WPM P1.03
Error Concealment Algorithm Using Interested Direction
for
JPEG 2000 Image Transmission
‘Pei-JunLee,
*Mei-Juan Chenand
‘Liang-GeeChen, Fellow,
ZEEE
’
Departmentof
ElectricalEng.
and Graduate institute of Electronics Eng., National Taiwan University, Taiwan *Departmentof
Electrical
Eng., National DongHwa
University,Taiwan
Abstract-The newly defined JPEG 2000 delivers scalable and
region-of-interest image with lower hit rate In internet and wireless communications. However, any data lass in JPEG 2000 bitstream will affect the consequent bitplanes and even possibly
destroy the whole picture, We propose a new algorithm to recover the damaged bitplanes data according to the interested directional pattern in-subband. The simulation results show that the proposed concealment algorithm can achieve much smoother edge on the reconstructed image.
1. INTRODUCTION
In recent years, intemet and wireless communication have grown astronomically. However, it is known that un-reliable wireless channels and the congestion of internet may inject errors into the transmitted hit-stream. Since multimedia will have much wider application in the future, error resilience issue has become necessary for imageivideo transmission. JPEG2000 contains error resilience tools to detect and localize the errors, and also to resynchronize the decoding process. These tools can minimize the effect of errors on image quality. But standard error resilience tools cannot recover the lost data. Consequently, additional methods that can conceal the lost data are required during the decoding process. Various methods of concealing errors have been proposed. Zero is used to replace the missing wavelet coefficients for JPEG2000 standard [I ,2]. [3] propose an algorithm for interpolating wavelet coefficients from neighboring lost coefficients. However, this algorithm cannot he used for real time error concealment.
It is known that once an error occurs, the remainder of the embedded bit-stream is useless and the subsequent decoding steps will produce erroneous results such that the whole image is influenced. For solving this problem, replacing all missing wavelet coefficients by zeros may affect lots of significant non-zero coefficients such
that
some high frequency components are lost. [4] conceals the lost bitplane by using the concept of cross-suhband correlation. The method does not utilize low frequency suhhand (LL hand and the lowest level). Therefore, when upper low frequency suhhand is lost, there is still significant image deterioration.In
this paper, we propose an efficient algorithm to recover damaged wavelet coeficients, in which the interested direction sets of in-subband are used to estimate the position of significant nodes of damaged hitplane. By the algorithm, the reconstructed image also preserves much smoother edge.0-7803-7721-4/03 $17.00 0 2003 IEEE 182
11. THE PROPOSED CONCEALMENT ALGORITHM
In the relative literatures, there are many efficient wavelet coding techniques to detect significant nodes. [ 5 ] utilizes the interested direction sets of in-subhand to express the relationship among suhands. We use the interested direction sets of in-subhand to determine the lost hitplane data. The direction patterns are shown in Fig.1. For different suhbands, the interested directions are different.
(I) In HL subhands, we only consider the directions v-l to v-4, their 180-degree rotation, and reflection with respect to x
and y axes, respectively.
(It) We consider h-l to h-4 in LH subhands, their 180-
degree rotation, and reflection with respect to x and y axes, respectively.
(Ill) In HH subhands, the interested direction set is d-1, we should consider both v-l and h-l directions and their 180- degree rotation.
For each lost hit in a bitplane, the following algorithm is to show the recovering procedure for the lost bit.
Step 1: Make sure which subhand is the lost bit belonging to, and then consider the corresponding interested directions in Fig.1.
Step2: There are two cases as follows to be considered. Casel: In the upper bitplane, the same location of the lost bit is “0”. If its interested direction pattern is satisfied, the lost hit is set to
“I”.
Othewise, check the interested direction pattem with concealed bits in the current bitplane, if it is satisfied, then the lost bit is set to “I”, else set to“0”.
Case2: In the upper bitplane, the same location of the lost hit is “ I ” . We only need to check the interested direction
pattem with concealed hits
in
the current hitplane, if it is satisfied, then the lost bit is set to “l”, else set to “0”.v- 1 “-2 v-3 “-4 d- I
h-l h-2 h-3 h-4
: upper bitplane significant node
: current bitplane damaged node
When the compressed image bitstream comes into the decoder, every pass will he decoded. Then errors can he detected by the error resilience mechanism in JPEG2000.
If
errors are detected, the relevant bitplane data will he discarded. The above algorithm can recover the lost bitplane data. The recovered are collected for the coefficients of IDWT and for reconstructing the image. The proposed scheme is summarized as the flowchart in Fig.2.Bilrircam
A
DecodingI
every parr Reconstructed Discard hitplaneI
missing wavelet coefficients by zeros to recover damaged wavelet coefficients such that the recovered wavelet coefficients will he much similar with noise free data. The experiments show that we have the objective results with at least 2 dB improvement better than those without error resilient mechanism, and the subjective result is with much smoother edge on the reconstructed image by the proposed concealment algorithm.
. .
such as, resynchronization marker for each subband, segmentation marker for each bitplane, and termination at each coding pass. When the error occurs, in decoding with error detection mechanism, those missing wavelet coefficients are simply replaced by zeros; 3) The procedure is the same as case 2) in encoding process, except that, in decoding process, the proposed error concealment algorithm is adopted.The experimental results demonstrate that the proposed algorithm improves performance significantly
in
terms of subjective measurements compared with other methods as shown in Fig. 3. Fig.3(d) is the reconstructed image by our proposed algorithm, which has at least 2 dB improvement better than the other two methods Fig.3(h) and Fig.3(c). It also Illustrates that Fig.3(d) has much smoother edges than the other two figs in "mirror" and Lena's "shoulder" and "face", etc. The overall performance show that the proposed algorithm is much more efficient and robust for recovering the lost datain
JPEG 2000 bitstream information.IV. CONCLUSION
In this paper, we have proposed a new technique to improve the error resilience ability for JPEG2000. The proposed approach utilizes interested direction sets of in-suhband and undamaged hitplane information instead of replacing the
(c) (d)
Fig.3 The subjective results for "Lena" image by BER=lOe-3, bit-rate=lhpp in the lowest HL suhhand. (a) reconstructed image by error free, PSNR=36.21dB ; (b) error resilience only by resynchronization marker, PSNR=14.49dB (case I
+
error); (c) missing wavelet coefficients replaced by zeros, PSNR=27.66dB (case 2+
error); (d) reconstructed image by our proposed algorithm, PSNR=29.57dB (case 3+
error).REFERENCE
[ I ] I. Moccagutla, S. Soudagar, 1. Liang, and H . Chen, "Enor-Resilient coding in JPEGZOOO and MPEG-4," IEEE Journal on Seleeled Areas in Communicotionr, Vo1.18, No.6, pp. 899-914, June 2000.
C. Chrirtopoulos, A. Skodras, and T. Ebrahimi, "The JPEGZOOO Still
image coding system: An ovewiew," I E E E Trans. on Consumer
Elelronics, Vo1.46, No.4, pp. 1103-1 127, Nov. 2000.
L.Atzori, S. Corona, and D.D Giuto, "Error recovery in JPEGZOOO image transmission," I E E E lnlernalional Conference on Acouslics.
Speech. and SignolProcessing, Vo1.3, pp.1733-1736,ZOOl
P.J. Lee and L.G. Chen, "Bit-plane error recovery via cross rubband for image transmission in JPEGZWO", IEEE Inlernolionol Conference on
Mullimedia and Expo. Switzerland, Amgut 2002.
Jia Wang, Songyu Yu and Jun Sun, "Wavelet image coding based on directional dilation," SPIE Visual Communications and Image Processing, Vol. 4671,pp.1175-1184,2002. [Z] [3] [4] [5]