Penelusuran Asal dan Deteksi Manipulasi Citra pada Kanal Whatsapp Menggunakan Watermark Ganda DWT dan Tanda Tangan Pascakuantum Ml-DSA-44

This Abstract has been read 0 times

Abstract

Images sent through WhatsApp lose their metadata and are recompressed, so a hash or a companion signature file can no longer prove their origin and integrity. This article proposes a dual-watermark scheme in the Discrete Wavelet Transform (DWT) domain: a 32-bit robust identifier that traces back to the issuer's certificate, and a semi-fragile 256-bit fingerprint per 32×32-pixel block compared through Bit Error Rate (BER) to localize manipulation blindly. The certificate is signed with the post-quantum signature ML-DSA-44 (FIPS 204) and stored on a server, while the pixels carry only the identifier. In a real WhatsApp Web transfer of 60 images, the identifier is read and the signature is valid for all of them, no intact image is judged malicious (0/48) while hash matching rejects all of them, and local manipulation is detected in 9 of 12 images. On the mobile app, compression is heavier (BER about 0.28): the identifier is still read in 10 of 10 images, no intact image is judged malicious (0/7), but the three manipulated images are not detected either (0/3). Hence image origin can be proven over WhatsApp, whereas manipulation detection has only been measured under light compression and needs per-channel threshold calibration. On 200 Chest X-Ray images without a channel, localization precision is 1.0000 and recall 0.99995 at a PSNR of 32.9 dB. An ablation on 48 uncompressed images shows that the baseline BER of intact images comes from 8-bit rounding and pixel clipping in very dark or very bright areas, not from the fingerprint itself.

Keywords: Dual Watermark, Image Authentication, WhatsApp, Discrete Wavelet Transform, ML-DSA