(For USM Staff/Student Only)

EngLib USM > Ω School of Electrical & Electronic Engineering >

Investigation and development of convolutional neural network based image splicing detection / Siti Mastura Binti Md Hasim

Investigation and development of convolutional neural network based image splicing detection_ Siti Mastura Binti Md Hasim_E3_2017-MFAR
Pengesan penyambatan imej telah menjadi satu bidang kajian yang sangat penting di seluruh dunia. Kepentingan untuk mengesan penyambatan imej tidak terhad kepada pihak yang berkuasa sahaja malah kepada semua pengguna biasa. Pengesan penyambatan imej memerlukan beberapa langkah untuk dipenuhi dan set data yang besar diperlukan. Kajian ini bermatlamat untuk menyiasat dan membina kaedah berdasarkan konvolusi rangkaian neural (CNN) untuk mengesan penyambatan imej. Tiga eksperimen awal telah dilakukan berdasarkan kajian sebelum ini untuk menyiasat bagaimana pra-pemprosesan membezakan prestasi CNN. Dari eksperimen awal yang dijalankan, satu rangka kerja dengan pengurangan nombor lapisan CNN telah dicadangkan tanpa sebarang pra-pemprosesan. Pengesahan silang dengan sepuluh lipatan telah digunakan untuk mendemonstrasi prestasi CNN. Eksperimen awal telah menunjukkan prestasi CNN sangat terjejas dengan saiz imej input. Oleh itu, reka bentuk yang dicadangkan telah diuji dengan tiga imej input yang berlainan saiz iaitu 28×28 piksel, 64×64 piksel dan 128×128 piksel. Dari pengesahan silang, 64×64 piksel imej input telah dikonklusikan sebagai saiz yang paling sesuai untuk pengesan penyambatan imej menggunakan CNN. Di akhir kajian ini, dapat dilihat bahawa dengan menggunakan reka bentuk yang dicadangkan, CNN dapat digunakan tanpa sebarang pra-pemprosesan. Image splicing detection is an area of studies that have been studied widely all around the world recently. The importance to do image splicing detection is not only for the authorities but also for common user. Image splicing detection requires several steps to be completed and a huge dataset is needed to be used. This study is aimed to investigate and develop CNN based method for image splicing detection. Three preliminary experiments are done according to previous work to observe how pre-processing affects CNN performance. Based on the preliminary experiments, an architecture with reduced number of CNN layers are proposed without any pre-processing. Ten-fold cross validation is used to demonstrate CNN performance. Preliminary experiments shows that CNN performance are critically affected by input image size. Therefore, the proposed architecture are tested with different input image sizes. Three different input image sizes are tested which are 28×28 pixel, 64×64 pixel and 128×128 pixels. From cross validation is can be concluded that 64×64 pixels input image is the most suitable input image size for CNN image splicing detection. At the end of this study, it is observed that by using the proposed architecture, CNN can be used for image splicing detection without any pre-processing.
Contributor(s):
Siti Mastura, Md Hasim - Author
Primary Item Type:
Thesis
Language:
English
Sponsor - Description:
Pusat Pengajian Kejuruteraan Elektrik & Elektronik -
First presented to the public:
8/1/2017
Original Publication Date:
4/23/2018
Previously Published By:
Universiti Sains Malaysia
Place Of Publication:
School of Electrical & Electronic Engineering
Citation:
Extents:
Number of Pages - 87
License Grantor / Date Granted:
  / ( View License )
Date Deposited
2018-04-23 14:42:09.976
Date Last Updated
2020-05-29 17:31:53.425
Submitter:
Mohd Fadli Abd. Rahman

All Versions

Thumbnail Name Version Created Date
Investigation and development of convolutional neural network based image splicing detection / Siti Mastura Binti Md Hasim1 2018-04-23 14:42:09.976