A Robust Anaglyph 3D Video Watermarking based on Multi-sprite

Generation

Dorra Dhaou

, Saoussen Ben Jabra

1,2

and Ezzeddine Zagrouba

Universit

e de Tunis El Manar, Institut Sup

erieur d’Informatique El Manar, LR16ES06 Laboratoire de Recherche en

Informatique, Mod

elisation et Traitement de l’Information et de la Connaissance (LIMTIC),

2 Rue Abou Raihane Bayrouni, 2080, l’Ariana, Tunisia

Universit

e de Sousse, Ecole Nationale d’Ing

enieurs de Sousse (ENISo), BP 264 Sousse Erriadh 4023, Tunisia

Keywords:

Anaglyph 3D Video, Watermarking, Multi-sprites, DWT, LSB, Invisibility, Robustness.

Abstract:

Collusion presents a malicious attack for video watermarking techniques. In the case of anaglyph 3D video,

this attack is not yet considered. In fact, only several watermarking techniques were proposed for this type

of media and they are not robust against dangerous attacks such as MPEG compression and collusion. In

this paper, a robust anaglyph 3D video watermarking technique is proposed. It is based on multi-sprites as

a target of insertion. This allows obtaining a robustness against collusion attacks. First, several sprites are

generated from original video. Then, a hybrid embedding scheme based on the least signiﬁcant bit and the

discrete wavelet transformation based method is applied on every sprite to insert signature. This improves

invisibility and robustness against usual attacks. Experimental results show a high level of invisibility and a

good robustness against collusion, compression and against additional attacks such as geometric and temporal

attacks.

1 INTRODUCTION

3D videos become more and more popular and

increase daily due to the 3D technology evolution

and the high speed of internet access. In fact, 3D

video is getting a huge attention from public recently

over 2D video. There are various display methods

to see videos in 3D like the anaglyphic technique,

the polarized light technique and the active shutter

technique. The last two methods require a speciﬁc

and expensive hardware. Indeed, an expensive

display with an alternate liquid crystal shutter glasses

which is used with an electronic device is needed for

the active shutter system while a polarized display

is required for the polarized system like a pair of a

costly polarized ﬁlter glasses and others displays to

obtain 3D videos. Finally, the anaglyphic system for

3D image which consists of two superimposed images

representing the same scene from different angles

where the left view in red and the right one in cyan

(green + blue). It is a printed image to be observed in

relief by using 3D glasses with two ﬁlters of different

colors (chromatically opposite) placed in front of each

eye of the observer. The visual cortex of the brain

fuses the two colored images into the perception of

a 3D scene. Anaglyph 3D system is the display

method that has taken the attention of researchers

lately in many domains because it is cheaper than

other systems and it is the easy way to make the

3D visual experience succeeded without any special

hardware, just only cyan-red glasses.

Due to the rapid evolution of the 3D technology,

the transmission of 3D video in the internet became

easier and everyone can access to their content.

However, 3D videos cannot be distributed without any

kind of protection from dangerous attacks. Digiatal

watermarking techniques are the most efﬁcient and

suitable solutions to protect 3D videos thanks to the

security and copyright protection provided to digital

data (text, audio, image, video). Indeed, it consists

of embedding a mark within a host 3D video, then of

trying to extract it after any attack applied on marked

3D video.

Different watermarking works dedicated for 2D

video have been proposed in the literature (Bayoudh

et al., 2017b; Kerbiche et al., 2017) in recent years but

concerning 3D videos, the ﬁeld is still immature due

to the variety of display ways and the 3D information

data complexity. Regarding 3D anaglyph videos,

some existing works were proposed and did not

resist to most important attacks such as compression

and collusion. This last attack tries to approximate

260

Dhaou, D., Ben Jabra, S. and Zagrouba, E.

A Robust Anaglyph 3D Video Watermarking based on Multi-sprite Generation.

DOI: 10.5220/0007930102600267

In Proceedings of the 16th International Joint Conference on e-Business and Telecommunications (ICETE 2019), pages 260-267

ISBN: 978-989-758-378-0

and eliminate the embedded mark to obtain the

original video. In order to resist this dangerous

attack, several techniques of 2D video watermarking

proposed using static mosaic or static multi-sprites as

target of signature embedding (Bayoudh et al., 2017a;

Bayoudh et al., 2015). In fact, the similar physical

points in all frames forming the video should be

marked similarly in order to withstand the collusion

attack. That is why, the best embedding target is the

mosaic image created from a video sequence.

In this paper, a robust anaglyph 3D video

watermarking technique is presented. It is based on

multi-sprite generation from every set of 25 frames

composing the original video. The signature will

be embedded in each sprite using a hybrid insertion

scheme based on Discrete Wavelet Transform (DWT)

and Least Signiﬁcant Bit (LSB) insertion methods.

This allows obtaining a high level of invisibility and

maximizing robustness against usual attacks.Besides,

the use of multi-sprites as an embedding target will

allows robustness against collusion and compression

attacks.

The rest of this paper is planned as follows:

section 2 deals with the related works of both

anaglyph 3D images and videos watermarking

techniques. Section 3 presents the proposed

technique for anaglyph 3D video based on DWT

and multi-sprites generation. Section 4 analyzes

the proposed scheme via the experimental results.

Section 5 presents a comparative study between the

existing works and the proposed method. Finally,

section 6 draws the conclusion and some perspectives.

2 ANAGLYPH 3D IMAGE AND

VIDEO WATERMARKING

Various watermarking techniques dedicated for

anaglyph 3D images are proposed in the literature and

they can be characterized by two criteria: the selected

insertion domain and the chosen channel to embed the

signature.

Regarding the ﬁrst criterion, the authors choose

to insert the mark into the coefﬁcients bits recovered

either after applying DWT (Zadokar et al., 2013; Devi

and Singh, 2016), Discrete Cosine Transformation

(DCT) (Munoz-Ramirez et al., 2015), or Fractional

Fourier Transformation (FrFT) (Y and Krishna,

2016). The most of anaglyph 3D watermarking

approaches are based on DWT domain to insert the

signature by adding the mark coefﬁcient bits of high

frequencies sub-bands to those of the chosen original

image view in order to ameliorate the invisibility and

the robustness compromise of the proposed technique

(Ruchika and Parth, 2015; Prathap and Anitha, 2014).

The existing techniques based on DWT domain

provide the highest invisibility level where the Peak

Signal to Noise Ratio (PSNR) value is up to 50 dB and

the best robustness level against usual attacks such

as geometric attacks compared with the DCT and the

FrFT based methods.

Concerning the second criterion, the mark is

inserted either into blue or red image (Ruchika

and Parth, 2015) or into all blue, red and depth

images composing the anaglyph image (Sanjay R. and

Ravindra B., 2015; Zadokar et al., 2013). The marked

blue or red image will be then recombined with

the unmarked host images to ﬁnd the ﬁnal marked

anaglyph image. In the majority of the existing

watermarking techniques, the writers insert the mark

into the blue images of anaglyph image due to the

insensitivity of the human eye to this color. This

insertion process gives a good visual quality level

and decreases the robustness against the majority of

attacks (Prathap and Anitha, 2014).

Concerning anaglyph 3D videos, few number of

existing works are proposed until now. In (Waleed

et al., 2013), a blind DWT based watermarking

technique is proposed. The mark is inserted in all

the blue images of all frames composing the host

anaglyph 3D video since the blue color change is

slightly perceived by the human visual system. The

problem of this technique is that when the mark is

embedded only in blue images, a lack of robustness

may be caused if the attack targets them then the

embedded mark will be easily lost. Whereas in 2015,

(Salih et al., 2015) an another blind watermarking

scheme based on DWT is suggested. The mark is

inserted into all blue images of all frames where

a scene change is detected in the anaglyph 3D

video to obtain a good invisibility level. In this

proposed technique, the embedding of mark just into

scene change frames detection can provoke a fragile

watermarking scheme in a video with only one or few

scenes. Moreover, in (Dhaou et al., 2018) a Groups of

Pictures (GOP) decomposition based watermarking is

proposed using the DWT domain where the mark is

embedded into the three images types (I, B and R)

of each set of GOP. In fact, the embedding method

targets cyan images and red images in B and R images

respectively. Whereas in I images, it targets all cyan,

red plus depth images. So, if any manipulation tries to

remove the mark embedded in any images type, it can

be then extracted from the other types. The proposed

scheme provides a high imperceptibility level and

robustness against usual manipulations and it can’t

resists to collusion attack.

A Robust Anaglyph 3D Video Watermarking based on Multi-sprite Generation

261

3 PROPOSED APPROACH

Based on the study of the art, all existing

watermarking techniques proposed for anaglyph

3D videos present a good invisibility level with

robustness against several usual attacks. However,

these proposed techniques did not consider malicious

attacks like MPEG compression and especially

collusion attack. To ﬁnd a good solution to

this problem, a new robust watermarking technique

dedicated for anaglyph 3D video is proposed in this

paper. This technique is based on multi-sprites as a

target of embedding. Multi-sprites are chosen thanks

to their high quality of reconstruction compared with

single mosaic.

The proposed method architecture is illustrated in

Fig. 1 and is composed into three main steps. In

fact, giving an original anaglyph 3D video, a sprite

will be generated from every set of 25 frames. Then,

signature will be embedded in each sprite. Finally,

marked 3D anaglyph video will be reconstructed

using the marked sprites.

3.1 Multi-sprite Generation

A sprite (mosaic image) is a panoramic view

containing all information spread in the video

sequence. It will characterize the evolution of the

video in a single high resolution image and a unique

physical point will represent each duplicated point

along the video. The mosaic generation composed of

three basic stages (Zeng et al., 2014). First, it consists

in detecting features from all the video frames to

select all interest points. The SIFT descriptor which is

invariant to certain transformations (rotation, scaling,

noises) is generated from all extracted interest points.

Second, all extracted feature points from images

are matched by calculating the distance between

them to compare the features descriptor of each

points. Next, the random sample consensus is

applied to remove the expected mismatching between

the corresponding images. Then, by applying

the appropriate transformation, either both the two

consecutive images, or the image and the active

sprite are aligned. Finally, the image is integrated

using a median or average intensity ﬁltering. Image

integration is done in order to create the sprite by

choosing the equivalent part of the transformed image

and linking it with the ﬁnal sprite.

A sprite can describe all the frames composing

the video in a unique image, but it can causes many

difﬁculties like an important time processing during

generation and a distortion effect in mosaic image.

In fact, if a video sequence has too much frames,

the recovered mosaic will has a large size. This

can causes ﬁrst complexity during the generation of

mosaic and it can be applied only if all frames are

available.

Multi-sprites are considered as a solution in order

to avoid the problems mentioned previously. In

fact, the multi-sprite generation consists in generating

multiple mosaic images from a video sequence in

the same scene rather than one mosaic image using

diverse parameters (Barhoumi et al., 2011) to select

the best frame number per sprite. They are generated

with the same way of single sprite but the only

difference between them is that the whole video

sequence is ﬁrst divided into sub-sequences basing on

a pre-determined parameter. Next, each sub-sequence

is manipulated as a whole video sequence and the

single sprite generation method is done to get the

wanted sprites. If one sprite is produced with too

much frames, the time processing will be increased

and the quality of the reconstructed sprite will degrade

the video quality.

In order to reduce these two problems, the video

will be simply divided into unit of a second segment.

So, for every second, the set of frames constituting it

is considered as a sub-sequence and a sprite will be

generated. In this case, the generation of one sprite

for each 25 frames per second is proposed. Fig. 2

presents sprites generated from the sequence video

”BigBuckBunny”, where each sprite is generated

from 25 frames.

Several watermarking techniques based on mosaic

images as an embedding target have been proposed

for 2D video contents (Koubaa et al., 2012). These

techniques are robust against malicious attacks such

as collusion. As a matter of fact, this last one

consists in estimating the inserted mark and then

eliminating it from marked videos in order to recover

original ones. There are two types of collusion

attacks: In the ﬁrst type, the attacker takes a number

of samples of the same video, which are marked

with different signatures, and essays to discover

similar frames. The unmarked video is obtained

by averaging the neighboring marked video frames,

which permits mixing the different signatures. In the

second type, the attacker takes various videos marked

with the same signature. To obtain unmarked videos,

the attacker has to estimate the inserted signature

by averaging the different marked videos, and then

eliminate it from all these videos.

Different 2D video watermarking techniques,

which withstand collusion attacks are proposed, but

regarding anaglyph 3D video there are no existing

technique robust against collusion. In fact, a

video watermarking based on the Spread Spectrum

SECRYPT 2019 - 16th International Conference on Security and Cryptography

262

Figure 1: The general ﬂowchart of the proposed approach.

Figure 2: Sprites generated from ”BigBuckBunny”.

(SS) technique was put forward in (Koubaa et al.,

2012), where the mark was embedded in the mosaic

image obtained from the whole video sequence using

a spatial domain, and which was robust against

MPEG4 compression, frame dropping and collusion

attacks. A video watermarking based on a dynamic

multi-sprite generation and SURF descriptor was

put forward in (Bayoudh et al., 2017a) to resist

collusion attack where a signature was inserted

into the different sprites. Another scheme robust

against collusion attacks was proposed in (Bayoudh

et al., 2015), where the mark was embedded in

mosaic images generated from each group of frames

of the host video using the LSB algorithm. In

(Kerbiche et al., 2018), a video watermarking based

on feature regions detected using the crowdsourcing

technique was presented. After the detection of

regions of interest, the mark was inserted utilizing

a multi-frequency domain in the generated mosaic

image. This approach was robust against most

manipulation and collusion attacks.

3.2 Signature Embedding and Detection

Methods

The signature embedding process is based on a hybrid

insertion using the DWT and LSB techniques where

the signature is embedded in each generated sprite

to enhance both the robustness and invisibility of

the proposed method. The signature embedding

technique is divided into different steps. First, based

on sprite size, the signature spreading is applied.

Then, the signature and each sprite are decomposed

using 3

level of wavelet transformation which is

obtained by decomposing the sub component LL

and critically sub sampled to LL

, HL

, LH

, and

. This process will be then applied for the sub

component LL

in order to obtain LL

, HL

, LH

and HH

. In fact, to obtain marked coefﬁcients,

the high frequency coefﬁcients (HH

) and (LH

) of

the signature are added to those of the sprite using

an invisibility factor. High frequency sub-bands are

chosen because they are the good areas for signature

embedding since the HVS is insensitive to these

high frequency components and they provide a good

tradeoff between capacity and invisibility. At the

second step, the low frequency sub-band (LL

) in

each sprite is going through an LSB embedding

technique where the signature is embedded in the

least signiﬁcant bits of each low frequency sub-band

(LL

) of sprite pixels which contains the total of

A Robust Anaglyph 3D Video Watermarking based on Multi-sprite Generation

263

image energy and information. Embedding the mark

in low frequency make the scheme robust against

compression attacks but it degrade the visual quality

of the marked media that is why we use to embed the

mark in the least signiﬁcant bit of (LL

) sub-band.

Indeed, the LSB technique is the well-known and

traditional method used to insert the mark in a host

media due to its simplicity (Sharma, 2012). It consists

in embedding the mark bits in the least signiﬁcant

bits i.e the ﬁrst bit of the original media pixels since

the insertion in the ﬁrst bit does not affect on the

visual quality of the marked media. The marked

sprites will be obtained by applying the 3-level of

Inverse Decomposition Wavelet Transform (I-DWT)

to the corresponding marked sub-bands. After all, the

marked sub-sequences will be reconstructed from the

obtained marked sprites and recombined in order to

generate the marked 3D anaglyph video.

The detection step consists of verifying the

presence of the mark in a given anaglyph 3D video

where the original generated sprites are required

during the detection. In fact, multi-sprites will be

generated from the marked anaglyph 3D video. Then,

the 3

level DWT is applied on each sprite and the

, HH

and LH

sub-bands are selected. The least

signiﬁcant bits are extracted from the LL

sub-band

coefﬁcients of each sprite to recover the bits of

embedded spread mark. In addition, the coefﬁcients

of the spread signature are extracted from the selected

high frequencies sub-bands HH

and LH

and the

I-DWT is applied to recover the mark. Finally, the

spread mark is divided into a set of marks and the one

which have the best correlation is chosen.

4 EXPERIMENTAL RESULTS

To evaluate the proposed approach, a set of

robustness and invisibility tests are applied on

ﬁve original anaglyph 3D videos, which have

different characteristics such as background texture,

movement, resolution and number of frames. In

fact, the ﬁrst video presents slow movement, textured

frames and it is composed of 650 frames with

a resolution of 1280 × 720 whereas the second

sequence has a rapid movement, textured frames and

is composed of 300 frames with the same resolution

as the ﬁrst video. The third video has a medium

movement, textured 250 frames and a resolution of

480 × 360. The fourth and the last videos present

a slow movement, textured frames and are composed

of 150 and 400 frames respectively with a resolution

of 1280 × 720 and 640 × 320 respectively. As a

signature, a binary image with size of 32 × 32 is used.

After signature embedding, no differences

between original (Fig. 3.a) and marked (Fig. 3.b)

anaglyph 3D videos can be observed.

Figure 3: a- Original frames. b- Marked frames.

In order to prove the high level of invisibility

of the proposed approach, the average of the PSNR

values is calculated between all original and marked

frames forming the video sequence. In fact, usually

a high level of invisibility is obtained with a PSNR

superior to 40 dB. The obtained PSNR values for

anaglyph 3D test videos are illustrated in ﬁgure 4.

They show that the proposed watermarking technique

provides a high invisibility with the embedded mark

in different test videos where the minimum PSNR

value is about 54 dB and its maximum is about 59

dB.

Figure 4: Invisibility evaluation using the PSNR values.

To evaluate the robustness of the suggested

scheme, various usual and malicious attacks are

experiment. First, these attacks are applied on the

marked anaglyph 3D video, and then the Normalized

Correlation value (NC) and the Bit Error Rate (BER)

SECRYPT 2019 - 16th International Conference on Security and Cryptography

264

are calculated between original and extracted mark to

prove the robustness against an attack by quantifying

the visual quality of the extracted mark. In fact,

the NC and BER reﬂect respectively the degree of

similarities and dissimilarities between original and

extracted mark. The maximum NC value is close to

1 if there is a correspondence between original and

extracted mark and it converges to 0 if there is a

difference. Contrary to NC value, the BER value is

close to 0 if there is a similarity between original and

extracted mark and it converges to 1 if not.

Foremost, the usual and spatial attacks are

evaluated by applying geometric attacks such as

rotation with various angles (50

◦

and 90

◦

), scaling

(reduction of 15% and 50% and enlargement of

250%), cropping and resizing. Indeed, The obtained

NC values are between 0.8 and 0.9 and the BER

values are between 0.001 and 0.003. These results

conﬁrm the robustness of the proposed method

against geometric attacks due to the invariance of

DWT to these attacks and the repetition of the mark

in each sprite of the sub-sequence video. Second,

different noise attacks such as salt & pepper, Gaussian

and Speckle noise are applied to the marked anaglyph

3D video. The obtained NC and BER values are

about 0.9 and 0.002 respectively almost times and

the detection of the mark is possible after these noise

attacks. Then, the average and Gaussian ﬁlters are

applied to the marked anaglyph 3D video. The BER

was almost about 0.002 with a correlation about 0.85

almost times. This robustness was enhanced because

of the invariance of the used DWT transformation. In

addition, the blurring attack is applied on the marked

video by using a low-pass ﬁlter with a Gaussian

variance equal to 5 and the detection of the mark

succeeds where the NC and BER values are about 0.9

and 0.001 respectively for most of video test.

At a second step, temporal manipulations are

evaluated by implementing various transformations

on the marked anaglyph 3D videos. First, the

frame based attacks are applied on the marked video

like frame suppression and swapping. Due to the

embedding of the mark in the different sprites of

the video, the detection of the signature succeeds

where the average of the NC and the BER values

were about 0.9 and 0.001, respectively. Then,

the histogram equalization and intensity adjustment

manipulations are applied on the different marked

video tests by enhancing the contrast of the marked

video sequences. After these attacks, the mark still

readable with an average of NC and BER values

about 0.9 and 0.002, respectively. Moreover, The

proposed approach is evaluated against the MPEG-4

compression with a variable bit rate (from 2 Mbps

to 512 kbps) by using the mediacoder software

application

which was adapted to transcode the

marked video. The obtained results show the

robustness against compression because of the use

of multi-sprites as an embedding target where the

average of the NC value was close to 0.9 and the

corresponding average BER value was about 0.001.

Finally, the collusion attack which is considered as

the most dangerous and malicious attack in video

watermarking is tested. It consists in estimating

the embedded mark and removing it easily from the

marked video without damaging its quality. The two

collusion types are performed during the evaluation.

The recovered results of both BER and NC value

are respectively close to 0.002 and 0.9 and they

conﬁrm the robustness of the suggested scheme

against collusion attack due to the use of multi-sprites

which allows marking the same physical points of

each video frame similarly.

Fig. 5 and Fig. 6 illustrate respectively the

average of the obtained NC and BER values after

applying the usual, spatial and temporal attacks on the

different marked video tests.

5 COMPARATIVE STUDY

To prove the efﬁciency of the suggested method,

obtained results are compared with three anaglyph 3D

watermarking existing methods (Waleed et al., 2013;

Salih et al., 2015; Dhaou et al., 2018) based on the

invisibility and robustness criteria. The PSNR values

are calculated for existing and proposed approaches.

These values show that the proposed scheme presents

a good level of invisibility with a PSNR value of about

59 dB. The video reconstruction from multi-sprites

caused a little invisibility degradation compared with

others approaches.

The robustness comparison showed that the

proposed approach is robust against almost of

usual such geometric attacks, noises, ﬁltering,

scaling, blurring, intensity adjustment, histogram

equalization, frame based attacks in addition to the

most malicious attacks which are MPEG compression

and collusion. However, the other existing methods

cannot resist these two last dangerous attacks and the

frame based attacks such as frame suppression and

swapping. In brief, Table 1 summaries the robustness

results for each existing techniques.

http://www.mediacoderhq.com

A Robust Anaglyph 3D Video Watermarking based on Multi-sprite Generation

265

Figure 5: Robustness results: the average of NC values obtained after applying attacks for different test videos.

Figure 6: Robustness results: the average of BER values obtained after applying attacks for different test videos.

Table 1: Robustness comparison.

Attacks Noise Filtering Blurring Rotation Cropping Scaling Frame

Supp. /

Swap.

Histogram

Eq.

Intensity

adj.

JPEG MPEG-4 Collusion

Proposed

approach

X X X X X X X X X X X X

(Waleed

et al.,

2013)

X X X X X X - X X X - -

(Salih

et al.,

2015)

X X X X X X - X X X - -

(Dhaou

et al.,

2018)

X X X X X X X X X X X -

6 CONCLUSIONS

This paper presented a new anaglyph 3D video

watermarking approach which is robust against

MPEG compression and collusion attacks. In fact, it

is based on multi-sprites which present an efﬁcient

embedding target for 2D and 3D videos. In order

to maximize the trade-off between invisibility and

robustness against usual and malicious attacks, a

hybrid scheme based on LSB and DWT insertion was

applied for every sprite generated from each set of 25

frames. The signature was embedded in both high and

low frequencies, obtained after DWT composition,

by applying the LSB on LL sub-band and DWT on

LH and HH sub-bands. The choice of multi-sprites

as an insertion target provides a robustness against

collusion attack and a high quality of marked video

reconstruction compared with single sprite generated

from the whole video. Besides, the use of DWT as

embedding domain and the LSB technique maximizes

the level of invisibility and the robustness against

usual attacks.

Experimentations show that the proposed method

is robust against several attacks like geometric

SECRYPT 2019 - 16th International Conference on Security and Cryptography

266

Figure 7: Invisibility comparison based on the PSNR

values.

attacks, compression, noises, ﬁltering, frame

suppression and against collusion with an average

NC value higher than 0.8 and average BER value

lower than 0.006. Moreover, the proposed technique

allows obtaining a high visual quality level. In

addition, the comparison of the proposed method

with existing works shows the good performance of

the proposed technique to resist the majority type of

attacks with a good invisibility level. As future work,

the invisibility level can be enhanced by improving

the embedding scheme and the generation step of

multi-sprites.

REFERENCES

Barhoumi, W., Bakkay, M. C., and Zagrouba, E. (2011).

An online approach for multi-sprite generation based

on camera parameters estimation. Signal, Image and

Video Processing, 7(5):843–853.

Bayoudh, I., Ben Jabra, S., and Zagrouba, E. (2015). On

line video watermarking-a new robust approach of

video watermarking based on dynamic multi-sprites

generation. In VISAPP, pages 158–165.

Bayoudh, I., Ben Jabra, S., and Zagrouba, E. (2017a).

Online multi-sprites based video watermarking robust

to collusion and transcoding attacks for emerging

applications. Multimedia Tools and Applications,

77(11):14361–14379.

Bayoudh, I., Ben Jabra, S., and Zagrouba, E. (2017b). A

robust video watermarking for real-time application.

In International Conference on Advanced Concepts

for Intelligent Vision Systems, pages 493–504.

Devi, H. S. and Singh, K. M. (2016). A robust

and optimized 3d red-cyan anaglyph blind image

watermarking in the dwt domain. Contemporary

Engineering Sciences, 9:1575–1589.

Dhaou, D., Ben Jabra, S., and Zagrouba, E. (2018).

An efﬁcient group of pictures decomposition based

watermarking for anaglyph 3d video. In the

13th International Joint Conference on Computer

Vision, Imaging and Computer Graphics Theory

and Applications (VISIGRAPP 2018, VISAPP), pages

501–510.

Kerbiche, A., Ben Jabra, S., Zagrouba, E., and Charvillat,

V. (2017). Robust video watermarking approach

based on crowdsourcing and hybrid insertion.

In International Conference on Digital Image

Computing: Techniques and Applications (DICTA),

pages 1–8.

Kerbiche, A., Ben Jabra, S., Zagrouba, E., and Charvillat,

V. (2018). A robust video watermarking based on

feature regions and crowdsourcing. Multimedia Tools

and Applications, 77(20):26769–26791.

Koubaa, M., Elarbi, M., Amar, C. B., and Nicolas, H.

(2012). Collusion, mpeg4 compression and frame

dropping resistant video watermarking. Multimedia

tools and applications, 56(2):281–301.

Munoz-Ramirez, D. O., Reyes-Reyes, R., Ponomaryov,

V., and Cruz-Ramos, C. (2015). Invisible digital

color watermarking technique in anaglyph 3d images.

In 12th International Conference on Electrical

Engineering, Computing Science and Automatic

Control (CCE), pages 1–6.

Prathap, I. and Anitha, R. (2014). Robust and blind

watermarking scheme for three dimensional anaglyph

images. Computers and Electrical Engineering,

40(1):51–58.

Ruchika, P. and Parth, B. (2015). Robust watermarking

for anaglyph 3d images using dwt techniques.

International Journal of Engineering and Technical

Research (IJETR), 3(6):55–58.

Salih, J. W., Abid, S. H., and Hasan, T. M. (2015).

Imperceptible 3d video watermarking technique based

on scene change detection. International Journal of

Advanced Science and Technology, 82:11–22.

Sanjay R., Z. and Ravindra B., R. (2015). A robust

dwt watermarking for 3d images. International

Journal on Emerging Trends in Technology (IJETT),

2(1):210–214.

Sharma, P. K. (2012). Analysis of image watermarking

using least signiﬁcant bit algorithm. International

Journal of Information Sciences and Techniques

(IJIST), 2:666–673.

Waleed, J., Jun, H. D., Hameed, S., Hatem, H., and Majeed,

R. (2013). Integral algorithm to embed imperceptible

watermark into anaglyph 3d video. International

Journal of Advancements in Computing Technology,

5(13):163.

Y, R. and Krishna, D. R. (2016). Digital watermarked

anaglyph 3d images using frft. International

Journal of Computer Trends and Technology (IJCTT),

41(2):77–80.

Zadokar, S. R., Raskar, V. B., and Shinde, S. V. (2013).

A digital watermarking for anaglyph 3d images. In

International Conference on Advances in Computing,

Communications and Informatics (ICACCI), pages

483–488.

Zeng, L., Zhang, S., Zhang, J., and Zhang, Y.

(2014). Dynamic image mosaic via sift and dynamic

programming. Machine vision and applications,

25(5):1271–1282.

A Robust Anaglyph 3D Video Watermarking based on Multi-sprite Generation

267