Modeling of Image Copyright Protection using Discrete Cosine

Transform Hash and Blockchain

Carles Juliandy

, Ronsen Purba

, Roni Yunis

, Darwin

Information Technology Master Department, STMIK Mikroskil, Medan, Indonesia

Information System Department STMIK Mikroskil, Medan, Indonesia

Keywords: Discrete Cosine Transform Hash, Blockchain, Copyrights Protection, Images

Abstract: Protecting copyright is an important issue because now such works as an image can be sold online for

making income. With the rapid development of distribution media, a centralized management system cannot

protect copyrights properly. Because now some research about image plagiarism can detect image

modification as plagiarism, but cannot detect rotated image as plagiarism. Application of Discrete Cosine

Transform (DCT) hash with adding looping steps can detect rotated images as plagiarism. Uploaded image

we looped it rotate 22,5

○

and saved the hash value from DCT hash each time rotate until 180

○

and then

compare each hash with the first hash to get the rotating plagiarism image. After that, a combination with

Blockchain which is a decentralized management system is a solution to protect copyrights now, with the

application of Blockchain and digital signature, making it difficult for other people to make changes to the

data which is stored in the block. This research results showed that the use of DCT hash can reach accuracy

until 99,67% to detect rotating image as plagiarism, and the mining time of the Blockchain with 10.000

blocks and difficulty target 5 needed 1591204,671 seconds.

1 INTRODUCTION

Protecting copyright was an important issue because

creativity of the author. Currently works such as

images can be sold online for income. Copy of the

media that easy to do has an impact on the

modification of media that is easily done too (Cho

and Jeong, 2019; Mehta, 2019; Ravindran, Zacharia

& Roy, 2018). Image plagiarism was used or

modified some or whole parts of the image without

any permission and give credit to the author (Aghav

et al. 2014; Ovhal et al. 2016). To detect an image

modification, the use of a cryptographic hash can

cause an avalanche effect, which was the effect

where a small change of input value can lead to a

drastic change of output value that can make it

difficult to detect if there was any modification of

the image. Perceptual hash was hashing algorithm

that differently from the cryptographic hash, that can

keep away from the avalanche effect, which is a

small change in input value will affect some or none

bit change (Mehta, 2019; Drmic et al., 2017).

Recently centralized management system can’t

protect the copyright as well as the decentralized

management system. Blockchain as a decentralized

management system that is immutable, integrity,

traceability, and transparency can protect it better

because Blockchain doesn’t need a centralized

server and interference from network members (Cho

and Jeong, 2019; Kibet, Simon and Karume, 2018).

The used of Blockchain to protect the image

Knirsch (Knirsch et al. 2018) researched the use of

Blockchain and smart contracts using the concept of

digital signature to provide private key and public

key to the proof of possession and can be forensic

evidence when claiming the copyright of the image.

(Mehta, 2019) research the use of Blockchain and

perceptual hash, where the Blockchain and smart

contracts are used to record all transactions at the

image marketplace. Perceptual hash and hamming

distance are used to detect the similarity of two

images. In this research, the perceptual hash can

detect any modification as plagiarism, but it can’t

detect the modification of 90

○

rotation image as

plagiarism.

The use of perceptual hash to detect the image

similarity has been researched by some researchers

before. Aghav et al., 2014 researched the capability

128

Juliandy, C., Purba, R., Yunis, R. and Darwin, .

Modeling of Image Copyright Protection using Discrete Cosine Transform Hash and Blockchain.

DOI: 10.5220/0010304600003051

In Proceedings of the International Conference on Culture Heritage, Education, Sustainable Tourism, and Innovation Technologies (CESIT 2020), pages 128-134

ISBN: 978-989-758-501-2

of perceptual hash to detect the rotation image as

plagiarism. This research generated a hash value

every time rotate the image 22,5

○

and then compare

it with the real hash value of the image in the

dataset. Rivas et al., 2017 researched the use of

perceptual hash to detect similarity images when

uploaded on social media. Drmic (Drmic et al. 2017)

compared each well-known algorithm in perceptual

hash such as average hash, differential hash, discrete

cosine transform (DCT) hash, and wavelet hash.

This research showed that DCT hash is the most

robust perceptual hash algorithm to detect similarity

image.

In this paper, we introduced the model that not

only detected the similarity of the uploaded image

but can protect the copyright of the image that is

already saved. The proposed model combined the

use of Blockchain technology with the digital

signature and perceptual hash to (1) protect and

prevent the attempt to change important information

such as hash value, image owner name, image name,

uploader name, and image added date, (2) to detect

the similarity of modification image such as gamma

correction, resize, rotate, crop, and salt and pepper

noise that try to upload especially to detect rotation

of 90

○

, 180

○

, and 270

○

that previous research can’t

solve it, (3) with ECDSA digital signature it can be

an additional security to protect and proof the

possession of the received upload image.

As mention above about the proposed model, it

can be explained that the contribution of this

research is the use of looping steps for DCT hash to

detected the plagiarism image, especially for the

rotation image which cannot be detected by the

previous researcher model. The improved DCT hash

is combined with Blockchain technology and the

digital signature to improve the security of data, so it

cannot be changed once it is saved into the block.

The remaining paper is structured as follows:

Section 2 provides background research related to

perceptual hash and blockchain application. Section

3 presents research methodology or our approach to

detect the similarity of the uploaded image and

prevent the attempt to change important data about

the image that is already saved before and to proof

of possession with the use of ECDSA. Section 4

provided the result and discussion and Section 5

provided the conclusion of the paper.

2 RELATED WORKS

Blockchain was a decentralized management system

invented by Satoshi Nakamoto in 2008 and

implemented in 2009. Bitcoin is the first application

that implemented this technology to handle the

transaction of cryptocurrency. As a result, Bitcoin

did not need a third party to validate the transaction.

All transaction in Bitcoin is validated by together

agreement which is called Consensus. Blockchain

isn’t a standalone technology, it consists of

cryptography, mathematics, algorithm, economic

model, combine of peer to peer network (P2P), and

consensus algorithm which is agreed by everyone

who joined to the network (Wang et al. 2018). The

use of Blockchain is well known because it's capable

to secure data inside, it prevented other people who

want to change the data which is already saved in

the block. But this technology still has an

opportunity to be hacked, if the attack is offense

more than 50% (50%+1) of network members at the

same time. But this is something that almost

impossible to do because it needed many resources

in computing. (Lin and Liao 2017).

The related works about Blockchain and

perceptual hash have been done by some

researchers. In 2014 (Aghav et al. 2014) research the

use of DCT hash which is one of the perceptual hash

algorithms to detect rotation image modification.

This research generated the hash value every 22,5

○

rotation and then compared it with the hash value of

each image that previously saved before. Compared

the hash value is using a hamming distance if the

hamming distance value below the threshold image

will be rejected, if the hamming distance value

above the threshold it will looping rotate the image

and repeat the step before until the image rotated

180

○

clockwise and anti-clockwise. This research

model is when there is just one hamming distance

value below the threshold all processes will stop and

it will be considered as plagiarism image. Bhowmik

and Feng, (2017) researched the use of Blockchain

store the watermark of unique information that

consisting of transaction history and the hash value

of image which can be used to find a similarity

image. The result of this research is using history

transactions and the value of image hash, it can be

defined as the part of the image that is edited or be

changed. Knirsch et al., (2018) research the use of

smart contracts with digital signature to handle the

claim of copyright possession by generated private

key and public key. In this research private key is

kept by the author of image, and the public key is

used by another to verified the possession of image.

But the analysis of this research is focus on evaluate

the operational cost implementation of this methoed

indeed of evaluate the effectiveness of this method.

The conclusion of this research is if the more image

Modeling of Image Copyright Protection using Discrete Cosine Transform Hash and Blockchain

129

size than the more cost is needed to implement this

method.

Jnoub and Klas, (2019) researched the use of

Blockchain to protect the image and the copyright

by register the ownership information and copyright

into the block of Blockchain. In this research, the

image is not stored in Blockchain, but just store tow

hash value of the image which is extracted directly

from the image, and the second hash value is

extracted using speed up robust feature (SURF).

This research efficient the storage of the Blockchain

because it just stored text value. Mehta (2019)

researched the use of perceptual hash to detect the

similarity of the modification image. Blockchain is

used in this research to record the transaction for

someone who uploaded the plagiarism image and get

a financial penalty because of that. In this research,

there is a problem when the proposed method failed

to detect the rotate image of 90

○

as a plagiarism

image. Andi, Purba, and Yunis, (2019) researched

the use of Blockchain to prevent plagiarism in a

scientific publication. This research uses the

combination of Blockchain, SHA-256, and ECDSA

to protect the publication data, so it’s nearly

impossible for other people who tried to change the

data. The benefit of this research is it can prevent the

reviewer or someone who handles the publication to

used, changed, or modified the paper which needs to

be published. Because the paper is already signed

with ECDSA and only can be changed with the

private key.

3 RESEARCH METHODOLOGY

In this paper, we proposed the model to solve three

problems: (1) the use of cryptographic hash to detect

the similarity of two images can cause the avalanche

effect, wherewith the small change of input value

can cause a drastic change in output value, which

make it impossible to detect the similarity image, (2)

the implementation of perceptual hashing (DCT

hash) with looping step can detect other

modification as plagiarism, such as gamma

correction, resize, crop, rotate, and salt and pepper

noise. Which is for the rotation it failed to detect 90

○

rotated image as plagiarism in previous research, (3)

the implementation of Blockchain to protect the

image, so it difficult for the owner to prove that the

image is too many.

This research used the same dataset as previous

research from (Mehta, 2019) the dataset is from

Berkeley Segmentation Dataset (BSDS 500) which

contained 500 images and then modified the dataset

with 42 modifications that fitted (Mehta, 2019)

dataset So at last, the dataset has 21.500 images (500

original images and 21.000 modified. The 42

modification of the dataset is consisted of:

 Rotation (in degrees): 5°, 10°,15°,20°,25°

(clockwise)

 Gamma correction: 0.5, 1.0, 1.5, 2.0, 2.5, 3.0,

3.5

 Salt and pepper noise: 0.05, 0.10, 0.15, 0.20,

0.25, 0.30, 0.35, 0.40, 0.45, 0.50, 0.55, 0.60,

0.65, 0.70, 0.75, 0.80, 0.85, 0.90, 0.95, 1.00

 Crop (fixed aspect ratio mode): 5%, 10%, 15%,

20%, 25%

 Resize (size reduction - fixed aspect ratio

mode): 5%, 10%, 15%, 20%, 25%

This research proposed a new model of

implementation of perceptual hash and Blockchain

to detect similarity and prevent the attempt to

change the data in Blockchain. The perceptual

hashing algorithm in this research is used the DCT

hash algorithm which is the most robust perceptual

hash algorithm (Drmic et al. 2017).

The DCT hash algorithm is a perceptual hashing

algorithm that concludes binaries value which

represented the image. This algorithm is based on

cosine transform which is first it reduced the image

into lower pixels than it transformed to grayscale

after that it got the DCT value from that reduced and

grayscale image. After that, it counts the average

value of that DCT value, and that compared it one

by one that DCT value with the average value to get

the binary number of an image.

The proposed model is shown in Figure. 1 below

Figure 1: Proposed Model of Perceptual Hash dan

Blockchain.

The proposed model consisted of two big part

that can be explained below:

CESIT 2020 - International Conference on Culture Heritage, Education, Sustainable Tourism, and Innovation Technologies

130

3.1 Front-end

In the front-end menu, the image owner or uploader

will fill the upload form that consisted of the image

name, image owner name, uploader name, and

upload date. After filled the upload form, before

submit, the uploader needed to check the plagiarism.

This check will process on the back-end that will

return the result if the image is considered

plagiarism or not plagiarism. If the image is

considered plagiarism the submit button cannot be

clicked, if the image is not plagiarism the submit

button will be available to click.

3.2 Back-end

On the back-end side after receiving the request to

check the plagiarism, the image will convert into a

hash value using the DCT hash algorithm and the

hash value will store temporarily in the array. After

that the looping process will start, it started to rotate

the image 22,5

○

until 157,5

○

clockwise after that it

will generate and store the image hash in the array

that was created before every time it rotated 22,5

○

After rotation is 157,5

○

image will rotate 22,5

○

anti-

clockwise until 180

○

and generate and store it again

in the array. Now we have 16 hash value of the

uploaded image (1 from original, 7 from clockwise

rotation, and 8 from anti-clockwise rotation). This

array with all that hash value will compare with all

image hash that is already in the dataset. To

compared this image hash we used a hamming

distance algorithm to compare bit per bit hash. The

hamming distance result will compare to a threshold

value that is already set. In this research, the

threshold value is 14 (based on previous research by

(Drmic et al. 2017)). If there is just one compared

result below the threshold value it will be detected as

a plagiarism image, and if all of the compare results

above of threshold result in it return a value as free

from the plagiarism act to the front end. After that, it

can press the submit button and can be continued to

submit. The submission process will store the data

that already fill in the upload form include the

original image hash and convert all that data with the

SHA-256 algorithm to create a new block and that

data will be signed with the ECDSA algorithm. The

private key will be store by the image owner as proof

of possession that the owner has that image. The

public key will be used by another member to verify

that the sign value is true and never changed before.

In this proposed model, we tried two different

testing methods that are qualitative test and a

quantitative test. In the qualitative test we test the

function of all proposed models, from testing the

DCT hash model to detect some modification

images such as rotation, crop, resize, gamma

correction, and salt and pepper noise. After that, we

tested the Blockchain model to protect the data in

the block, and next is to testing the ECDSA to verify

the data that already sign. In the quantitative tested

we test the accuracy of the DCT hash model to

detect the rotate image as plagiarism. The testing is

using 3 scenarios, try to upload 200, 400, and 600

images, and the second quantitative test is to test the

mining time of the Blockchain model if there is

someone who tried to change the data in the block.

Our testing show that this model is capable to detect

the plagiarism rotate the image and protect the

image data from the possible plagiarism act.

4 RESULTS AND DISCUSSIONS

In this research, we build the application based on

the proposed model using node.js. In the testing, we

separate the test into 2 parts, qualitative test to show

the functionality of the proposed model from the

DCT hash to detect plagiarism and Blockchain with

ECDSA to protect the data that already saved. The

second part test is a quantitative test to show the

accuracy of the DCT hash model to detect the rotate

image as plagiarism and to show how the

Blockchain model to handle the attempt to change

the data.

For the qualitative test we try the DCT hash

model with upload the original image that doesn’t in

the dataset before, and some modification images

such as rotation, crop, resize gamma correction, and

salt and pepper noise. The result showed that the

functionality of the DCT hash model works well.

When uploaded the original image that doesn’t in

the dataset before, the model can detect it as a non-

plagiarism image and can continue to submit the

process. The resulted testing can be seen in Figure 2.

Figure 2: Successful Received Image

Modeling of Image Copyright Protection using Discrete Cosine Transform Hash and Blockchain

131

For the modification image, the model can detect

it well as a plagiarism image, we have tried it with

several modifications and all of them can be

detected as plagiarism. The result of this testing can

be seen in Figure 3. Below, Figure 3. is the rotate

○

image.

Figure 3: Rotation Plagiarism Image.

Another qualitative test is testing the capability

of Blockchain and ECDSA to prevent an attempt to

change the data inside. For the Blockchain model,

we tried to change the block data. For example, we

tried to change the name of the image that was

already saved in the block. This change made the

whole block after the changed block become an

invalid block. This can be seen in Figure. 4 which is

a valid block with green color before any changed

data and in Figure. 5 below which is the invalid

block became red color with changed data from

image 2 to image 3.

Figure 4: Valid Block Before Change Data.

Figure 5: Invalid Block After Change Data.

This qualitative test showed that our proposed

model can protect the data well. It prevented other

people to change the value inside. To make it a valid

block it needed to mine block per block when on the

other hand it seems like something impossible for

the usual computer except supercomputer.

For ECDSA testing we try to change the hash

value of the image that is already signed before,

after that we try to validate the signed text with the

public key from the first signed. The result shows

that the public key can’t validate the data because

the data is different when it first signed. This means

that our proposed model of ECDSA can protect the

signed data well and prevent it from changing. The

result of ECDSA testing can be seen in Figure 5.

Figure 6: Fail Validation ECDSA.

For the quantitative test, we have tested the

accuracy of DCT hash with 3 different scenarios.

The first scenario is uploaded 200 images, the

second scenario is uploaded 400 images, and the

third scenario is uploaded 600 images. the

CESIT 2020 - International Conference on Culture Heritage, Education, Sustainable Tourism, and Innovation Technologies

132

measurement of this quantitative test is using the

confusion matrix with the operator

 True Positive (TP) for actually the plagiarism

images and the DCT recognized it as a

plagiarism image.

 True Negative (TN) for actually the not

plagiarism images and the DCT recognized it

as not plagiarism image.

 False Positive (FP) for actually the not

plagiarism images and the DCT recognized it

as a plagiarism image.

 False Negative (FN) for actually the plagiarism

images and the DCT recognized it as a not

plagiarism image.

For the quantitative test, we have tested the

accuracy of DCT hash with 3 different scenarios.

The first scenario is uploaded 200 images that

consist of 50 original images that don’t in the dataset

before, 50 random images from the dataset and

rotated it with 90

○

, 50 random images from the

dataset and rotated it with 180

○

, and 50 random

images from the dataset and rotated it with -90

○

. The

detailed testing of the first scenario can be seen in

Table 1

(Rc is Received and Rj is Rejected).

Table 1: First Scenario Testing Result.

Image

type

Image

qty

Result

Rc Rj TP TN FP FN

Original

Ima

e 50 49 1 0 49 1 0

Rotated

○

50 0 50

50 0 0 0

Rotated

-90

○

50 1 49 49 0 0 1

Rotated

180

○

50 0 50

50 0 0 0

Total: 200 149 49 1 1

For the first scenario with a confusion matrix, it

gained 49 TN and 149 TP. In percentage, it gained

99% accuracy this is because there is one image that

not a plagiarism image, and the DCT hash detected

it as a plagiarism image. On the other hand that is

one plagiarism image and the DCT hash cannot

detect it as a plagiarism image.

For the second scenario, the image type is the

same as the previous scenario, but the difference is

the amount of testing image. In this scenario, we use

100 images of each image type. The total image of

this testing is 400 images. In this scenario, we get

100% accuracy with the confusion matrix. The result

of this scenario can be seen in Table 2.

Table 2: Second Scenario Testing Result.

Image

type

Image

qty

Resul

Rc R

TP TN FP FN

Original

Ima

e 100 100 0 0 100 0 0

Rotated

○

100 0 100 100 0 0 0

Rotated

-90

○

100 0 100 100 0 0 0

Rotated

180

○

100 0 100 100 0 0 0

Total: 400

300 100 0 0

For the second scenario with a confusion matrix,

it gained 100 TN and 300 TP. In percentage, it

gained 100% accuracy this is because all of the

images detected like the images should be,

plagiarism as plagiarism, and not plagiarism as a not

plagiarism.

For the third scenario, the image type is the same

as the previous scenario, but the difference is the

amount of testing image. In this scenario, we use

150 images of each image type. The total image of

this testing is 600 images. In this scenario, we get

100% accuracy with the confusion matrix. The result

of this scenario can be seen in Table 3.

Table 3: Third Scenario Testing Result.

Image

type

Image

qty

Result

Rc Rj TP TN FP FN

Original

Image 150 150 0 0 150 0 0

Rotated

○

150 0 150 150 0 0 0

Rotated

-90

○

150 0 150 150 0 0 0

Rotated

180

○

150 0 150 150 0 0 0

Total: 600 450 150 0 0

For the third scenario with a confusion matrix, it

gained 150 TN and 450 TP. In percentage, it gained

100% accuracy this is because all of the images

detected like the images should be, plagiarism as

plagiarism, and not plagiarism as a not plagiarism.

From three scenarios testing, we get the average

accuracy of this model is

(99%+100%+100%) / 3 = 99,67%.

For the second quantitative test, we test the

mining time of the Blockchain. Mining time was a

time to use for mine or to find a fit public key for

Blockchain private key. We tested mining time in

second by using the number different of blocks from

Modeling of Image Copyright Protection using Discrete Cosine Transform Hash and Blockchain

133

10, 50, 500, 1000, 5000, 10000 blocks with different

difficulty target (DT) from 1 to 5. The details of this

testing can be seen in Table 4 below.

Table 4: Mining Time Testing Result.

Number

Blocks

Minin

Time

(

Second

)

DT1 DT2 DT3 DT4 DT5

10 0,2 0,6 4,7 11,341 28,578

50 0,8 3,0 6,9 15,321 39,465

500 33,0 60,8 133,8 289,561 621,59

1000 156,1 412,0 865,5 1832,5 4120,5

5000 2.962,6 7167,5 15012,3 37891,5 81952,91

10000 75643,5 160824,5 340156,8 721051,3 1591204,6

The result of this testing showed that the number

of blocks and difficulty target is correlated to mining

time. It meant that the more blocks and the higher

the difficulty target you want to mine and use, the

more time you need. For example, we can see the

highest mining time to mining 10000 blocks with the

difficulty target is 5, it needed about 1.591.204,671

seconds, or in the day it about 18 days more.

This result meant that Blockchain is a robust

technology to protect the data that are already saved

inside.

5 CONCLUSION AND FUTURE

RESEARCH

author for their works and idea, that why protecting

the copyright of the image is important. Previously

researched image copyright protection using

perceptual hash and Blockchain can detect

plagiarism when uploading images, but for some

cases like the rotated image like 90 degrees. The

proposed model with the implementation of DCT

hash and Blockchain can protect the image

modification image especially the rotated image

with an accuracy of 99,67%. The implementation of

the Blockchain proved that is capable to protect the

data that already in the block. With the mining time

needed when in the largest difficulty target, it shows

that it was nearly possible to change the data. For

future research, it can be implemented using another

algorithm that can detect the modification images

easier and combined well with the implementation

of Blockchain so the new model can more efficient

in time consumption and more accurate to detect the

modification image.

REFERENCES

Aghav, Sushila, et al. 2014. “Mitigation of Rotational

Constraints in Image-Based Plagiarism Detection

Using Perceptual Hash.” 2(1): 28–32.

Andi, Ronsen Purba, and Roni Yunis. 2019. “Application

of Blockchain Technology to Prevent The Potential Of

Plagiarism in Scientific Publication.” Proceedings of

2019 4th International Conference on Informatics and

Computing, ICIC 2019.

Bhowmik, Deepayan, and Tian Feng. 2017. “The

Multimedia Blockchain: A Distributed and Tamper-

Proof Media Transaction Framework.” International

Conference on Digital Signal Processing, DSP 2017-

Augus.

Cho, Sunghyun, and Chiyoung Jeong. “A Blockchain for

Media : Survey.” 2019 International Conference on

Electronics, Information, and Communication

(ICEIC): 1–2.

Drmic, Andrea, et al. 2017. “Evaluating the Robustness of

Perceptual Image Hashing Algorithms.” : 995–1000.

Jnoub, Nour, and Wolfgang Klas. “Detection of Tampered

Images Using Blockchain Technology.” 2019 IEEE

International Conference on Blockchain and

Cryptocurrency (ICBC): 70–73.

Kibet, Alex, Prof Simon, and Maina Karume. 2018. “A

Synopsis of Blockchain Technology.” 7(11): 789–95.

Knirsch, Fabian, et al. 2018. “EVALUATION OF A

BLOCKCHAIN-BASED PROOF-OF-POSSESSION

IMPLEMENTATION Center for Secure Energy

Informatics, Salzburg University of Applied Sciences,

Puch / Salzburg , Austria.” 865082(865082).

Lin, Iuon Chang, and Tzu Chun Liao. 2017. “A Survey of

Blockchain Security Issues and Challenges.”

International Journal of Network Security 19(5): 653–

59.

Mehta, Rishabh. “Decentralised Image Sharing and

Hashes.” 2019 11th International Conference on

Communication Systems & Networks (COMSNETS)

2061: 1–6.

Ovhal, Prajakta Mahendra, et al. 2016. “Plagiarized Image

Detection System Based on CBIR To Cite This

Version : HAL Id : Hal-01284675 Plagiarized Image

Detection System Based on CBIR.” 4(3).

Ravindran, M K, Balaji Zacharia, and Antony Roy. 2018.

“Plagiarism and Copyright , Acknowledgements,

Disclosure and Conflicts of Interest.” : 207–14.

Rivas, Alberto et al. 2017. “On Hashes Extraction.” 1(d):

87–94.

Wang, Huaimin et al. 2018. “Blockchain Challenges and

Opportunities: A Survey.” International Journal of

Web and Grid Services 14(4): 352.

CESIT 2020 - International Conference on Culture Heritage, Education, Sustainable Tourism, and Innovation Technologies

134