Pengembangan Sistem Deteksi Steganografi Medis Menggunakan Deep Steganalysis Network

Widyanto Wibowo, Farhan (2026) Pengembangan Sistem Deteksi Steganografi Medis Menggunakan Deep Steganalysis Network. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025221209-Undergraduate_Thesis.pdf Restricted to Repository staff only Download (9MB) | Request a copy

Abstract

Steganografi adaptif keluarga UNIWARD menyembunyikan perubahan piksel pada area bertekstur sehingga sulit dideteksi, termasuk pada citra medis grayscale yang integritasnya penting bagi kepercayaan diagnostik dan privasi pasien. Karena steganalisis berbasis deep learning umumnya dikembangkan pada citra natural, domain citra medis belum banyak dievaluasi. Penelitian ini mengembangkan sistem deteksi steganografi citra grayscale berbasis deep residual steganalysis, dengan front-end Spatial Rich Model (SRM) statis dan Truncation Linear Unit yang diikuti backbone CNN keluarga SRNet dalam dua varian, yaitu SRNet Ringkas (srnet_inspired) dan SRNet Penuh (srnet_exact). Keduanya dilatih melalui pipline identik dengan pair-preserving split, kalibrasi temperature scaling, dan ambang batas optimal berbasis F1-score. Sistem diuji pada BOSSBase dengan S-UNIWARD, citra medis Brain Tumor pada payload 0,4 bit per piksel, dan citra natural 256x256 untuk studi sensitivitas payload 0,4 hingga 0,5 bit per piksel. SRNet Penuh konsisten lebih unggul dengan AUROC 0,8288 pada BOSSBase, sedangkan pada citra medis kedua backbone mencapai kinerja yang nyaris sempurna dengan AUROC 0,9999 pada payload 0,4 bit per piksel, jauh melampaui citra natural pada payload yang sama, dan AUROC meningkat monoton seiring bertambahnya payload. SRNet Ringkas meraih sekitar 94,6 hingga 100 persen F1-score SRNet Penuh dengan kurang dari sepertiga jumlah parameter, sehingga lebih efisien untuk lingkungan dengan sumber daya terbatas.
========================================================================================================================================
Adaptive steganography of the UNIWARD family hides pixel changes in highly textured regions so that they are difficult to detect, including in grayscale medical images whose integrity is important for diagnostic trust and patient privacy. Because deep learning based steganalysis is generally developed on natural images, the medical image domain has not been widely evaluated. This research develops a grayscale steganography detection system based on deep residual steganalysis, with a static Spatial Rich Model (SRM) front-end and a Truncation Linear Unit followed by an SRNet family CNN backbone in two variants, namely Compact SRNet (srnet_inspired) and Full SRNet (srnet_exact). Both are trained through one identical pipeline with a pair-preserving split, temperature scaling calibration, and an F1-score based optimal threshold. The system is evaluated on BOSSBase with S-UNIWARD, the Brain Tumor medical images at a payload of 0.4 bits per pixel, and 256x256 natural images for a payload sensitivity study at 0.3 to 0.5 bits per pixel. Full SRNet is consistently superior with an AUROC of 0.8288 on BOSSBase, while on the medical images both backbones achieve near-perfect classification with an AUROC of 0.9999 at a payload of 0.4 bits per pixel, far exceeding natural images at the same payload, and the AUROC increases monotonically as the payload grows. Compact SRNet attains about 94.6 to 100 percent of the F1-score of Full SRNet using less than one third of the parameters, making it more efficient for resource constrained environments.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Steganalisis, Steganografi Adaptif, Spatial Rich Model, Convolutional Neural Network, Citra Medis. ============================================= Steganalysis, Adaptive Steganography, Spatial Rich Model, Convolutional Neural Network, Medical Image
Subjects:	T Technology > T Technology (General) > T57.5 Data Processing
Divisions:	Faculty of Information and Communication Technology > Informatics > 55201-(S1) Undergraduate Thesis
Depositing User:	Farhan Widyanto Wibowo
Date Deposited:	28 Jul 2026 03:43
Last Modified:	28 Jul 2026 03:43
URI:	http://repository.its.ac.id/id/eprint/138438

Actions (login required)

View Item