Discover Networks Computer science
Network aware blind EEG data hiding using quantization index modulation
Abstract
Reliable data hiding in electroencephalography (EEG) needs blind payload recovery, high signal fidelity, and an embedding strategy that considers multichannel structure. This study proposes a reliability-adaptive blind DCT–RS–QIM framework combining an orthonormal temporal discrete cosine transform, Reed–Solomon coding, dithered scalar quantization index modulation, fixed carrier-reliability profiling, and bounded synchronization search. For multichannel EEG, a build-only Pearson graph and its Laplacian eigenbasis direct embedding toward selected graph-smooth components; the single-channel configuration uses the same coding, embedding, and synchronization architecture without a multichannel graph. Evaluation covered fixed-profile capacity, clean fidelity, payload recovery, graph-related structural measures, negative decoder controls, steganalysis, and runtime. At 45 B, clean exact recovery reached 100% for 400 Bonn records and 90 Graz A records. Mean normalized PSNR was 43.7510 ± 1.8222 dB for Bonn and 55.5594 ± 2.5057 dB for Graz Graph. Relative to the all-channel PSNR-matched Direct control, Direct achieved slightly higher exact recovery (83.016% versus 80.159%) and lower pre-ECC BER (3.9497% versus 5.6406%). The primary sensor-space measures did not show a clear Graph advantage. Graph produced lower values on specific alternate-graph and method-aligned distortion measures, but these effects are metric-specific and do not establish general topology superiority; the lower high graph-frequency fraction is basis-dependent and expected from the Graph design. Graph met the predefined subject-paired non-inferiority criterion on the seven-attack endpoint across the same nine subjects but had higher computational cost. Bounded synchronization enabled exact recovery under the tested circular shifts, and no false-valid frames were observed in the evaluated negative decoder controls. The framework therefore provides a reproducible blind data-hiding approach under an explicitly defined receiver model and a controlled Graph–Direct comparison with mixed findings.