Error Resilient Image Transmission over Wireless Fading ... - CiteSeerX

M. Padmaja et. al. / International Journal of Engineering Science and Technology Vol. 2(5), 2010, 1242-1249

Error Resilient Image Transmission over Wireless Fading Channels M Padmaja [1] [1]

Assistant Professor, Dept of ECE, V R Siddhartha Engineering College, Vijayawada [2]

[3]

M Kondaiah [2] K Sri Rama Krishna [3]

PG Student, Dept of ECE, V R Siddhartha Engineering College, Vijayawada

Professor and Head, Dept of ECE, V R Siddhartha Engineering College, Vijayawada

Abstract: Nowadays transferring images and video over wireless channels is becoming more use of the medium. However, a wireless medium is not very reliable in the way that it adds unwanted components and noise to the wireless transmission. So there may be a loss of data. Due to this possible loss of data without the capability of resending a correct version we need a system to protect and correct such losses. The transmission of images over wireless channels is examined using reorganization of the compressed images into error-resilient, product-coded streams. The product-code consists of Turbo-codes or Reed–Solomon codes which are optimized using an iterative process. The wireless channel used for the testing phase is a Rayleigh Fading channel with Additive White Gaussian Noise (AWGN) added as a noise component. In this paper we propose protection techniques for image transmission and compare the performance of various protection methods for all JPEG standards. The proposed image standard model was found to perform very well in protecting the images against quality degradation during transmission over wireless channels. The strength of the protection plays a large part in the protection of the image and should be chosen to suit the particular channel in use. The default protection also provides very strong protection for the user who does not wish to choose their own setting. Keywords: JPEG, Source coding, Channel Coding, Additive White Gaussian Noise (AWGN). 1. INTRODUCTION

Wireless image transmission has been a very demanding feature in recent multimedia communications. However, a wireless environment is very prone to the fading phenomenon that hears a high error rate channel. The specific application of image transmission over wireless channels has deservedly attracted much attention since it requires not only careful design of the coding methodology for the compression of images, but also appropriate selection of the set of channel codes for effective forward error- correction. A variety of error resilient techniques [I, 2] employing product codes based on RCPC/CRC and RS codes for channel protection of other [3] streams have been recently proposed in the literature. In [41, a scheme based on Turbo-codes was presented which outperformed the method in [2] for image transmission over wireless channels. In [5] a real-time optimization algorithm was presented for the transmission of independently decodable packet streams over varying channels. The system utilizes the packetization scheme of [6]. In [7] the system of [5] was improved by combining the product code scheme of [1]. The methodology in [8] takes into consideration the dependencies between information in the compressed stream in order to cluster dependent layers and protect them according to their importance. The scheme proposed in the present paper is based on the image source coder, which generates error-resilient streams. The source coder is used in Conjunction with the application of a product code consisting of Turbo codes and Reed-Solomon codes. Due to the systematic form of Turbo codes, the immediate extraction and decoding of source information from the channel-coded stream is possible in case the stream is not corrupted. Whenever the stream is corrupted, the product codes will correct several errors. Uncorrectable errors are localized and the corrupted portion of the stream is discarded. The resulting error resilient transmission system is evaluated and is shown to outperform the best-performing known schemes for the transmission of images over wireless channels. The proposed image standard was created for this explicit purpose, protecting the important portions of the code stream whilst also providing a means of recovering lost data due to transmission errors. Modern image coding schemes have been conceived with a full scalable approach during the standardization process: it has been shown that this approach may be performed in an embedded Way, that is, the scalability is inherent to the image encoding

ISSN: 0975-5462

1242

M. Padmaja et. al. / International Journal of Engineering Science and Technology Vol. 2(5), 2010, 1242-1249 technique, and the final image improves with respect to PSNR or resolution as the codestream data bits reach the receiver side. 2. PROPOSED PROTECTION TECHNIQUE FOR THE JPWL STANDARD The entire process will follow a main flow as follows: Image encoder

Error Protection encoder Error prone Wireless channel

Image decoder

Error Protection decoder

Fig.1: Block diagram of Encoder and Decoder

2.1 Image Encoder Image encoder encodes the image into block coding of wavelet coefficients. The final bitstream is composed by a succession of layers which include information from independent codeblocks since their decoding does not require prior decoding of other code blocks.

1 3

Subband 1

2 4 N‐2

Subband 2

N‐1 N

Subband N

Fig.2: Each subband forms a group of layers that can he independently protected and decoded.

One of our primary goals during the design and implementation of the system proposed in the present paper was the transmission of information in such a way so that the corrupted portion of information can be discarded without affecting the decodability of the rest of the information. For this reason, we propose the division of the wavelet coefficients to be transmitted into N disjoint sets Jn, n=1,. . . ., N. in the wavelet domain so that

 Jn  n

and

 Jn =W n

Where W is the set including all the coefficients of the wavelet representation. If the disjoint sets of coefficients are channel-coded appropriately into channel packets. Then the erasure of a packet during transmission would not prevent the uncorrupted information from being decoded. Although numerous combinations of disjoint sets of coefficients can he conceived. In practice, since blockwise coding is performed. the subbands of the wavelet

ISSN: 0975-5462

1243

M. Padmaja et. al. / International Journal of Engineering Science and Technology Vol. 2(5), 2010, 1242-1249 decomposition were chosen in the present paper as a reasonable compromise (see Fig.1 ) between coding efficiency and information decoupling. 2.2 Error Protection encoder and Decoder The error protection encoder and decoder consist in techniques to protect a codestream against transmission errors as shown in below

Info

CRC

Parity1

Parity2

Fig.3: Arrangement of data in Channel-coded row

CRC

Turbo Coder

CRC

Turbo Coder

FEC FEC Subband 2

CRC

Turbo Coder

FEC FEC FEC Subband 3

CRC

Turbo Coder

: : : :

CRC

Turbo Coder

FEC FEC FEC FEC Subband N

CRC

Turbo Coder

Subband 1 FEC Subband 2

Fig.4; Product code based on Turbo codes and RS codes

The residual errors descriptor specifies the locations of residual errors in the codestream. The residual errors are the errors which cannot be corrected by the error protection decoder. This information is typically generated during the error protection decoding process. This information can subsequently be used in the image decoder to prevent decoding corrupted portions of the stream. 3. CHANNEL RATE ALLOCATION Since the bitstreams that are generated by image consist of layers with Unequal importance, UEP should generally be applied for their efficient protection from channel errors. The proposed UEP algorithm takes into account the importance of each packet and allocates more channel symbols (Turbo code bytes and Reed- Solomon symbols) to packets cawing important information and fewer to other packets, in this way, packets that contribute with higher distortion improvement to the eventual image quality are better protected than the rest. The problem formulated as above can be solved optimally under a specific target rate constraint by

Packet 1

s 1 c 1

Packet 2

r 2 s 2 c2

r 3 s 3 c 3 ISSN: 0975-5462

1244


Packet 3

:

:

:

Packet N r N s N c N Fig.5. Location of RS bytes. Source bytes and Turbo code bytes

Horizontal and vertical direction, the beginning of the first source byte in a packet is placed immediately after the last RS symbol On the other hand; the Turbo-code stream begins after the last source symbol. Specifically. We assume that in the nth, row of the product code array. There



D

N



D n

(1)

n  1

are rn RS symbols, sn source bytes and cn Turbo-code bytes. The resulting product code array is schematically shown in Fig. 2. The transmission of each packet in Fig. 2 stimulates a reduction in the average (expected) distortion of the image reconstructed after transmission. Since transmitted packets are independent of each other, the eventual distortion reduction D is the cumulative sum of the reductions achieved by the transmission of each packet separately, i.e. Where Dn is the average distortion reduction caused by the transmission of the nth packet. Our intention is to determine the optimal rn, sn, cn for n = 1,. . . , N by maximization of the average distortion reduction D subject to the constraint

rn  sn  cn  Rp  sn  Rp  rn  cn for n 1,2....N Where Rp 

(2)

Rs  c In order to simplify our optimization task, we make two assumptions regarding the allocation of N

channel rate. The expected distortion depends on the number of packets that are erased during transmission. In practice, an erasure occurs when the Turbo decoder is unable to recover the information in a corrupted packet. The average distortion reduction caused by the transmission of the nth packet is given by

Dn  (1  pn(cn)).Dn( sn)  Pn(cn).Dn

(3)

Then the average distortion reduction in the nth packet is corrupted is N

D n   Pe ( x ).Dn ( r x 1

N  x  1

 rn ) (4)

To gain insight regarding the term Dn (rN  x 1  rn ) assume that x packets are erased (see Fig. 6) during transmission. This means that only the source symbols in product code columns in which there are at least x RS symbols can be recovered. Since the level of RS protection is monotonically non-decreasing with n, the portion of the stream that can be recovered is determined by the end of the RS stream in the N -x + 1 packet. Thus, in every column on the left of the axis in Fig. 6. there are at least x packets carrying RS symbols which guarantee the recovery of the erased information. The probability Pe(x) that exactly x packets, out of N packets in total. are erased depends on the number t j, j = 1,. . . , Q, of packets protected at the jth protection level, where Q is the number of available Turbo protection levels.

ISSN: 0975-5462

1245


pe ( x ) 

u1

u Q 1

u2

  ...... 

x1  0 x 2  0

p t1 ( x1). Pt 2 ( x 2 )..

x Q 1  0

..... PtQ 1 ( x Q  1 ). P x Q  x 

Q 1

x l 1

(5) l

Using (2). (3). and (4). and taking into consideration the fact that Dn (rN  x 1  rn ) is zero.

| Packet 1 Packet 2

: |

:

| |

Packet N Correctable bitstream Erased Packet

| Uncorrectable bitstream Undecodable Source

Turbo Protection

Symbols

Symbols

Correctly received

Decodable Source

Packet

Symbols

Reed Solomon Symbols

Fig.6: When x packets are erased. the correctable portion of the bitstream lies on the left of the axis defined by the end of the RS stream in the N - x + 1 packet (since symbols in this stream are protected by at least x RS symbols). Information symbols that lie on the right of the axis are decodable only if they are part of uncorrupted packets.

N+1-n, then eq(1) will be

N

D { (1 pn(cn)).Dn(Rp rn cn)  n1

N n

pn(cn). ( pe( x).Dn(rN  1  x  rn))} x 1

(5)

Our intention is to maximize the distortion reduction given by (5). For the efficient solution of the optimization problem, a two-stage procedure is followed first the RS code is kept constant and the Turbo-code stream is

ISSN: 0975-5462

1246

M. Padmaja et. al. / International Journal of Engineering Science and Technology Vol. 2(5), 2010, 1242-1249 optimized. Subsequently. The Turbo rate-allocation determined in the previous step is kept constant and the RS stream is optimized. The above procedure is repeated several times until convergence. In particular, the Turbo optimization problem is treated by exhaustive search among the allowable combinations. Since the RS rate is monotonically nonincreasing with n, the RS rate-allocation problem is tackled using a fast, packet-wise bisection procedure which calculates. for the nth packet, the optimal length of the RS stream in the allowable range [ rn  1, rn  1] . It should be noted that since the appropriate amounts of RS and Turbo protection are determined using a two-step process, and not jointly, the above procedure does not guarantee global optimization. In practice, however, the proposed allocation algorithm yields very satisfactory results. 4. QUALITY TESTING

The PSNR is defined at (Peak signal-to-noise ratio) as:

Maxi 2 PSNR(dB)  10log10 MSE MAXi is the maximum value a pixel can take in the image. A standard grayscale will have an 8 bit resolution with a maximum value of 255. The MSE is defined as:

1 m1 n1 MSE   I (i, j)  K(i, j) mn i 0 j 0

2

It can be seen from the equations that a larger PSNR value indicates a better image quality. 5. RESULTS AND DISCUSSION The packet protection of the proposed standard will provide us with the best indication of the performance of the standard. In the event that the header fails to be properly protected, the image will fail to decode and as such the packet protection be completely irrelevant. The first set of results that will be investigated are those which left the data sensitivity and addressing mode as the defaults for the Open JPEG codec. 5.1 Table: For Lena Image

SNR

ISSN: 0975-5462

PSNR (JPEG)

PSNR (Proposed)

9

0

0

12

0

0

15

0

14.616

18 21

0 7.108

13.64 17.387

24

14.543

17.78

27

13.534

24.682

30

13.018

28.666

33

17.358

31.582

36

33.981

37.92

39

67.676

43.167

42 45

71.504 Inf

49.857 51.066

1247


(a) (b)

Fig.7 (a) Input lena Image (b) JPEG Protected Image

(c)

and (c) Proposed Protected Image

JPEG protection of lena image SNR Vs PSNR(dB)

proposed protection of lena image SNR Vs PSNR(dB)

80

60 JPEG protection

Lena-proposed protection

70 50 60 40 P S N R (dB )

P S NR(dB )

50 40

30

30 20 20 10 10 0

5

10

15

20

25 SNR(dB)

30

35

40

45

0

5

10

15

20

25 SNR(dB)

(a)

30

35

40

45

(b)

Fig.8: (a) performance of JPEG protection with respect to SNR vs PSNR in dB’s and (b) performance of Proposed Protection with respect to SNR vs PSNR in dB’s

ISSN: 0975-5462

1248

M. Padmaja et. al. / International Journal of Engineering Science and Technology Vol. 2(5), 2010, 1242-1249 Simulation of BER/SER for QPSK with Gray coding( Rayleigh multipath and AWGN) 0 10 BER-simulated SER-simulated

Rayleigh Fading Envelope(variance=0.5) 10 5

-1

0 Amplitude/RM S(dB)

B E R /S E R

10

-2

10

-3

10

-5 -10 -15 -20 -25

-4

10

0

5

10

15 20 SNR=Eb/No(dB)

(a)

25

30

35

-30

20

40

60

80

100 120 symbols

140

160

180

200

(b)

Fig.9: (a) simulation for QPSK in Rayleigh Fading Channel and (d) Its Fading Envelope performance

6. CONCLUSIONS A novel method was proposed for the communication of images over wireless channels. The proposed protection standard model performed very well at protecting the images codestream from a large array of error levels. It would be useful for anyone wanting to implement wireless stream of images. It also exploits the block-based structure of the image streams and employs product codes consisting of Turbo codes and erasure-correction codes in order to deal effectively with burst errors. A framework for the optimal unequal error protection was also proposed. Performance evaluation showed the superiority of the proposed scheme in comparison to well-known wireless transmission schemes. REFERENCES: [1] [2] [3] [4] [5] [6] [7] [8]

G. Sherwood and K. Zeger, “Error Protection for Progressive h a w Transmission Over Memoryless and Fading Channel? IEEE Trans. Communications, vol. 46, no. 12, pp.1555-1559, Dec. 1998. D.G. Sachs. A. Raghavan, and K. Ramchandran. ”Wireless image transmission using multiple-description based concatenated codes,” in Dora Compression Conference, 2000. A. Said and W. A. Pearlman,”A New Fast and Efficient Image Codec Based on Set Partitioning in Hierarchical Trees;’ IEEE Trans. Circuits Syst. Video TEchnol., vol. 6, pp. 243- 250, June 1996. N. Thomos, N. V. Boulgouris, and M. G. Strintzis. “Wireless Image Transmission Using Turbo Codes and Optimal Unequal Error Protection:’ in Proc. IEEE Int. Conference on Image Processing, Barcelona. Spain. Sep. 2003. V. Stankovic, R. Hamzaoui, and Z. Xiong, “Fast forward error protection algorithms for transmission of packetized multimedia bitstreams over varying channels:’ in Proc. IEEE In: Conference on Communication, Anchorage, AK, May 2003. X. Wu, S. Cheng, and Z. Xiong, “on packetization of embedded multimedia bitstreams.’ IEEE Trans. Multimedia, vol. 3, no. 1, pp. 132-140, Mar. 2001. V. Stankovic, R. Hamzaoui, and Z. Xiong, “Product code error protection of packetized multimedia bitstreams:’ in on Image Processing Proc. IEEE In/. Conference, Barcelona, Spain. Sep. 2003. J. Thie and D. Tauhman, “optimal erasure protection assignment for scalable compressed data with small packets and short channel codewords:’ EURASIP Journal on Applied Signal Processing, Feb. 2004.

ISSN: 0975-5462

1249