<![CDATA[Peningkatan distribusi media digital menyebabkan risiko pelanggaran hak cipta dan manipulasi citra semakin tinggi, khususnya pada dokumentasi digital institusi. Namun, metode watermarking konvensional masih menghadapi keterbatasan dalam menjaga keseimbangan antara ketidaktampakan (imperceptibility) dan ketahanan watermark (robustness) terhadap berbagai serangan citra. Penelitian ini mengusulkan metode watermarking citra digital berbasis hybrid Discrete Cosine Transform (DCT), Hessenberg Decomposition (HD), Singular Value Decomposition (SVD), dan diffusion refinement untuk meningkatkan stabilitas watermark sekaligus mempertahankan kualitas visual citra. Proses embedding dilakukan pada domain frekuensi DCT melalui modifikasi singular value hasil dekomposisi Hessenberg, kemudian diperkuat menggunakan diffusion refinement berbasis DPM-Solver++ multistep untuk meningkatkan kualitas rekonstruksi citra dan stabilitas ekstraksi watermark. Metode diimplementasikan menggunakan Python dan diuji pada delapan citra dokumentasi digital berwarna menggunakan berbagai skenario serangan, meliputi Gaussian blur, JPEG compression, histogram equalization, sharpening, dan salt-and-pepper noise. Hasil eksperimen menunjukkan rata-rata PSNR sebesar 49,29 dB, SSIM sebesar 0,9488, dan NC sebesar 0,9472 pada kondisi normal. Pada serangan JPEG compression, metode masih mempertahankan nilai SSIM sebesar 0,937 dan PSNR sebesar 40 dB. Hasil tersebut menunjukkan bahwa metode yang diusulkan mampu mempertahankan kualitas visual citra sekaligus meningkatkan ketahanan watermark terhadap berbagai distorsi citra digital. Abstract The increasing distribution of digital media has raised the risk of copyright infringement and image manipulation, particularly in institutional digital documentation. However, conventional watermarking methods still face limitations in maintaining a balance between imperceptibility and robustness against various image attacks. This study proposes a hybrid digital image watermarking method based on Discrete Cosine Transform (DCT), Hessenberg Decomposition (HD), Singular Value Decomposition (SVD), and diffusion refinement to improve watermark stability while preserving image visual quality. The embedding process is performed in the DCT frequency domain through singular value modification derived from Hessenberg decomposition, followed by DPM-Solver++ multistep diffusion refinement to enhance image reconstruction quality and watermark extraction stability. The proposed method was implemented using Python and evaluated on eight color digital documentation images under various attack scenarios, including Gaussian blur, JPEG compression, histogram equalization, sharpening, and salt-and-pepper noise. Experimental results achieved an average PSNR of 49.29 dB, SSIM of 0.9488, and NC of 0.9472 under normal conditions. Under JPEG compression attacks, the method still maintained an SSIM value of 0.937 and PSNR of 40 dB. These results demonstrate that the proposed method effectively preserves image visual quality while improving watermark robustness against various digital image distortions.]]>
