculated by

hDR,n(mi, ni) =

hDR

_

YDB(m+mi, n+ni), YDB(m, n)

_

(m

i

,n

i

)∈Ω

DB

hDR

_

YDB(m+mi, n+ni), YDB(m, n)

_

(8)

where hDR is the ﬁlter coefﬁcient which controls the contribution of

the input YDB(m+mi, n+ni) to the output. Fuzzy ﬁlters are signal

dependent and help removing the ringing artifact while preserving

the details. They depend on the distance between the surrounding

pixel YDB(m+mi, n+ni) and the pixel of interest YDB(m, n). Fol-

low the spatial-rank relation, the ﬁlter coefﬁcients hDR must fulﬁll

the constraints

lim

|Y

DB

(m+m

i

,n+n

i

)−Y

DB

(m,n)|→0

hDR

_

YDB(m+mi, n+ni), YDB(m, n)

_

=1

(9)

lim

|Y

DB

(m+m

i

,n+n

i

)−Y

DB

(m,n)|→∞

hDR

_

YDB(m+mi, n+ni), YDB(m, n)

_

=0

(10)

and

h

_

YDB(m+mi,1, n+ni,1), YDB(m, n)

_

≥h

_

YDB(m+mi,2, n+ni,2), YDB(m, n)

_

if | YDB(m+mi,1,n+ni,1)−YDB(m, n) |

≤| YDB(m+mi,2, n+ni,2)−YDB(m, n) | . (11)

1270

c

c

c

c

c

c

c

c

c

c

c

c

c

c

a

a

a

c

c

c

c

c

c

c

a

a

a

c

c

c

a

a

a

s

s

s

c

c

c

c

s

s

s

c

c

c

c

s

s

s

c

c

c

c

c

c

c

c

(1,−1)

(m

i

,n

i

)=(1,−1)

(0,−0)

'

'

'

'

'

'

'

'

'

(−1,2)

(m

i

,n

i

)=(−1,0)

Fig. 3. Localized coefﬁcients of fuzzy ﬁlters.

hDR is referred to as the membership function and there are

many functions which fulﬁll these requirements. An example is the

Gaussian membership function,

hDR

_

YDB(m+mi, n+ni), YDB(m, n)

_

=exp

_

−

_

Y

DB

(m+m

i

,n+n

i

)−Y

DB

(m,n)

_

2

2σ

2

_

(12)

where σ represents the spread parameter of the input and controls

the strength of the fuzzy ﬁlter. Note that the contribution of the input

YDB(m, n) to the output is always highest compared to the contri-

bution of other samples

h

_

YDB[m, n], YDB[m, n]

_

=1

≥ h

_

YDB[m+m

, n+n

], YDB[m, n]

_

∀(m

, n

). (13)

For the same | YDB[m+m

, n+n

**]−YDB[m, n] |, the higher the σ
**

value, the higher the contribution of YDB[m+m

, n+n

] relatively

compared to the contribution of YDB[m, n] to the output. This im-

plies that YDB[m, n] will be more averaged to YDB[m+m

, n+n

].

Smaller σ values will keep the signal YDB[m, n] more isolated from

its neighboring samples.

An example of calculating the coefﬁcients of the localized fuzzy

ﬁlters is shown in Fig.3. The top right area visualizes the way cal-

culating fuzzy coefﬁcient at pixel (m, n) = (1, −1). To estimate

the spatial-rank relation between pixel of interest (m, n) = (1, −1)

and its surrounding pixel at (2, −2), the conventional fuzzy ﬁlter

is based only on the value of these two pixels using (12). If these

two pixels represent an edge with similar values, its small differ-

ence leads to coefﬁcients having values closed to unity (as observed

from (9)). They would dominate the coefﬁcients in other directions

which are different than the edge direction. This will cause the

painting-like effect. To reduce this effect, the spatial-rank relation

is proposed to be a weighted combination of the spatial-rank rela-

tion of surrounding pairs of pixels where their position difference are

(mi, ni) = (1, −1). These relations are shown as the 45

o

arrows in

the top right of Fig.3. The pairs of pixels are determined by the shift

positions (mii, nii) ∈ V , where V is the 3 × 3 local window cen-

tered by pixel at (m, n) = (1, −1) in this example. Another exam-

ple with (m, n) = (−1, 2) and direction with (mi, ni) = (−1, 0)

is shown in the bottom left area of Fig.3.

hDR in this case is not calculated using (12) but is a lowpass

averaging of the ﬁlter coefﬁcient inside the window V which is cen-

tered by the pixel of interest at (m, n)

hDR

_

YDB(m+mi, n+ni), YDB(m, n)

_

(14)

=

(m

ii

,n

ii

)∈V

_

WDB(mii, nii)×

hDR,m

ii

,n

ii

_

YDB(m+mi +mii, n+ni +nii), YDB(m+mii, n+nii)

_

_

E

U

V

DCT

E

Deblocking

E

IDCT

E

2 T

E

Deringing

E

U

DR

V

DR

Y

E

DCT

E

Deblocking

E

IDCT

E

Deringing

c

E

Y

DR

Fig. 4. Block diagram of the proposed localized ﬁltering scheme.

where WDR(mii, nii) is the weighting factor of each hDR,m

ii

,n

ii

to

the ﬁlter coefﬁcient hDR. hDR,m

ii

,n

ii

is calculated by

hDR,m

ii

,n

ii

_

YDB(m+mi +mii, n+ni +nii), YDB(m+mii, n+nii)

_

=exp

_

−

_

YDB(m+mi +mii,n+ni +nii)−YDB(m+mii,n+nii)

_

2σ

2

_

.

(15)

For chroma components, the localized algorithm cannot base on

U and V because they don’t have enough information. In this case,

the algorithm is extended for chroma components by using the same

localized fuzzy ﬁlter of luma component. This helps match the color

to the structure in the luma component and reduce the color bleed-

ing. The whole algorithm is shown in Fig.4. At ﬁrst, all components

are transformed to DCT domain and are deblocked using the local-

ized linear ﬁlter. They are then converted back the pixel domain by

IDCT transform. Next, the deblocked Y component is implemented

to form the coefﬁcients at the localized fuzzy ﬁlter, which is later

used for deringing. For U and V components, they are upsampled

to the same size of Y component, then are deringed by the localized

fuzzy ﬁlter from luma.

4. SIMULATION RESULTS

Simulations are performed to verify the effectiveness of the local-

ized ﬁlters in artifact reduction. The simulated database includes

different types of images from video sequences: Foreman, News,

Mother, Silent, Soccer, Bus. They are compressed using JPEG com-

pression with different scaling factors k of the quantization matrix.

For both deblocking and deringing localized ﬁlters, the weighting

matrices WDB and WDR are [1, 1, 1; 1, 3, 1; 1, 1, 1] /11. The local

windows ΩDB, ΩDR and V are all set as 3 × 3 pixels. The thresh-

old Th in the deblocking part is set to 250k

2

. Only 2 pixels at each

side of the block borders are replaced by the deblocked pixels to

avoid the blurry effect which is mentioned in Section 3. The spread

parameter of the fuzzy ﬁlter σ is also proportional to the scaling fac-

tor as 5k. For comparison, other methods of artifact reduction such

as Chen’s method [6], Liu’s method [7] and Kong’s method [10] are

also simulated. Fig.5 shows the zoomed-in result for the 6

th

frame of

News sequence which is compressed with scaling factor of 3. Com-

paring to the compressed image in Fig.4(b), the enhanced images

using Chen’s method (Fig.4(c)), Liu’s method (Fig.4(d)) and Kong’s

method (Fig.4(e)) can reduce most of blocking artifact and ringing

artifact. However they also introduce other artifacts such as blurri-

ness (Chen’s and Liu’s methods) or painting-like effect with dom-

inant edges and very few details (Kong’s method). The enhanced

images using Chen’s method and Liu’s method are rather blurry and

still have some ringing artifacts at strong edge areas. The enhanced

image using Kong’s method is sharper with less ringing artifacts than

the enhanced image using Chen’s method or Liu’s method. But it has

the painting-like effect with cleared out details, as can be seen at the

face of the spokeswoman. The enhanced image using the proposed

localized ﬁlter in Fig.4(f) achieves the best visual quality comparing

1271

(a) Original (b) Compressed (28.45dB)

(c) Chen’s method (28.32dB) (d) Liu’s method (28.46dB)

(e) Kong’s method (28.90dB) (f) Proposed method (29.25 dB)

Fig. 5. Comparison for 6

th

frame of News sequence.

Table 1. Comparison of PSNR in units of dB for different methods.

Images 3Q Chen Liu Kong Localized Filters

News 28.45 28.32 28.46 28.90 29.15

Silent 28.85 29.40 29.32 29.67 29.76

Foreman 29.09 29.45 29.44 30.09 30.15

Soccer 27.35 27.54 27.50 27.58 27.92

Mother 32.36 32.95 32.81 33.16 33.39

Bus 26.29 25.80 25.93 26.52 26.64

Average gain 3Q +0.1809 +0.1794 +0.5873 +0.7694

4Q −0.0553 +0.0008 +0.4074 +0.5456

2Q +0.0407 −0.0391 +0.3319 +0.4955

to other methods. It both effectively removes blocking and ring-

ing artifacts and still keeps the details and edge sharpness. A close

inspection at the dress of the dancing woman concludes that the pro-

posed method is more effective in color bleeding reduction, com-

paring to other methods. The PSNR of enhanced image using the

proposed method also achieves highest value (29.25 dB), compar-

ing to the compressed image (28.45dB), Chen’s method (28.32dB),

Liu’s method (28.46dB) and Kong’s method (28.90dB).

The comparison in term of PSNR values is shown Table.1.

Over many types of images which are compressed with scal-

ing factor of 3, the proposed localized ﬁlters provide the highest

PSNR improvement of +0.7694 dB comparing to Chen’s method

(+0.1809 dB), Liu’s method (+0.1794 dB) and Kong’s method

(+0.5873 dB). It also outperforms other methods in different scal-

ing factors, as also shown in Table.1. The localized fuzzy ﬁlter

can be also applied to JPEG2000 compressed images or H.264

sequences to reduce the ringing artifacts.

5. CONCLUSIONS

The paper proposes a novel method for coding artifact reduction

using a conditionally localized DCT-based ﬁlter for deblocking and

a localized fuzzy ﬁlter for deringing. These ﬁlters effectively remove

the coding artifacts while avoiding the blurry effect from linear ﬁl-

tering and painting-like effect from non-linear fuzzy ﬁlters. Future

works will focus on adaptive localized ﬁlters and advanced methods

for color bleeding reduction.

6. REFERENCES

[1] K. Dabov, A. Foi, V. Katkovnik, and K. Egiazarian, “Im-

age Denoising by Sparse 3-D Transform-Domain Collabora-

tive Filtering,” IEEE Trans. Image Process., vol. 16, pp. 2080–

2095, August 2007.

[2] T. Jarske, P. Haavisto, and I. Defee, “Post-ﬁltering methods for

reducing blocking effects from coded images,” IEEE Trans.

Cosumer Electronics, vol. 40, pp. 521–526, August. 1994.

[3] Y.F. Hsu and Y.C. Chen, “A New Adaptive Separate Median

Filter for Removing Blocking Effects,” IEEE Trans. Cosumer

Electronics, vol. 39, pp. 510–513, August 1993.

[4] P. List, A. Joch, J. Lainema, G. Bjontegaard, and M. Kar-

czewicz, “Adaptive Deblocking Filter,” IEEE Trans. Circuits

Syst. Video Technol., vol. 13, pp. 614?619, July 2003.

[5] A. Nosratinia, “Embedded post-processing for enhancement

of compressed images,” Proc. IEEE Data Compression Conf.,

pp. 62–71, 1999.

[6] T. Chen, H.R. Wu, and B. Qiu, “Adaptive postﬁltering of trans-

form coefﬁcients for the reduction of blocking artifacts,” IEEE

Trans. Circuits Syst. Video Technol., vol. 11, pp. 594–602, May.

2001.

[7] S. Liu and A.C. Bovik, “Efﬁcient DCT-domain blind measure-

ment and reduction of blocking artifacts,” IEEE Trans. Circuits

Syst. Video Technol., vol. 12, pp. 1139–1149, December. 2002.

[8] J. Hu, N. Sinaceur, F. Li, K.W. Tam, and Z. Fan, “Removal of

Blocking and Ringing Artifacts in Transform Coded Images,”

Proc. IEEE Int. Conf. Acoustics, Speech and Signal Process-

ing, vol. 4, pp. 2565–2568, April 1997.

[9] S.H. Oguz, Y.H. Hu, and T.Q. Nguyen, “Image coding ringing

artifact reduction using morphological post-ﬁltering,” Proc.

IEEE Int. Work. Multi. Signal Process., pp. 628–633, 1998.

[10] H.S. Kong, Y. Nie, A. Vetro, H. Sun, and K. Barner, “Adap-

tive fuzzy post-ﬁltering for highly compressed video,” Prof. of

IEEE Int. Conf. Image Proc., pp. 1802–1806, 2004.

[11] D.T. Vo and T.Q.Nguyen, “Quality Enhancement for Motion

JPEG using Temporal Redundacies,” IEEE Transactions on

Circuits and Systems for Video Technology., vol. 18, pp. 609–

619, May 2008.

[12] D.T. Vo, T.Q.Nguyen, S. Yea, and A. Vetro, “Adaptive Fuzzy

Filtering for Artifact Reduction in Compressed Images and

Videos,” IEEE Transactions on Image Processing., vol. 18,

pp. 1166–1178, May 2009.

1272

n) ≤ T h m=0 n=0 (5) Fig. not all pixels in the blocks YDB (m.n ni ) YDB(m | ∞ Y + + − . n) = 1 (9) hDR YDB(m+mi. n+ni ) (7) where hDR (mi .0 = Xq where the shifted motion vector is considered equal to 0. To increase this cross-block correlation. Finally. For highly compressed images.i 4 the high DCT frequencies tend to have zero value. It is shown that the difference between the block of interest and the temporally surrounding blocks must be small enough so that these blocks are rather related to each other to achieve the quality improvement. YDB(m. n) otherwise. n) is compressed to Yq (m. Because of the block-size processing. n+ni ).n (mi. a condition of quality enhancement was considered for the linear temporal ﬁlter over many frames. (11) 1270 . (4) and h YDB(m+mi. n) to form the output YDR (m. ni ).2.ni (u. Assume that a normalized fuzzy ﬁlter hDR.n is applied to a set ΩDR of neighboring samples YDB (m+mi .2 . respectively.mi . n) ≥ h YDB(m+mi. n) (mi. n) = 0 (10) ∀(mi . n) | .ni )∈ΩDB N− N− 1 1 hDR.ni (u. a condition of the small difference between the block of interest and the spatially surrounding blocks is implemented as follows 1 N− N− 1 Yq (m + mi . this section will discuss a novel localized fuzzy ﬁlters. n) by N × N blockwise processing with a transform T followed by a quantization matrix kQ where Q is the standard quantization matrix and k is the ˜ ˜ scaling factor.n+ni.2 . As can be seen in (2).n ni ) YDB(m | 0 Y + + − . ni) = hDR YDB (m+mi. YDB(m.mi .n i i where T h is a threshold. 2.ni (u. a DCT-based domain low pass ﬁlter [6] is applied through the surrounding N × N blocks Yq (m+ mi . 3. 0) ≥ WDB (mi . n) near to block borders Yq (m. n+ni ) and the pixel of interest YDB (m. Only surrounding blocks which satisfy this condition will be locally ﬁltered using coefﬁcients WDB (mi . n) if (m. v) = ku kv N m=0 n=0 for m. Fuzzy ﬁlters are signal dependent and help removing the ringing artifact while preserving the details.ni)∈ΩDB ˜ WDB (mi . .1 )−YDB(m.1 . . as shown in Fig. In [11]. Y and Yq are the transformed coefﬁcients of Y and its quantized value. . The deblocking ﬁlter should be only applied to pixels near to block borders. The enhanced N × N block Yq is reconstructed by inverse DCT transform. n+ni ) is Yq. . n) hDR YDB (m+mi. ni )YDB (m+mi. (2) where WDB is the lowpass ﬁlter coefﬁcients for deblocking and ΩDB is the local window centered at location (m. ni ) = 1 (mi . To avoid the reconstructed output Yq biasing from Yq . n = 0.m . as described in [10][12].n)→ lim hDR YDB(m+mi. N −1. n+ni.ni )∈ΩDR 2 m n ˜ Yq (m+mi . n). these coefﬁcients are recreated because of the non-zero val˜ ues of Yq. . ni ). comparison results are presented in Section 4. YDB(m. ni ). They depend on the distance between the surrounding pixel YDB (m+mi . . n+ni ). it will blur the images due to its lowpass ﬁltering.1 . Follow the spatial-rank relation. n+ni. (6) To reduce the ringing artifact.2 ). . The localized DCT-based deblocking ﬁlter is applied to the DCT coefﬁcients at the same coefﬁcient of these local DCT coef˜ ﬁcients to form the enhanced DCT coefﬁcients Yq ˜ Yq (u. n) is calculated by hDR. n+ni ).ni )∈ΩDB (8) where hDR is the ﬁlter coefﬁcient which controls the contribution of the input YDB (m+mi.1 ). Section 5 gives the concluding remarks. n) which consists all ˜ ˜ (mi . n+ni ) to the output. Similarly in this paper for image enhancement. n + ni ) together with the block of interest Yq (m.0. N −1 (1) ˜ and Yq.xq m Ti c 1 3 ' E n i 2 xq. the ﬁlter coefﬁcient must satisfy WDB (mi . ni .mi . v). n+ni. YDB(m. n) if | YDB(m+mi. ni ) ∈ ΩDB (3) | DB(m mi. n+ni ). ni )Yq. n+ni ) around the input YDB (m.n (mi. n) | ≤| YDB(m+mi. n) = (mi. CONDITIONALLY LOCALIZED DCT-BASED DEBLOCKING FILTERS Assume the image Y (m. YDB (m. the ﬁlter coefﬁcients hDR must fulﬁll the constraints | DB(m mi. v = 0. Assume that the DCT transform (type II) of block Yq (m+ ˜ mi . the correlation between the N × N block of pixels are destroyed. n). n+ni )Cu Cv Yq. Fuzzy ﬁlters are non-linear ﬁlters which are based on the real-valued spatial-rank relation. v) = (mi .2 )−YDB(m. v) for u. so that the DCT coefﬁcients of the center blocks have highest contribution to the output. n) = Yq (m. 2. . Translation between blocks of image xs and x. YDB (m. LOCALIZED FUZZY DERINGING FILTERS Although the localized deblocking ﬁlters as described in Section 2 can reduce the blocking artifacts and ringing artifacts.n)→ lim Normally WDB (0. n + ni ) − Yq (m.

For U and V components. To estimate the spatial-rank relation between pixel of interest (m. For both deblocking and deringing localized ﬁlters. Smaller σ values will keep the signal YDB [m.4(c)).mii. n) = exp − YDB (m mi. hDR in this case is not calculated using (12) but is a lowpass averaging of the ﬁlter coefﬁcient inside the window V which is centered by the pixel of interest at (m.nii YDB (m+mi +mii. This helps match the color to the structure in the luma component and reduce the color bleeding. 0 0− ) Fig. ni ) = (1. They are compressed using JPEG compression with different scaling factors k of the quantization matrix.4(e)) can reduce most of blocking artifact and ringing artifact. The enhanced image using the proposed localized ﬁlter in Fig. n) to the output is always highest compared to the contribution of other samples h YDB [m. n+ni ). 1 1− ) Y E DCT E Deblocking E IDCT E Deringing YDR E U E DCT V c E Deblocking E IDCT E T E Deringing UDR E 2 VDR a a c 'c 'c 'c a a a c 'c ' c 'c a a a a c 'c 'c 'c ( 12 −. where WDR (mii . The enhanced image using Kong’s method is sharper with less ringing artifacts than the enhanced image using Chen’s method or Liu’s method. News. nii ) ∈ V .nii YDB (m+mi +mii. 3. YDB (m+mii.ni) (1 1 = . These relations are shown as the 45o arrows in the top right of Fig.n ni ) YDB (m. n) = (−1. n] |. Localized coefﬁcients of fuzzy ﬁlters. If these two pixels represent an edge with similar values. which is later used for deringing. Soccer. They are then converted back the pixel domain by IDCT transform. This will cause the painting-like effect.c c c c c c c c c c c c c c c c (mi. Fig. The enhanced images using Chen’s method and Liu’s method are rather blurry and still have some ringing artifacts at strong edge areas.n+ni +nii)−YDB (m+mii. 1.mii. The local windows ΩDB . the conventional fuzzy ﬁlter is based only on the value of these two pixels using (12).3. Liu’s method [7] and Kong’s method [10] are also simulated.3. n+n ]−YDB [m. 2σ 2 (15) Fig. The threshold T h in the deblocking part is set to 250k2 .nii )∈V For chroma components. n] ∀(m . In this case. The top right area visualizes the way calculating fuzzy coefﬁcient at pixel (m. n+n ].mii . all components are transformed to DCT domain and are deblocked using the localized linear ﬁlter. n+ni +nii). the algorithm is extended for chroma components by using the same localized fuzzy ﬁlter of luma component. the weighting matrices WDB and WDR are [1. An example of calculating the coefﬁcients of the localized fuzzy ﬁlters is shown in Fig. −1) in this example. Bus.5 shows the zoomed-in result for the 6th frame of News sequence which is compressed with scaling factor of 3. then are deringed by the localized fuzzy ﬁlter from luma. 1. However they also introduce other artifacts such as blurriness (Chen’s and Liu’s methods) or painting-like effect with dominant edges and very few details (Kong’s method).) ( i. Silent. where V is the 3 × 3 local window centered by pixel at (m.4(d)) and Kong’s method (Fig.nii to the ﬁlter coefﬁcient hDR . The pairs of pixels are determined by the shift positions (mii . n) = (1. 4. its small difference leads to coefﬁcients having values closed to unity (as observed from (9)). Liu’s method (Fig. YDB (m. n+nii) 1271 . 2) and direction with (mi .n) 2 + + − 2σ 2 (12) where σ represents the spread parameter of the input and controls the strength of the fuzzy ﬁlter. the enhanced images using Chen’s method (Fig. hDR YDB (m+mi. n] to the output. Note that the contribution of the input YDB (m.mii .nii is calculated by hDR. nii )× hDR. the higher the σ value. n) hDR YDB (m+mi. i) ( 10) m n =− . (13) For the same | YDB [m+m . Next. 1.4(b). the spatial-rank relation is proposed to be a weighted combination of the spatial-rank relation of surrounding pairs of pixels where their position difference are (mi . n+n ]. −1) and its surrounding pixel at (2. An example is the Gaussian membership function. −1). ni ) = (−1. Block diagram of the proposed localized ﬁltering scheme. 1. n+nii) = exp − YDB(m+mi +mii. Mother. To reduce this effect. n+ni ). 1. Only 2 pixels at each side of the block borders are replaced by the deblocked pixels to avoid the blurry effect which is mentioned in Section 3. n] = 1 ≥ h YDB [m+m . This implies that YDB [m. 1. YDB (m+mii. YDB [m. YDB (m. the higher the contribution of YDB [m+m .n+nii) . 1] /11. they are upsampled to the same size of Y component. the deblocked Y component is implemented to form the coefﬁcients at the localized fuzzy ﬁlter. −1).4. −2). 0) is shown in the bottom left area of Fig. n+ni +nii). 3. as can be seen at the face of the spokeswoman. hDR is referred to as the membership function and there are many functions which fulﬁll these requirements. n]. YDB [m.3. n) = (mii . nii ) is the weighting factor of each hDR. SIMULATION RESULTS Simulations are performed to verify the effectiveness of the localized ﬁlters in artifact reduction. But it has the painting-like effect with cleared out details.− ) s c s s s c s s s c s s c c c c c c c c c (. the localized algorithm cannot base on U and V because they don’t have enough information. For comparison. hDR. Another example with (m. The spread parameter of the fuzzy ﬁlter σ is also proportional to the scaling factor as 5k. n+n ] relatively compared to the contribution of YDB [m. The whole algorithm is shown in Fig. 4. The simulated database includes different types of images from video sequences: Foreman. (. ΩDR and V are all set as 3 × 3 pixels. They would dominate the coefﬁcients in other directions which are different than the edge direction. n) = (1. n) = (1. n] more isolated from its neighboring samples. n ).4(f) achieves the best visual quality comparing (14) WDB (mii . n] will be more averaged to YDB [m+m . Comparing to the compressed image in Fig. At ﬁrst. other methods of artifact reduction such as Chen’s method [6].

50 32. 1166–1178. A.H.5456 +0. 18.5873 +0. K. “Embedded post-processing for enhancement of compressed images. 11.Q. Video Technol. “Removal of Blocking and Ringing Artifacts in Transform Coded Images.1809 dB). Oguz.1. pp. Multi. and K. 1999. 2565–2568. Cosumer Electronics. vol. Hu. “Quality Enhancement for Motion JPEG using Temporal Redundacies. 40. Chen.46dB) (e) Kong’s method (28. 4. 609– 619. G. 12.25 dB).. “Adaptive Fuzzy Filtering for Artifact Reduction in Compressed Images and Videos.35 32.0553 +0. 6. Y. [4] P. 1802–1806.0391 Kong 28.F.” IEEE Trans. “Image Denoising by Sparse 3-D Transform-Domain Collaborative Filtering.C. Circuits Syst. [2] T. These ﬁlters effectively remove the coding artifacts while avoiding the blurry effect from linear ﬁltering and painting-like effect from non-linear fuzzy ﬁlters. CONCLUSIONS (a) Original (b) Compressed (28. Nie.80 +0. 2001.” IEEE Transactions on Image Processing.” Proc. P. Sinaceur. and K. IEEE Data Compression Conf. December.4074 +0.39 26. H.Nguyen.29 3Q 4Q 2Q Chen 28.1794 dB) and Kong’s method (+0. vol.81 25. The PSNR of enhanced image using the proposed method also achieves highest value (29.. August 1993. pp. It also outperforms other methods in different scaling factors. Bovik. 62–71. Image Proc.45dB) The paper proposes a novel method for coding artifact reduction using a conditionally localized DCT-based ﬁlter for deblocking and a localized fuzzy ﬁlter for deringing.” IEEE Trans.15 29. [11] D. pp.1. N. Hsu and Y. H. Egiazarian.. Barner..” Proc.25 dB) Fig. Wu. vol.” IEEE Trans. Circuits Syst. pp. Defee.. August. Nguyen.58 33. Images News Silent Foreman Soccer Mother Bus Average gain 3Q 28.09 27.45dB). Katkovnik. Comparison of PSNR in units of dB for different methods. Vo. Hu. [12] D. pp.” Proc. 16.Q.54 32. Qiu.16 26..ing factor of 3. 594–602. Table 1. Signal Process.52 +0. Chen’s method (28. Dabov. 1994.15 27. Fan. Liu’s method (28. Circuits Syst. Karczewicz.3319 Localized Filters 29. [10] H. Cosumer Electronics. T.0008 −0. Lainema.7694 +0.32 29..40 29. Liu’s method (+0. Conf. “Post-ﬁltering methods for reducing blocking effects from coded images.4955 to other methods. 2080– 2095.46dB) and Kong’s method (28. Nosratinia. 13.H. Work.” IEEE Transactions on Circuits and Systems for Video Technology.” IEEE Trans. and T. 39. pp. Vetro. A close inspection at the dress of the dancing woman concludes that the proposed method is more effective in color bleeding reduction. Joch. Future works will focus on adaptive localized ﬁlters and advanced methods for color bleeding reduction. pp. pp. pp.Q.67 30.1809 −0. 614?619. S. Sun.Nguyen. Comparison for 6th frame of News sequence. 510–513. “Adaptive fuzzy post-ﬁltering for highly compressed video.T. Acoustics. Over many types of images which are compressed with scal- 1272 .09 27. “Image coding ringing artifact reduction using morphological post-ﬁltering.32dB). and B.W.” IEEE Trans. 1139–1149.C. Bjontegaard.90dB).. “Adaptive Deblocking Filter. vol. It both effectively removes blocking and ringing artifacts and still keeps the details and edge sharpness.1794 +0. V. 5.64 +0. Vetro.76 30. pp.. A. Kong. 628–633. of IEEE Int. vol. [3] Y.5873 dB).45 27.32 29. 5. (c) Chen’s method (28. Video Technol. Y.46 29. April 1997. 521–526. Liu and A. as also shown in Table. and Z. Chen.R. “Adaptive postﬁltering of transform coefﬁcients for the reduction of blocking artifacts.95 25.264 sequences to reduce the ringing artifacts. pp. Vo and T.32dB) (d) Liu’s method (28. May 2009.7694 dB comparing to Chen’s method (+0. August 2007.” IEEE Trans. List. May. 18. A.45 28. Image Process. [7] S. IEEE Int.92 33.36 26. comparing to the compressed image (28. Haavisto. July 2003. “Efﬁcient DCT-domain blind measurement and reduction of blocking artifacts.T. Conf. [8] J. Tam. vol.93 +0. [9] S.90dB) (f) Proposed method (29. F. REFERENCES [1] K. vol. Li. and M. 1998. Jarske.44 27. vol. 2004. Foi. and I. vol. Speech and Signal Processing. 2002. IEEE Int. pp.” Prof. J. [5] A. May 2008. Yea.S.0407 Liu 28. comparing to other methods. the proposed localized ﬁlters provide the highest P SN R improvement of +0.85 29. “A New Adaptive Separate Median Filter for Removing Blocking Effects.90 29. and A. The localized fuzzy ﬁlter can be also applied to JPEG2000 compressed images or H. [6] T. Video Technol. The comparison in term of P SN R values is shown Table.

Sign up to vote on this title

UsefulNot useful- Filter Design
- Simple Mountains in GIMP
- ds_2Gb_DDR3(B-ver)based_SODIMMs(Rev.0.4).pdf
- [IJET-V2I3_1P10] Authors
- ds_1Gb_DDR3(B-ver)based_SODIMM(Rev.1.0)
- 14_IJAR
- Inverse Systems
- 277-D020
- A Blind Robust Watermarking Scheme
- 03 Morph
- lecture5.pdf
- Advan
- sarkar_ACM_MMSec_09.pdf
- Least Squares Subspace Projection Approach to Mixed Pixel Classiﬁcation for Hyperspectral Images
- Data-Driven Motion Estimation With Spatial Adaptation
- 09e4150ca30325aab0000000.pdf
- Corner Point 1
- Gu 3212401245
- Z Transform
- Image Sensor Architecture
- 4. IJECE - Block-Level Secure Steganography - Poonam
- stego
- SKIN TONE DETECTION USING DWT
- 2008
- Cmos Image Sensors
- [Tutorial] CMOS Image Sensors.pdf
- Veda Based Car License Plate Recognition
- Download Asset
- Gr Gb Difference in 3M CMOS Image Sensor
- ProductBrief_IMX135
- Localized Based