Codec-conditioned Image Hiding under a Matched Training Budget: Evidence from JPEG, WebP, HEIF and AVIF Recompression

Deep image hiding is usually hardened against JPEG alone, whereas shared images are re-encoded as JPEG, WebP, HEIF or AVIF, often repeatedly.Conditioning the hiding network on the channel codec is an intuitive remedy, tested here under a matched step budget.A codec-conditioned transformer and its unconditioned predecessor, each trained under its own recipe for 30,000 steps on identical data and channels with three seeds, are compared on 1,000 paired images through real local encoders, without resizing or platform processing.The predecessor is superior on both axes: cover fidelity 25.62 ± 0.13 against 22.14 ± 1.62 dB, and secret recovery after JPEG at quality 80 of 27.62 ± 0.21 against 22.03 ± 0.10 dB, with run-level differences of 5.5 to 6.7 dB over ten channel conditions.The predecessor's optimiser settings do not improve the conditioned model, and its complete recipe collapses it to a trivial solution.Ablating the conditioning changes recovery by at most 0.35 dB, and declaring the correct codec instead of a wrong one changes it by +0.12 dB on average.Cover fidelity peaks early and then falls by 2.10 to 3.26 dB in 9 of 14 conditioned runs.A steganalysis detector identifies every model with accuracy above 0.98, so none of the models is covert.

Authors

Institutions

Publication Details

Journal
International journal of intelligent engineering and systems
Published
2026-11-17
DOI
https://doi.org/10.22266/ijies2026.1231.47
Primary Topic
Advanced Steganography and Watermarking Techniques
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Codec-conditioned Image Hiding under a Matched Training Budget: Evidence from JPEG, WebP, HEIF and AVIF Recompression

Ahmed Al-Shammari, Mohammed Tali Almalchy, Murtadha A. Alazzawi, Mustafa Sabah Taha et al.
International journal of intelligent engineering and systems
Advanced Steganography and Watermarking Techniques
article

Codec-conditioned Image Hiding under a Matched Training Budget: Evidence from JPEG, WebP, HEIF and AVIF Recompression

Ahmed Al-Shammari, Mohammed Tali Almalchy, Murtadha A. Alazzawi, Mustafa Sabah Taha, Imad Ali Hasoon
article en

Abstract

Deep image hiding is usually hardened against JPEG alone, whereas shared images are re-encoded as JPEG, WebP, HEIF or AVIF, often repeatedly.Conditioning the hiding network on the channel codec is an intuitive remedy, tested here under a matched step budget.A codec-conditioned transformer and its unconditioned predecessor, each trained under its own recipe for 30,000 steps on identical data and channels with three seeds, are compared on 1,000 paired images through real local encoders, without resizing or platform processing.The predecessor is superior on both axes: cover fidelity 25.62 ± 0.13 against 22.14 ± 1.62 dB, and secret recovery after JPEG at quality 80 of 27.62 ± 0.21 against 22.03 ± 0.10 dB, with run-level differences of 5.5 to 6.7 dB over ten channel conditions.The predecessor's optimiser settings do not improve the conditioned model, and its complete recipe collapses it to a trivial solution.Ablating the conditioning changes recovery by at most 0.35 dB, and declaring the correct codec instead of a wrong one changes it by +0.12 dB on average.Cover fidelity peaks early and then falls by 2.10 to 3.26 dB in 9 of 14 conditioned runs.A steganalysis detector identifies every model with accuracy above 0.98, so none of the models is covert.

International journal of intelligent engineering and systemsVol. 19(12)
University of Misan (IQ), University of Al-Qadisiyah (IQ), Imam Ja’afar Al-Sadiq University (IQ), Imam Alkadhim University College (IQ)
Advanced Steganography and Watermarking Techniques
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon