Pembangkitan Kamuflase Citra Menggunakan Model Unsupervised Diffusion

Lauw, Nathaniel Takeshi Inatori (2025) Pembangkitan Kamuflase Citra Menggunakan Model Unsupervised Diffusion. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5002211175-Undergraduate_Thesis.pdf Restricted to Repository staff only until 1 April 2027. Download (14MB) | Request a copy

Abstract

Pembangkitan kamuflase citra menjadi sangat penting saat terdapat konten pribadi dan/atau rahasia. Salah satu kunci utama pada teknik kamuflase citra adalah foreground inpainting atau pewarnaan ulang latar depan citra atau objek yang ingin dikamuflase. Studi terdahulu umumnya bergantung pada augmentasi background citra yang digunakan untuk meleburkan citra objek. Akan tetapi, augmentasi citra tergolong mahal secara komputasi dan hasil pewarnaan citra objek yang kurang membaur dengan warna background citra. Oleh karena itu, penelitian Tugas Akhir ini mengusulkan sebuah model unsupervised berdasarkan model diffusion yang dinamakan dengan Unsupervised Diffusion (UD). Model yang diusulkan bertugas untuk membangkitkan kamuflase citra tanpa augmentasi background citra. Penelitian Tugas Akhir ini juga mengevaluasi model yang diusulkan pada data citra kamuflase Camouflaged/Concealed Object Detection 10K (COD10K). Hasil eksperimen menunjukkan bahwa model Unsupervised Diffusion mampu membaurkan objek citra dengan background citra.
=================================================================================================================================
Camouflage image generation has been increasingly important when dealing with private and/or confidential contents. One fundamental technique in camouflage image is through foreground or object inpainting. Previous studies predominantly depended on background image augmentation which were used to faint objects. However, image augmentation required high computational cost, yet were not able to provide a seamless features to the background image. Therefore, this study proposes an unsupervised version of diffusion model, which called as Unsupervised Diffusion (UD). This proposed model is tasked to generate camouflage images without relying on background iamge augmentation. This study would also evaluate the proposed model in camouflage image dataset, Camouflaged/Concealed Object Detection 10K (COD10K). Presented results showcase the blending capability between object and background image.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Kamuflase citra, model difusi, citra, Camouflaging images, diffusion model, images
Subjects:	Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) Q Science > QA Mathematics > QA76.9 Computer algorithms. Virtual Reality. Computer simulation.
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Mathematics > 44201-(S1) Undergraduate Thesis
Depositing User:	Nathaniel Takeshi Inatori Lauw
Date Deposited:	11 Feb 2025 03:29
Last Modified:	11 Feb 2025 03:29
URI:	http://repository.its.ac.id/id/eprint/118011

Actions (login required)

View Item