Supplementary loss concealment technique for imagetransmission through data hiding

Kyung-Su Kim & Hae-Yeoun Lee & Heung-Kyu Lee

Published online: 17 March 2009# Springer Science + Business Media, LLC 2009

Abstract This paper addresses a loss concealment technique of image transmission in errorprone environments. Imperfect transmission of imagery results in the loss of lines or blocks.Previous concealment techniques utilize the spatial, spectral, or temporal redundancy of theimagery. We propose a supplementary loss concealment technique that uses a data-hidingmethod. We divide an image into 8×8 pixels blocks and generate block descriptioninformation of each block. The block description information is inserted into other blocks inthe same image through LSB-based data hiding. Missing lines or blocks duringtransmission are restored by extracting this block description information. The techniqueto resist against lossy compression is also considered. We have performed a simulation toshow the outstanding performance of the proposed technique in comparison with other lossconcealment methods.

Keywords Loss concealment . Data hiding . JPEG compression . Image transmission

1 Introduction

Invaluable images such as satellite imagery and medical imagery are lossless compressed ornot compressed for transmission. However, imperfect transmission of the image in errorprone environments may result in the loss of lines or blocks. In order to minimize qualitydegradation, loss concealment techniques have to be applied at the receiving side.

Loss concealment means that an error has occurred and has been detected, but the erroris too great to correct. Therefore, it tries to interpolate what missing lines or blocks shouldhave been from surrounding pixels and hence to minimize the effect of the damage.

Multimed Tools Appl (2009) 44:1–16DOI 10.1007/s11042-009-0265-0

K.-S. Kim :H.-K. LeeDepartment of Electrical Engineering and Computer Science, Korea Advanced Institute of Science andTechnology, Guseongdong, Yuseonggu, Daejeon 305-701, South Korea

H.-Y. Lee (*)School of Computer and Software Engineering, Kumoh National Institute of Technology, 1,Yanghodong, Gumi, Gyeongbuk, South Koreae-mail: haeyeoun.lee@kumoh.ac.kr

Forward error correction (FEC) coding is a loss concealment technique, which is theprocess of adding redundancy to a bit stream according to some rules known to both thetransmitter and the receiver. However, it suffers from recovering the losses of continuouslines or blocks. Also, since it causes consequent delay and increases in the transmissionbandwidth, it cannot be applied to satellite imagery transmission with the limited bandwidthand high error possibility.

Another approach to conceal the loss is post-processing at the receiver. It can improvethe transmission efficiency and no extra channel needs. Previous concealment techniquesutilized the spatial, spectral, or temporal redundancy of the received image [15]. Hybrid ordynamic switching of spatial and temporal loss concealment techniques is also possible [6,8, 10]. These techniques estimate the missing lines or blocks of the image in a heuristicmanner by assuming smoothness or continuity of pixel values.

1.1 Spatial or spectral loss concealment

Spatial or spectral loss concealment assumes that natural images are smooth and their pixelvalues are continuous on the spatial or spectral domain. Missing lines or blocks arereconstructed by interpolating the neighboring pixels [5, 16]. Just simple spatialinterpolation techniques often suffer from the blurring effect in the edges of the image.Several techniques considering the edges of the image have been reported to solve theproblem [18, 20]. A small neighborhood of correctly received pixels to find the directionsof edges passing through the missing block was utilized [13, 14]. Interpolation can beapplied not only to the spatial domain but also to the spectral domain such as the discretecosine transform (DCT) domain [3].

1.2 Temporal loss concealment

Temporal loss concealment assumes that video contents are smooth and continuous in time.In order to conceal the loss of image (or frame) lines or blocks, the missing lines or blocksof the current frame are replaced with those of the previous frame [11, 19]. Some enhancedthe traditional temporal error concealment methods. In [12], they exploited the correlationof the motion vector (MV) to recover the lost block. An enhanced edge-sensitive processingorder with a new MV searching algorithm was proposed [17]. These methods improved theperceptual quality compared to previous temporal based algorithms.

Data hiding techniques hide useful information directly into an original media itselfwithout any perceptual distortion. In aspect of transmission, it provides the same bit rate asthe media to be transmitted. Jayalakshmi et al. designed a wavelet-based algorithm, wherethe approximate band was embedded in two selected high-frequency sub-bands of theimage [7]. Another loss concealment technique was proposed by Gur and Altug [4]. Theyalso used discrete wavelet transform (DWT) and embedded downsized replicas of originalimages into wavelet sub-bands for erroneous channels. Main disadvantages are thatmodification of DWT coefficients affects the whole quality of a data-hidden image and thescaling factor for embedding information should be adjusted by the characteristics of agiven image. Also, embedding capacity without visual degradation is limited so that theloss concealment performance is low. In [2], they extracted a low resolution version of agiven image through DWT and embedded it in DCT domain. Some techniques combined adata hiding with spatial redundancy [1] or with genetic algorithm [9].

In this paper, we propose a supplementary loss concealment technique of imagetransmission that uses LSB-based data hiding. The presented method is not dependent on

2 Multimed Tools Appl (2009) 44:1–16

the characteristics of a given image and has high embedding capacity for the high lossconcealment performance. We first divide an image into 8×8 pixels blocks and generateBlock Description Information (BDI) of each block. The BDI of one block is then insertedinto the LSB plane of the other block in the same image. Loss of image lines or blocksduring transmission is restored by extracting this BDI that has been inserted into otherblocks before transmission. Our technique is supplementary because the inserted BDI of thebroken block has a possibility to be destroyed after transmission. In such cases, some well-known concealment technique should be combined. Our research focused on applicationswith lossless compression, but has extended the technique to consider lossy compression atlow compression ratios.

This paper is organized as follows. In Section 2, we describe the way to generate BDI.Section 3 proposes a supplementary loss concealment technique through LSB-based datahiding. Section 4 presents a loss concealment technique that resists to lossy compression.Simulation results are shown in Section 5. Section 6 concludes.

2 Block description information

We explain how to generate block description information. Block Description Information(BDI) is calculated from each block of the image and exactly describes the contents of eachblock. If we have this BDI in case of the loss of the image, the original image can besimilarly reconstructed.

The BDI should be inserted into the image itself. Therefore, the inserted BDI should notbe visible to the naked eye. It means that the size, called the capacity, of the BDI that can beinserted into the image is limited. For the optimum recovery of the loss of image lines orblocks, BDI with a few bits should be able to describe the image block as accurately aspossible. Thus, the use of compression techniques is inevitable to minimize the size of theBDI.

To generate BDI, we apply well-known JPEG compression based on Discrete CosineTransform (DCT) and quantization1. For each 8×8 pixels block of the image, we convertthe spatial domain into the frequency domain by applying DCT, which remove theredundancy of image pixels on the spatial domain. We quantize DCT coefficients using aquantization table shown in Fig. 1. Each quantized coefficient is represented by 8 bits. DCcomponent has a value ranging from 0 to 255. AC components have a value ranging from−128 to 128.

In order to minimize the affection of the original image by BDI insertion, we set limits tothe size of the BDI at 64 bits length. Increasing the size of the BDI can more accuratelyreconstruct the loss of image lines or blocks during transmission. However, the insertedBDI of the image will be perceptible.

Since each quantized coefficient is 8 bits and the size of the BDI is 64 bits, eightquantized coefficients are selected in zigzag order as shown in Fig. 2 and used as the BDIof each block in our concealment technique. By quantizing coefficients following DCT,most of nonzero coefficients come together around the DC component. Thus, we are able toreconstruct the image that is similar to the original image using only BDI.

1 JPEG2000 compression based on Discrete Wavelet Transform (DWT) can be applied for our lossconcealment technique.

Multimed Tools Appl (2009) 44:1–16 3

3 Loss concealment technique with LSB-based data hiding

This section describes our supplementary loss concealment technique through LSB-baseddata hiding. We first explain inserting BDI into the image and then describe extracting theinserted BDI and restoring the loss of missing lines or blocks. Figure 3 illustrates how theproposed technique actually works using the BDI.

3.1 Insertion of BDI

The first step for inserting BDI is to generate BDI for each block. The BDI is then insertedinto the LSB plane of other blocks of the same image. Our BDI insertion process isdepicted in Fig. 4.

1 2

3

4

5

9

6

8

13

7

14

15

10 12

11

Fig. 2 Eight selected coefficientsin the DCT block

16 11

12

14

12

13

10

14

16

16

19

24

24

26

40

40

58

57

51

60

69

61

55

56

14 17

18

24

22

35

22

37

55

29

56

64

51

68

81

87

109

104

80

103

113

62

77

92

49

72

64

92

78

95

87

98

103

112

121

110

120

103

101

99

Fig. 1 Quantization table ofDCT coefficients

4 Multimed Tools Appl (2009) 44:1–16

(Step A, B, and C) First, we divide an image into 8×8 pixels blocks as shown in Fig. 4.For each block, we apply quantization following DCT and select eight quantizedcoefficients in zigzag order as explained in Section 2. These eight quantized coefficientsare our 64 bits BDI.

(Step D) We insert the BDI of each block into the LSB plane of the other block, calledthe partner block, in the same image by a LSB-based data hiding technique. For example,the BDI of the block a1 is inserted into the LSB plane of the block c1. The size of BDI is

(a) Flowchart of BDI generation and embedding.

(b) Flowchart of BDI extraction and recovery.

Fig. 3 Block diagram of the proposed loss concealment technique. a Flowchart of BDI generation andembedding. b Flowchart of BDI extraction and recovery

Fig. 4 BDI insertion process

Multimed Tools Appl (2009) 44:1–16 5

64 bits and the size of the LSB plane of a 8×8 pixels block is also 64 bits. Therefore, byreplacing the LSB plane of the 8×8 pixels block with the BDI, we are able to insert the BDIinto the block of the image.

Because adjacent blocks have high probability to be broken at the same time, we do notinsert the BDI of each block into the LSB plane of the adjacent block. For example, we donot insert the BDI of the block a1 into the LSB plane of the block b1. When several lines orblocks are lost in the boundary of the block a1 and b1, it is difficult to restore the loss of theblock a1, because the BDI of the block a1 that has been inserted into the block b1 is alsobroken. Of course, when two blocks (a1 and c1) where a block c1 contains the BDI of theother block a1 is broken at the same time, it is impossible to restore the loss of the block a1.In such cases, we apply a spatial interpolation technique to restore the loss of the block a1.In the experiment, the distance between the block a1 and the block c1 is a quarter of theimage height away.

In aspect of the image quality degradation, the LSB-based data hiding techniquemodifies only the LSB plane of each block and that changes its pixel value by 1.Consequently, the LSB-based data hiding technique achieves a high perceptual quality.Signal processing applications measure PSNR (Peak Signal to Noise Ratio) values torepresent the quantitative quality of the image as follows:

PSNR ¼ 10 log102552

1MN

PM�1

i¼0

PN�1

j¼0O i; jð Þ �W i; jð Þk k2�

ð1Þ

where O(i,j) and W(i,j) are the original image and the BDI inserted image. M and N are asize of the image.

Let assume that BDI has 0 or 1 value selected randomly from Gaussian distribution. Theabsolute difference between the LSB value of the block and the BDI will be 0 or 1, whoseprobability is 0.5. Thus, we can theoretically calculate PSNR value of 51.14 dB after LSB-based data hiding.

In our experiments, we test 100 randomly selected images from Internet and calculatePSNR values after inserting the BDI into the LSB plane of the image. Based on the results,we achieve the average PSNR value of 51.12 dB.

3.2 Extraction of BDI and loss concealment

When the loss of image lines or blocks is founded during transmission, the first step is toextract the BDI of the damaged block that has been inserted into the LSB plane of the otherblock, called the partner block, in the same image. We then reconstruct the image andrestore the loss. Our BDI extraction and loss concealment process is depicted in Fig. 5.

(Step A) First, we divide an image into 8×8 pixels blocks as shown in Fig. 5. When theblock a1 including missing lines or blocks is founded, we identify the partner block c1which contains the BDI of the damaged block a1. By reading the LSB plane of the blockc1, we are able to extract the inserted 64 bits BDI.

(Step B, C, and D) The extracted 64 bits BDI is the 8 quantized coefficients of the8×8 pixels DCT block. To reconstruct the image, we allocate these 8 quantized coefficientsto the DCT block in zigzag order (see Fig. 2) where other coefficients are set as zero andapply inverse DCT following de-quantization. The reconstructed image is similar to theoriginal image. For example, the 64 bits BDI from block c1 is used to apply inverse DCT

6 Multimed Tools Appl (2009) 44:1–16

following de-quantization. The result from this BDI represents the image similar to theoriginal image of the block a1. After we reconstruct the image from the BDI, we restore theloss of image lines or blocks by replacing the damaged image pixels as the reconstructedimage pixels.

When two blocks c1 and a1 are broken simultaneously as explained in previous section,LSB-based data hiding fails to restore the loss of the block a1. In such cases, a well-knownspatial interpolation technique is used to restore the loss of the block a1. It is possible to usemore sophisticated interpolation techniques. This is the reason that our technique issupplementary. The presented method is to enhance the quality of the restored image ashigh as possible.

4 Loss concealment technique resisting to JPEG compression

The LSB data hiding based loss concealment method cannot be directly applied forapplications where the image should be compressed with JPEG or JPEG2000 because LSBinformation is easily destroyed by lossy compression. This section presents an alternativemethod resisting to JPEG compression. Since satellite images and medical images areextremely valuable, these images should be compressed with low compression ratios (i.e.,JPEG quality factor is more than 85). The 64 bits BDI information of block a1 is embeddedin all frequency DCT coefficients of block c1. The BDI insertion method proceeds as follows:

(Step A and B) First, all DCT coefficients, C[1], C[2], … , C[64], are selected anddivided by its quantization factor q[i] in the quantization table of Fig. 1 as follows:

C0 i½ � ¼ roundC i½ �aq i½ �

� �for 1 � i � 64

where α is a strength parameter.(Step C) The binary sequence of 64 bits BDI is inserted by replacing the LSB plane of

each C0 i½ � with the BDI. It is based on the fact that the hidden BDI should survive againstthe quantization performed during compression using smaller quantization multiplier comparedto α.

(Step D) BDI hidden coefficients are obtained by the following rule:

Cw i½ � ¼ aq i½ �C¶¶ i½ � for 1 � i � 64

where C00 i½ � is the modified coefficient in Step C.

Fig. 5 BDI extraction and lossconcealment process

Multimed Tools Appl (2009) 44:1–16 7

The BDI extraction method simply performs the first two steps, A and B, of the insertionmethod with the same α parameter.

C¶¶ i½ � ¼ roundCw i½ �aq i½ �

� �for 1 � i � 64

Then, we check the LSB plane of all quantized coefficients C00 i½ � to extract the inserted64 bits BDI.

5 Simulation results

This section reports on the performance of the proposed loss concealment technique. Weused 100 randomly collected images from Internet that included such images as Lena,Baboon, Lake, and Bridge and video frames as Akiko and Foreman. After inserting the BDIinto the LSB plane of images, the average PSNR value between original images and BDIinserted images was 51.12 dB.

To overcome bursty losses, the partner block does not locate near the original block. Inour experiment, the BDI of each block is inserted into the partner block that is located in aquarter of image heights away. To find a partner block, a circular embedding technique isemployed where we embedded the BDI of block A into block B and the BDI of block B intoblock C, and the BDI of block C into block A. The distance among the block A, the blockB, and the block C is a quarter of the image height. For examples, since a 512×512 imageis divided into 64×64 blocks, the BDI of the (1, 1)th block is inserted into the LSB plane ofthe (17, 1)th block, the BDI of the (17, 1)th block is inserted into the LSB plane of the(33, 1)th block, the BDI of the (33, 1)th block is inserted into the LSB plane of the (49, 1)th

block, and the BDI of the (49, 1)th block is inserted into the LSB plane of the (1, 1)th block.In order to show the performance of the proposed loss concealment algorithm withoutspatial interpolation, we constrained that the partner block of a block including the lossshould survive similarly to [4]. When both blocks are lost at the same time, the spatialinterpolation method should be applied and that will disturb to show the performanceenhancement from the data-hiding technique only.

5.1 Quality of BDI

We analyzed the quality of BDI. From the BDI inserted image, we extracted the inserted64 bits BDI and reconstructed the image using only this 64 bits BDI. Some results areshown in Fig. 6. Although the perceptual quality of the image was degraded, we couldreconstruct the image that would be considerably similar to the original image. Only byusing BDI, we achieved the average PSNR value of 27.14 dB between the original imageand the reconstructed image.

5.2 Performance of the proposed loss concealment technique

The performance of the proposed loss concealment technique was tested in comparisonwith a spatial interpolation technique and a Jayalakshmi method [7]. The spatial interpolationtechnique restores missing lines or blocks of the image by considering adjacent and not lossneighboring pixels. To simulate the loss of image lines or blocks during transmission, werandomly removed several lines or blocks of the image depending on the loss rate whosepercentage was ranged from 5% to 50%.

8 Multimed Tools Appl (2009) 44:1–16

To assess the quality of the restored images, we measured the PSNR value betweenoriginal images and restored images against the loss of image lines or blocks. Simulationresults are shown in Tables 1 and 2. Figure 7a shows the average PSNR value for all testimages in a variety of the loss rate where X axis stands for the block loss rate (%) and Yaxis stands for PSNR values (dB) and the variance of PSNR values for each method isdepicted in Fig. 7b.

In most of the loss rates, the proposed technique explained in Section 3 outperformedboth the Jayalakshmi method and the spatial interpolation technique. The PSNR value ofthe proposed technique demonstrates an improvement of approximately 7 dB to both spatialinterpolation method and the Jayalakshmi method at all distortion levels and thus our lossconcealment technique achieves satisfactory performance in terms of the PSNR value andperceptual quality. We observed that the variance values of the presented method werehigher than those of other two methods (about 4 dB). The reason is because the presentedmethod is affected by the characteristics of a given image. The BDI is generated from 8 lowfrequency coefficients in DCT domain and hence the images containing a large amount oflow frequency components rather than the middle or high frequency components show the

a

a

b

b

Fig. 6 a Original image and b reconstructed image using only 64 bits BDI in Akiko and Foreman

Table 1 Average PSNR values between original and restored images

Loss rate Proposed Jayalakshmi et al. [7] Spatial interpolation

5% 36.23 dB 33.85 dB 29.38 dB

10% 33.36 dB 30.40 dB 26.39 dB

20% 30.49 dB 27.25 dB 23.43 dB

30% 28.71 dB 25.32 dB 21.55 dB

40% 27.52 dB 23.50 dB 20.21 dB

50% 26.58 dB 22.35 dB 18.95 dB

Multimed Tools Appl (2009) 44:1–16 9

Table 2 Minimum, maximum, and variance of PSNR values for each method

Lossrate

5% 10% 20% 30% 40% 50%

Spatial interpolation method Min 23.56 dB 21.11 dB 18.42 dB 16.69 dB 15.17 dB 13.92 dB

Max 35.30 dB 31.27 dB 28.66 dB 26.54 dB 25.18 dB 24.18 dB

Var 4.65 4.6 4.7 4.25 4.32 4.25

Jayalakshmi et al. method[7]

Min 27.57 dB 25.42 dB 22.24 dB 20.53 dB 19.72 dB 18.59 dB

Max 38.67 dB 35.22 dB 32.07 dB 30.14 dB 28.32 dB 27.17 dB

Var 3.96 4.18 4 3.47 3.3 2.71

Proposed method Min 30.65 dB 27.77 dB 25.16 dB 23.45 dB 22.23 dB 21.22 dB

Max 44.52 dB 40.14 dB 38.42 dB 35.86 dB 35.11 dB 34.62 dB

Var 8.16 8.56 8.64 8.23 8.39 8.51

5 10 15 20 25 30 35 40 45 500

2

4

6

8

10

12

Percentage of block lost (%)

Var

ianc

e

Spatial interpolation method

Jayalakshmi method [9]

Proposed method

0 10 20 30 40 505

10

15

20

25

30

35

40

Percentage of block lost (%)

PSN

R (

dB)

Erroneous

Spatial interpolation method

Jayalakshmi method [9]

Proposed method

Fig. 7 a Average PSNR valuesbetween original images and re-stored images for 100 test imageson a variety of the loss rate and bthe variances of PSNR values foreach method. The confidenceinterval on the estimated mean ofPSNR values is 99.99%

10 Multimed Tools Appl (2009) 44:1–16

high concealment performance. Moreover, although the spatial interpolation techniqueachieved appropriate PSNR values, interpolated lines or blocks could be easily noticeableby the native eye.

As explained in Section 4, we have extended the proposed technique to JPEGcompressed images. The performance at JPEG compression is shown in Fig. 8 where Xaxis stands for the JPEG quality factor ranged from 85 to 100 and Y axis stands for thePSNR value (dB). The control parameter α is set to 0.35. For 100 test images, the proposedtechnique manages to keep an enhancement gap from 1 dB to 4 dB when compared withthe spatial interpolation method. Since the Jayalakshmi method could not recover theoriginal block or line from an erroneous image in the lossy compression format, weexcluded the result. Although the enhancement is smaller than that from losslesscompressed images, it is observed that the hidden BDI was not destroyed in the compressedform and we could reconstruct the loss of blocks with the extracted BDI.

6 Discussions and conclusions

To show the performance of the proposed technique, we intentionally eliminated severallines near to a mouth in Akiko and eyes in Foreman (see Fig. 9b). Figure 9c shows restored

86 88 90 92 94 96 98

10

15

20

25

30

(a) 5% block lossJPEG Quality

PSN

R (

dB)

86 88 90 92 94 96 98

10

15

20

25

30

(b) 20% block lossJPEG Quality

PSN

R (

dB)

86 88 90 92 94 96 98

10

15

20

25

30

(c) 35% block lossJPEG Quality

PSN

R (

dB)

86 88 90 92 94 96 98

10

15

20

25

30

(d) 50% block lossJPEG Quality

PSN

R (

dB)

Erroneous Spatial interpolation method Proposed method

Fig. 8 Average PSNR values between original images and restored images for 100 test images on a varietyof JPEG quality factor when loss rate is a 5% loss, b 20% loss, c 35% loss, and d 50% loss, respectively

Multimed Tools Appl (2009) 44:1–16 11

images by the proposed technique and Fig. 9d shows restored images by the spatialinterpolation technique.

In the spatial interpolation technique, no information that there has been the mouth inAkiko and the eye in Foreman is contained in neighboring pixels. Consequently, it isimpossible to restore the missing mouth or eyes by any interpolation technique. Although

a a

b b

c c

d d

Fig. 9 a Original image, b attacked image, c restored image by the proposed technique, and d restoredimage by the spatial interpolation technique in Akiko and Foreman

12 Multimed Tools Appl (2009) 44:1–16

more sophisticated interpolation techniques can be considered, it will be in the samesituation that there is no information about missing parts.

In the proposed technique, the information about the missing mouth in Akiko or themissing eyes in Foreman exists in other blocks of the same image and hence we are able torestore successfully the missing mouth or eyes.

Since we calculate the BDI of all blocks in the image, the size of BDI is fixed to the sizeof the block (64 bits) and that set limits to the perceptual quality of the reconstructed image.If region of interest (ROI) area can be constrained, it is possible to increase the size of BDIin ROI area and the perceptual quality of the reconstructed image in ROI area will behigher.

This technique was studied to be used for a satellite system in our institute, whichtransmitted the large-size acquired image to the image receiving processing system (IRPS)in the ground through wireless channels. However, there were many problems: (1) becauseof large image size, the image can not be transmitted at a contact time between the satelliteand the IRPS. Usually, it takes a long time to download the acquired image in the memoryof the satellite (i.e., since the satellite rotates the earth, several contacts between the satelliteand the IRPS are required). (2) the IRPS can not communicate with the satellite directly,which only receive the signal from the satellite. The communication with the satellite isperformed through the Mission Control Station (MCS). To implement the FEC scheme inthe satellite system, hardware and software requirements for the satellite, the IRPS, and theMCS will be highly complex and difficult to implement. Also, since the FEC schemerequires the image to be kept in the limited memory of the satellite, the satellite can notacquire another image until the FEC scheme is completed. In this situation, the FEC schemewas not a solution to solve the loss concealment problem.

When the presented technique is applied to transmit the acquired image in the memoryof the satellite, complex requirements in hardware and software are not necessary. Also, thesatellite with the limited memory can be used to acquire another image. Similarly to thesatellite communication, there are various applications where the FEC scheme is difficult toimplement. In these applications, we believe that the presented technique will be useful.

Imperfect transmission of priceless imagery requires loss concealment techniques at thereceiving side to minimize the degradation of the image quality. Different from previousconcealment techniques utilizing the spatial, spectral, and temporal redundancy, weproposed a supplementary loss concealment technique through LSB-based data hiding.Also, the technique was extended to resist against lossy compression. Simulation resultsshowed the outstanding performance of the proposed technique in comparison with thespatial interpolation technique. We believe that the proposed technique has a great merit inapplications with lossless compression and lossy compression having high qualityfactors. Further works are developing a loss concealment technique applicable for videoapplications.

Acknowledgements This work was supported by the Korea Science and Engineering Foundation (KOSEF)grant funded by the Korea government (MEST) (No. ROA-2007-000-20023-0), and Defense AcquisitionProgram Administration and Agency for Defense Development under the contract.

References

1. Anhari AK, Sodagari S, Avanaki AN (2008) Hybrid error concealment in image communication usingdata hiding and spatial redundancy. Proceedings of the Int. Conf. on Telecommunications, pp 1–5

Multimed Tools Appl (2009) 44:1–16 13

2. Bashiri D, Aghagolzadeh A, Museviniya J, Nooshyar M (2008) A novel still image error concealmentusing fragile watermarking in wireless image transmission and packet-switched networks. Proceedings ofthe Int. Symp. on Telecommunications, pp 792–797

3. Chung YJ, Kim JW, Kuo C-CJ (1999) Real-time streaming video with adaptive bandwidth control andDCT-based loss concealment. IEEE Trans Circuits Syst II Analog Digit Signal Process 46:951–956.doi:10.1109/82.775393

4. Gur G, Altug Y (2007) Image error concealment using watermarking with subbands for wirelesschannels. IEEE Commun Lett 11:179–181. doi:10.1109/LCOMM.2007.061055

5. Hemami SS, Meng TH-Y (1995) Transform coded image reconstruction exploiting interblockcorrelation. IEEE Trans Image Process 4:1023–1027. doi:10.1109/83.392344

6. Hwang M-C, Kim J-H, Duong DT, Ko S-J (2008) Hybrid temporal error concealment methods forblock-based compressed video transmission. IEEE Trans Broadcast 54(2):198–207. doi:10.1109/TBC.2008.917274

7. Jayalakshmi M, Merchant SN, Desai UB, Ahay G, Aanchan JVL, Srinath P, Shashank J (2006) Errorconcealment using digital watermarking. Proceedings of the IEEE Asia Pacific Conf. on Circuits andSystems, pp 1713–1716

8. Jo M-H, Song W-J (2007) Hybrid error concealments based on block content. IET Image Process 1:141–148. doi:10.1049/iet-ipr:20060290

9. Kang L-W, Leou J-J (2006) Two error resilient coding schemes for wavelet-based image transmissionbased on data embedding and genetic algorithms. J Vis Commun Image Representation 17:1127–1144.doi:10.1016/j.jvcir.2006.08.003

10. Kung W-Y, Kim C-S, Jay Kuo C-C (2006) Spatial and temporal error concealment techniques for videotransmission over noisy channels. IEEE Trans Circuits Syst Video Technol 16(7):789–802. doi:10.1109/TCSVT.2006.877391

11. Lee S-H, Choi D-H, Hwang C-S (2001) Loss concealment using Affine transform for H.263 coded videotransmissions. Electron Lett 37:218–220. doi:10.1049/el:20010147

12. Ma Y, Cai A, Guo J (2006) A novel temporal error concealment based on statistical pixel estimationmodel. Proceedings of the Int. Conf. on Digital Telecommunications, pp 21–27

13. Navathe B, Merchant SN (2006) Directional error concealment technique for intra-coded videoframes. J Electron Imaging 15:1–4. doi:10.1117/1.2204453

14. Shi Y, Zhu X, Yin H (2008) A fast and efficient spatial error concealment for intra-coded frames. ProcCongress Image Signal Process 1:264–267. doi:10.1109/CISP.2008.45

15. Wang Y, Zhu Q-F (1998) Error control and concealment for video communication: a review. Proc IEEE86:974–997. doi:10.1109/5.664283

16. Wang Y, Zhu Q, Shaw L (1993) Coding and cell-loss recovery in DCT based packet video. IEEE TransCircuits Syst Video Technol 3:248–258. doi:10.1109/76.224235

17. Wu T-H, Wu G-L, Chen C-Y, Chien S-Y (2008) Enhanced temporal error concealment algorithm withedge-sensitive processing order. Proceedings of the IEEE Int. Symp. on Circuits and Systems, pp 3466–2469

18. Zeng W, Liu B (1999) Geometric structured based loss concealment with Nobel applications in block-based low-bit-rate coding. IEEE Trans Circuits Syst Video Technol 9:648–664. doi:10.1109/76.767129

19. Zhang J, Arnold JF, Frater MR (2000) A cell-loss concealment technique for MPEG-2 coded video.IEEE Trans Circuits Syst Video Technol 10:659–665. doi:10.1109/76.845011

20. Zhu W, Wang Y, Zhu Q-F (1998) Second-order derivative-based smoothness measure for lossconcealment. IEEE Trans Circuits Syst Video Technol 8:713–718. doi:10.1109/76.728413

14 Multimed Tools Appl (2009) 44:1–16

Kyung-Su Kim received the B.S. degree in Computer Engineering from Inha University, Republic of Korea,in 2005, and the M.S. degree in Computer Science from Korea Advanced Institute of Science andTechnology(KAIST), Republic of Korea, in 2007. He is currently working toward his Ph.D. degree inMultimedia Computing Lab., Dept. of EECS, KAIST. His research interests include image/videowatermarking and fingerprinting, error concealment method, information security, multimedia signalprocessing, and multimedia communications.

Hae-Yeoun Lee received his MS and PhD degrees in computer science from Korea Advanced Institute ofScience and Technology, Korea, in 1997 and 2006 respectively. From 1997 to 2001, he was with the SatelliteTechnology Research Center, KAIST. From 2001 to 2006, he was with Satrec initiative, Korea. From 2006 to2007, he was a post-doctoral researcher in Weill Medical College, Cornell University, United States. He isnow with Kumoh National Institute of Technology, Korea. His major interests are digital watermarking,image processing, remote sensing and digital rights management.

Multimed Tools Appl (2009) 44:1–16 15

Heung-Kyu Lee received a BS degree in electronics engineering from Seoul National University, Seoul,Korea, in 1978, and MS and PhD degrees in computer science from Korea Advanced Institute of Science andTechnology, Korea, in 1981 and 1984, respectively. Since 1986 he has been a professor in the Department ofComputer Science, KAIST. He has authored/coauthored over 100 international journal and conferencepapers. He has been a reviewer of many international journals, including Journal of Electronic Imaging,Real-Time Imaging, and IEEE Trans. on Circuits and Systems for Video Technology. His major interests aredigital watermarking, digital fingerprinting, and digital rights management.

16 Multimed Tools Appl (2009) 44:1–16