DCT based watermarking method for social media

Kedmenec, Luka (2015) DCT based watermarking method for social media. Master's thesis - Graduate Programme. Grafički fakultet. [Mentor: Poljičak, Ante].

[img]
Preview
PDF
DB433_Kedmenec_Luka.pdf

Download (1MB) | Preview

Abstract

U ovom je radu razvijena steganografska metoda za zaštitu slika na društvenim mrežama. Metoda ugrađuje informacije poput loga ili serijskog broja. Prijenosom slika na društvenu mrežu dolazi do neželjene degradacije izvorne slike. Na temelju analize napada socijalne mreže razvijena je metoda koja se temelji na diskretnoj kosinusnoj transformaciji – DCT u kojoj se dodatne informacije skrivaju modificiranjem koeficijenata u kosinusnoj domeni. Koeficijent je izabran nakon temeljite analize odnosa šuma i signala (PSNR), indeksa strukturalne sličnosti (SSIM) i učestalost pogrešnih bitova (BER). Koeficijenti srednje frekvencije su se pokazali kao idealni za modifikaciju i ugradnju dodatnih informacija. Modificiranjem tih frekvencija ne degradiramo sliku u toj mjeri da je promjena vidljiva ljudskom oku, a još uvijek je u stanju izdržati različite vrste napada poput skaliranja, šuma, JPEG kompresije, zamućenja i ostalih. Metoda je testirana i rezultati učestalosti pogrešnih bitova (BER) kod ekstrakcije za različite napade su ispod 1%. Ova metoda je implementirana i testirana na 4 različite hardverske platforme s ciljem ubrzavanja izvršavanja koda koristeći OpenCL. Dobivena ubrzanja su postignuta izgradnjom kernela za vremenski najzahtjevnije funkcije DCT i iDCT.

Item Type: Master's thesis - Graduate Programme
Mentor name: Poljičak, Ante
Defence date: 17 July 2015
Abstract in english: In this memory a steganographic method for image protection on social network is developed. The method hides additional information such as a logo or serial number. Uploading the images on the social network has induced image degradation on the original images. Based on the analysis of this attack a method is developed based on the discrete cosine transform - DCT where additional information is hidden by modifying the coefficients in the cosine domain. The coefficient is chosen after thorough analysis based on Peak Signal to Noise Ratio (PSNR), Structuram Similarity Index (SSIM) and bit error rate (BER). It is concluded that the coefficients representing middle range of frequencies in the cosine domaine is best suited for modifications. The reason for this is that the modification of these frequencies does not degrade the picture enough for a human eye to see it but are still able to withstand various types of image attacks such as scaling, rotation, cropping and JPEG compression. The method was evaluated and it has shown that bit error rate (BER) for different attacks was below 1%. The method is implemented on 4 different hardware platforms using OpenCL and then tested for performance in speed and complexity of the implementation. Acceleration was based on creating a kernels for most time consuming functions DCT and iDCT.
Uncontrolled Keywords: steganografija, DCT, OpenCL, društvene mreže, hardverska implementacija, hardversko ubrzanje
Keywords in english: steganography, DCT, OpenCL, social networks, hardware implementation, hardware acceleration
Subjects: TECHNICAL SCIENCES > Graphic Technology
Institution: Grafički fakultet
City: Zagreb
Number of Pages: 56
Callnumber: polj 2015 ked
Inventory number: DB433
Depositing User: Antonia Grgurović
Status: Unpublished
Date Deposited: 25 Aug 2015 12:04
Last Modified: 25 Aug 2015 12:05
URI: http://eprints.grf.unizg.hr/id/eprint/2263

Actions (login required)

View Item View Item

Downloads

Downloads per month over past year