Pith. sign in

REVIEW 5 major objections 5 minor 22 references

An Efficient Digital Watermarking Technique for Small Scale devices

T0 review · 5 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A hybrid FWT-AQIM watermarking scheme claims real-time performance under 40 ms on a Raspberry Pi 5 with high robustness and imperceptibility.

desk verdict Modest engineering contribution with useful Pi 5 timings, but the headline robustness claims are contradicted by the paper's own tables and rest on a circular best-tile selection. read the letter →

arxiv 2506.06691 v1 pith:QZXEEOK7 submitted 2025-06-07 cs.MM cs.CR

Kaushik Talathi , Aparna Santra Biswas This is my paper
classification cs.MMcs.CR
keywords digitalwatermarkingFastWaveletTransformQuantizationIndexModulationIoTsmall-scaledevicesRaspberryPi5YCbCrcolorspacemosaicwatermark
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper tries to establish that a hybrid of the Fast Wavelet Transform (FWT) and Additive Quantization Index Modulation (AQIM) can embed and extract digital watermarks in under 40 milliseconds on a low-power device like the Raspberry Pi 5, while keeping the watermarked image nearly indistinguishable ($\mathrm{PSNR} \ge 34$ dB, $\mathrm{SSIM} \ge 0.97$) and resilient to common geometric and signal-processing attacks. The motivation is to provide lightweight, real-time watermarking for IoT, mobile, and other resource-constrained settings, where standard transforms like DWT-DCT or SVD are too computationally heavy. The scheme works in the YCbCr color space, embeds a mosaic watermark into the low-frequency FWT sub-band of the luminance channel, and reports near-zero bit-error rates with $\mathrm{NCC} \ge 0.998$ for QR-code watermarks, alongside a peak throughput of 11 MP/s and memory usage below about 170 MB for 512×512 images.

What carries the argument

The mechanism that carries the argument is the combination of FWT and AQIM. FWT is a fast multi-resolution decomposition that splits an image into approximation and detail sub-bands; embedding in the low-frequency approximation sub-band concentrates the watermark in perceptually important coefficients while limiting visual distortion. AQIM embeds each watermark bit by snapping a host coefficient to one of two interleaved quantization lattices, chosen by the bit value and a step size $\alpha$; the additive form applies a controlled shift without expensive norm computations. The watermark is replicated into a mosaic that matches the low-frequency sub-band dimensions, providing redundancy. At extraction, the best-matching mosaic tile is chosen by SSIM/NCC, which the paper says ensures accurate recovery under attacks.

What would settle it

Run the extraction algorithm on a watermarked 512×512 image while withholding the original host image; if the recovered watermark's BER rises well above the reported near-zero values and NCC drops below 0.998, the scheme's robustness claim is shown to depend on having the original image, a condition absent in many real-world uses.

Watch

Extended reading notes

Core claim

The central claim, put forward for a sympathetic reader, is that FWT-AQIM offers a practical balance of robustness, imperceptibility, and computational efficiency for small-scale devices. The method decomposes the luminance channel of an image with a fast wavelet transform, embeds the watermark into the low-frequency approximation coefficients using additive quantization index modulation with a controllable step size, and reconstructs the image. For extraction, the original and watermarked images are both decomposed and the watermark mosaic is recovered by applying the inverse AQIM rule and selecting the tile with the highest SSIM/NCC. In experiments on a Raspberry Pi 5, embedding and extraction each complete in about 30–40 ms for standard 512×512 images, with image quality staying above the reported thresholds and QR-code watermarks surviving cropping, rotation, scaling, JPEG, noise, and filtering with near-zero BER and NCC above 0.998.

Load-bearing premise

The extraction procedure requires the original, unwatermarked host image, so the reported near-zero error rates and NCC values apply only when the original is available at detection time, not in blind watermarking scenarios.

Editorial extensions

If this is right

  • Embedding and extraction complete in under 40 ms on a Raspberry Pi 5, with throughput up to 11 MP/s and memory usage around 160 MB for standard images, making real-time watermarking feasible on IoT-class hardware.
  • QR-code style watermarks (structured, error-correcting) survive cropping, rotation, scaling, JPEG, Gaussian noise, median filtering, and sandpaper noise with near-zero BER and NCC ≥ 0.998, supporting authentication and ownership applications.
  • Watermarking into the luminance low-frequency sub-band keeps imperceptibility high (PSNR ≥ 34 dB, SSIM ≥ 0.97), so watermarked images are visually indistinguishable from originals.
  • The mosaic redundancy and adaptive decomposition levels allow the method to scale to higher-resolution images (up to 2908×6000 in tests) while keeping processing under manageable times and memory use.
  • The speed advantage of about 0.03–0.04 s per operation over a comparison IoT watermarking method indicates the hybrid is competitive as a lightweight solution.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the original host image is typically not available at the detector (as in consumer devices), the scheme would need a blind extraction variant; otherwise the practical value is limited to settings where the owner retains and supplies the original.
  • The paper's robustness numbers are computed after picking the mosaic tile with the highest SSIM/NCC; a stricter evaluation that fixes the tile position or adds synchronization could lower the reported NCC and BER values.
  • The near-perfect results for QR-code watermarks likely derive from QR's built-in error correction; ordinary image watermarks show 9–12% BER even without attacks, so the headline robustness is not general.
  • A natural extension is to test against adversarial or deep-learning-based watermark removal, print-cam attacks, or combined attacks, which are absent from the current evaluation.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 5 minor

Summary. The paper proposes FWT-AQIM, a hybrid watermarking scheme combining Fast Wavelet Transform and Additive Quantization Index Modulation for resource-constrained devices. The method embeds a mosaic-based grayscale watermark (e.g., a 64×64 Lena image or a QR code) into the low-frequency sub-band of the Y channel in YCbCr space. Experiments on a Raspberry Pi 5 report PSNR ≥ 34 dB, SSIM ≥ 0.97, embedding/extraction under 40 ms, and high robustness, with near-zero BER and NCC ≥ 0.998 for QR watermarks. The paper claims that the method is lightweight, scalable, and suitable for real-time IoT content protection.

Significance. If the claims were reliable, a low-complexity watermarking method with good imperceptibility, high robustness, and fast operation on a Raspberry Pi 5 would be a useful contribution for IoT and mobile content protection. The paper addresses a relevant problem and provides a complete embedding/extraction pipeline. However, the central quantitative claims are contradicted by the paper's own tables, and the extraction procedure is non-blind and uses a selection step that inflates the reported robustness. No code or data are provided, and the comparison with prior work is incomplete. The potential significance is therefore not realized by the present version.

major comments (5)
  1. [§3.3] The extraction procedure is non-blind: Section 3.3 states that 'both the original and watermarked images undergo FWT decomposition.' This requires the original, unwatermarked host image at detection time, which is not available in the content-authentication and IoT deployment scenarios described in the abstract and introduction. The claimed practical applicability is therefore not supported.
  2. [§3.3] The reported robustness is inflated by oracle-like tile selection. Section 3.3 says the reconstructed watermark mosaic is evaluated using SSIM and NCC and that 'these metrics guide the selection of the best-matching watermark tile.' Since SSIM/NCC are computed against the known original watermark, the best tile is chosen with full knowledge of the ground truth, so the reported NCC and BER measure the best of many candidates rather than the fidelity of the extracted watermark. Table 2 shows the consequence: QR watermarks, whose structured redundancy benefits most from this selection, appear nearly perfect, while Lena watermarks degrade substantially.
  3. [Abstract and Tables 1–2] The abstract's blanket claims are contradicted by the paper's own data. Table 1 gives Cameraman QR SSIM = 0.9577, below the declared SSIM ≥ 0.97, and Tiffany Lena NCC = 0.8713, far below the declared NCC ≥ 0.998. Table 2 reports QR BER = 42.06% under 20% cropping (Airplane), 22.95% under scaling s=0.2, and 33.59% under sandpaper p=0.05. Additionally, the high-resolution Sunrise image takes 2.195 s for embedding in Table 1, directly contradicting the 'less than 40 ms' claim in the abstract and Section 5.
  4. [Table 3 and Section 4.3] The comparison in Table 3 is misleading. The 'Proposed Method' column appears to use best-case QR values from Table 1 rather than results under the same attack conditions and watermark type as the cited methods. No attack parameters, watermark type, or host images are specified for the proposed column, so the claimed superiority over Sk et al., Kumar Shrivastava et al., and Li et al. cannot be verified from the paper.
  5. [§4.2 and Table 1] The quantization step α and FWT decomposition level are chosen separately for each host–watermark combination (Table 1 lists α = 25, 30, 35, 60, 70, 90 and levels 2, 3, 4). The reported PSNR/BER values are therefore outcomes of per-case tuning, not an evaluation of the algorithm's default behavior. The paper does not describe a principled rule for selecting α or level, which limits the reproducibility and generality of the claimed performance.
minor comments (5)
  1. [Throughout] There are numerous typos and grammatical errors: 'Medial Filter attack' (§4.3.6), 'experimentation„' (§4), 'T est Img 1' (Table 2), 'p)= 0.01' (§4.3.8), and inconsistent capitalization of 'Quantization index Modulation' in the abstract. These should be corrected.
  2. [References] Reference [20] begins with '[1] Kim D-W, Kim J-K...' which is a formatting error.
  3. [Section 4.2] The text states 'SSIM scores exceeding 0.99' and 'PSNR around 37–38 dB' for the Lena watermark, but Table 1 lists several Lena rows with SSIM below 0.99 (e.g., Peppers 0.9962, Cameraman 0.9914) and QR rows with SSIM as low as 0.9577. The summary paragraph should be reconciled with the table.
  4. [Figure 2] The text references Fig. 2 for both embedding and extraction processes, but the figure appears to show only the embedding process. A separate extraction diagram would improve clarity.
  5. [Data availability] The Data Availability statement says 'The data will be available as on request' without providing a repository, protocol, or point of contact. Given the absence of code, numerical verification of the reported results is not possible.

Circularity Check

2 steps flagged · score 6.0 of 10

Reported robustness values are largely self-confirming: the extraction routine selects the best-matching mosaic tile using SSIM/NCC and then reports SSIM/NCC as the result, while α is hand-adjusted per experiment.

  1. self definitional [Section 3.3, 'Watermarking processes' (extraction description)]
    "Applying the inverse AQIM rule yields a reconstructed watermark mosaic, which is then evaluated using Structural Similarity Index (SSIM) and Normalized Cross-Correlation (NCC). These metrics guide the selection of the best-matching watermark tile, ensuring accurate recovery even under common attacks."

    The extracted watermark tile is chosen by maximizing SSIM/NCC against the original watermark, and then the same SSIM/NCC values (plus BER computed on the chosen tile) are reported as the robustness result. The reported NCC is therefore, by construction, the maximum over the candidate tiles rather than an independent extraction outcome. This directly manufactures the abstract's 'near-zero bit error rates and NCC ≥ 0.998' claim for QR watermarks: the selection procedure uses the ground-truth metric that is later used as the evaluation metric, making the headline robustness measurement self-confirming.

  2. fitted input called prediction [Section 3.3 and Section 4.2, Table 1 (quantization step selection)]
    "To adapt to different host–watermark combinations, users can select from multiple wavelet families and decomposition levels and adjust the quantization step α."

    Table 1 then reports PSNR, SSIM, BER, and NCC after choosing a different α for each host/watermark pair (e.g., α=30 for Lena watermarks, α=25 for QR watermarks, α=90 for large hosts). The reported quality and robustness numbers are thus outcomes of individually tuned parameters, not predictions of a fixed method. The headline guarantees (PSNR ≥ 34 dB, SSIM ≥ 0.97, near-zero BER/NCC ≥ 0.998) are conditional on these chosen settings, so part of the claimed performance is fitted input presented as measured performance.

full rationale

The paper does not rely on self-citations, imported uniqueness theorems, or machine-checked prior results; its computational claims (embedding/extraction time, memory, throughput) are not circular, though the abstract's 'less than 40 ms' is contradicted by Table 1 for high-resolution hosts. The central circularity is in the robustness evaluation. Section 3.3 states that SSIM and NCC are used to select the best-matching watermark tile; because those same metrics are the reported robustness scores, the near-zero BER and NCC ≥ 0.998 values are, for the QR watermark, the result of selecting the tile that best matches the known original watermark. This is a self-confirming measurement, not an independent extraction test. The per-experiment adjustment of the quantization step α (Table 1) further means the headline PSNR/SSIM/BER values describe tuned configurations rather than a parameter-free prediction. The non-blind requirement that the original image be available at extraction is a serious deployment limitation but is a correctness concern, not a circularity. Overall, the method has some independent content (speed, imperceptibility under chosen α), but the headline robustness numbers reduce by construction to the selection metric, justifying a partial-circularity score of 6.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central claims rest on two hand-chosen parameters (alpha and FWT level), several domain assumptions about wavelet-domain embedding, and two procedure choices: using the original image at extraction and selecting the best mosaic tile by the reported metric. No new entities are introduced. The lack of independent benchmarks or code makes the ledger heavier than the prose suggests.

free parameters (2)
  • Quantization step alpha = 25, 30, 35, 60, 70, 90 (per host/watermark)
    Chosen by hand for each host/watermark to balance PSNR and BER; the headline metrics depend on this value.
  • FWT decomposition level = 2, 3, or 4 depending on image
    Set adaptively per image; robustness, embedding capacity, and timing all depend on the level.
assumptions (4)
  • standard math Fast Wavelet Transform is an invertible multiresolution transform with energy compaction.
    The scheme assumes FWT decomposition and reconstruction preserve image content well enough for watermark embedding (Section 3.1).
  • domain assumption Low-frequency coefficients of the luminance (Y) channel are suitable carriers for invisible watermarks.
    The method embeds only in low-frequency Y sub-bands to minimize perceptual distortion (Section 3.3); this is a heuristic, not proven.
  • ad hoc to paper The original host image is available at extraction time.
    Section 3.3 states both original and watermarked images undergo FWT during extraction, making the scheme non-blind. No justification is given for why this is acceptable in target IoT scenarios.
  • ad hoc to paper Selecting the mosaic tile with the highest NCC/SSIM gives an unbiased measure of extraction quality.
    Section 3.3 says these metrics 'guide the selection of the best-matching watermark tile,' so the reported NCC is an optimized maximum, not an independent measurement.

how reviews work

0 comments
Cite this review

Pith. "Pith review of An Efficient Digital Watermarking Technique for Small Scale devices." pith.science (2026). https://pith.science/paper/QZXEEOK7

@misc{pith2026250606691,
  author       = {Pith},
  title        = {Pith review of: An Efficient Digital Watermarking Technique for Small Scale devices},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QZXEEOK7}},
  note         = {Machine review of arXiv:2506.06691}
}
read the original abstract

In the age of IoT and mobile platforms, ensuring that content stay authentic whilst avoiding overburdening limited hardware is a key problem. This study introduces hybrid Fast Wavelet Transform & Additive Quantization index Modulation (FWT-AQIM) scheme, a lightweight watermarking approach that secures digital pictures on low-power, memory-constrained small scale devices to achieve a balanced trade-off among robustness, imperceptibility, and computational efficiency. The method embeds watermark in the luminance component of YCbCr color space using low-frequency FWT sub-bands, minimizing perceptual distortion, using additive QIM for simplicity. Both the extraction and embedding processes run in less than 40 ms and require minimum RAM when tested on a Raspberry Pi 5. Quality assessments on standard and high-resolution images yield PSNR greater than equal to 34 dB and SSIM greater than equal to 0.97, while robustness verification includes various geometric and signal-processing attacks demonstrating near-zero bit error rates and NCC greater than equal to 0.998. Using a mosaic-based watermark, redundancy added enhancing robustness without reducing throughput, which peaks at 11 MP/s. These findings show that FWT-AQIM provides an efficient, scalable solution for real-time, secure watermarking in bandwidth- and power-constrained contexts, opening the way for dependable content protection in developing IoT and multimedia applications.

Figures

Figures reproduced from arXiv: 2506.06691 by the authors.

Figure 1
Figure 1. Multiple Levels FWT Decomposition 3.2. Quantization Index Modulation (QIM) QIM is a well-known watermarking technique that maps host signal co￾efficients onto one of two interleaved quantization lattices, each of which represents a binary character, in order to encode information. The organized lattice arrangement guarantees resilience against common signal degrada￾tions, while the method provides imperceptibility w… view at source ↗
Figure 2
Figure 2. FWT-AQIM based watermarking process For extraction, both the original and watermarked images undergo FWT decomposition to retrieve their low-frequency coefficients. Applying the in￾verse AQIM rule yields a reconstructed watermark mosaic, which is then evaluated using Structural Similarity Index (SSIM) and Normalized Cross￾Correlation (NCC). These metrics guide the selection of the best-matching watermark tile, ensur… view at source ↗
Figure 3
Figure 3. demonstrates the cropping attack on standard and non-standard host images with different crop ratios. BER increases from around 11–15 % to 37 % as the crop ratio r increases from 3% to 20%, while NCC decreases from 0.99 to 0.28 for Lena watermarks.The QR-code watermark shows superior robustness, with BER remaining near 0% and NCC > 0.55 until an extreme crop ratio [PITH_FULL_IMAGE:figures/full_fig_p015_3.png] view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: Illustration of the rotation attack 4.3.3. Scaling Attack The image is first resized by a factor s (up- or down-sampled) and then scaled back to its original dimensions using linear interpolation. This dou￾ble resampling introduces smoothing, interpolation blur, and an…
Figure 5
Figure 5. Figure 5: Illustration of the scaling attack 15 [PITH_FULL_IMAGE:figures/full_fig_p016_5.png]
Figure 6
Figure 6. Figure 6: Illustration of the gaussian noise attack [PITH_FULL_IMAGE:figures/full_fig_p017_6.png]
Figure 7
Figure 7. Figure 7: Illustration of the JPEG compression attack [PITH_FULL_IMAGE:figures/full_fig_p018_7.png]
Figure 8
Figure 8. Figure 8: Illustration of the median filter attack [PITH_FULL_IMAGE:figures/full_fig_p018_8.png]
Figure 9
Figure 9. Figure 9: Illustration of the resize attack 4.3.8. Sandpaper Noise (Salt-and-Pepper) Attack Sandpaper attack randomly sets pixels to black or white with probability p, modeling impulse noise from transmission errors or dust. For p) = 0.01, half a percent of all pixels are set to…
Figure 10
Figure 10. Figure 10: Illustration of the sandpaper attack [PITH_FULL_IMAGE:figures/full_fig_p020_10.png]
Figure 11
Figure 11. Figure 11: Graphical representation of various attacks for standard and non-standard [PITH_FULL_IMAGE:figures/full_fig_p022_11.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

22 extracted references · 19 canonical work pages

  1. [1]

    Information 2020;11:110

    Begum M, Uddin MS.Digital Image Watermarking Techniques: A Re- view. Information 2020;11:110. https://doi.org/10.3390/info11020110

    work page doi:10.3390/info11020110 2020

  2. [2]

    Computers & Electrical Engineering 2018;72:589–605

    Sk A, Masilamani V.A novel digital watermarking scheme for data authentication and copyright protection in 5G net- works. Computers & Electrical Engineering 2018;72:589–605. https://doi.org/10.1016/j.compeleceng.2018.02.045

    work page doi:10.1016/j.compeleceng.2018.02.045 2018

  3. [3]

    In: Tuba M, Akashe S, Joshi A, editors

    Kumar A.A Review on Implementation of Digital Image Watermark- ing Techniques Using LSB and DWT. In: Tuba M, Akashe S, Joshi A, editors. Information and Communication Technology for Sustainable Development, vol.933, Singapore: SpringerSingapore; 2020, p.595–602. https://doi.org/10.1007/978-981-13-7166-0_59

    work page doi:10.1007/978-981-13-7166-0_59 2020

  4. [4]

    PEN 2022;10:283

    Al-Kadei FHMS, Hasan SN.Improve a secure blind wa- termarking technique for digital video. PEN 2022;10:283. https://doi.org/10.21533/pen.v10i2.2859. 24

    work page doi:10.21533/pen.v10i2.2859 2022

  5. [5]

    International Journal of Information Management 2019;45:262–75

    Kaw JA, Loan NA, Parah SA, et al.A reversible and se- cure patient information hiding system for IoT driven e-health. International Journal of Information Management 2019;45:262–75. https://doi.org/10.1016/j.ijinfomgt.2018.09.008

    work page doi:10.1016/j.ijinfomgt.2018.09.008 2019

  6. [6]

    https://doi.org/10.18178/joig.11.1.61-71

    Reyes-Ruiz L, Fragoso-Navarro D, Garcia-Ugalde F, et al.Ro- bust Dual Digital Watermark Applied to Antique Digitized Cin- ema Images: Resistant to Print-Scan Attack, JOIG 2023;11:61–71. https://doi.org/10.18178/joig.11.1.61-71

    work page doi:10.18178/joig.11.1.61-71 2023

  7. [7]

    Signal Processing 2019;164:386–401

    Yao Y, Zhang W, Wang H, et al.Content-adaptive reversible visible watermarking in encrypted images. Signal Processing 2019;164:386–401. https://doi.org/10.1016/j.sigpro.2019.06.034

    work page doi:10.1016/j.sigpro.2019.06.034 2019

  8. [8]

    Signal Processing 2025;234:109999

    Xie X, Jiang J, Zhang J, et al.Reversible adversarial vis- ible image watermarking. Signal Processing 2025;234:109999. https://doi.org/10.1016/j.sigpro.2025.109999

    work page arXiv 2025

Show all 22 references
  1. [9]

    Multimed Tools Appl 2020;79:20149–97

    Kumar S, Singh BK, Yadav M.A Recent Survey on Multimedia and Database Watermarking. Multimed Tools Appl 2020;79:20149–97. https://doi.org/10.1007/s11042-020-08881-y

    2020 doi

  2. [10]

    Kapse AS, Belokar S, Gorde Y, et al.Digital Image Security Using Dig- ital WatermarkingIRJET 2018:5(3):163–166

    2018

  3. [11]

    Microsyst Technol 2018;24:2539–48

    Gupta G, Gupta VK, Chandra M.An efficient video water- marking based security model. Microsyst Technol 2018;24:2539–48. https://doi.org/10.1007/s00542-017-3689-x. 25

    2018 doi

  4. [12]

    Mul- timed Tools Appl 2022;81:15961–75

    FkirinA,AttiyaG,El-SayedA,etal.Copyright protection of deep neural network models using digital watermarking: a comparative study. Mul- timed Tools Appl 2022;81:15961–75. https://doi.org/10.1007/s11042- 022-12566-z

    2022 doi

  5. [13]

    Int J Onl Eng 2023;19:155–73

    Hussein Tuama Hazim, Nawar Alseelawi, ALRikabi HTHS.A Novel Method of Invisible Video Watermarking Based on Index Mapping and Hybrid DWT-DCT. Int J Onl Eng 2023;19:155–73. https://doi.org/10.3991/ijoe.v19i04.37581

    2023 doi

  6. [14]

    2018 10th International Conference on Knowledge and Smart Technology (KST), Chiang Mai: IEEE; 2018, p

    Panyavaraporn J, Horkaew P.DWT/DCT-based Invisible Digital Water- marking Scheme for Video Stream. 2018 10th International Conference on Knowledge and Smart Technology (KST), Chiang Mai: IEEE; 2018, p. 154–7. https://doi.org/10.1109/KST.2018.8426150

    2018

  7. [15]

    Watermarking Techniques for Copy- right Protection of Videos

    Kothari AM, Dwivedi V, Thanki RM.Singular Value Decomposition (SVD)-Based Video Watermarking. Watermarking Techniques for Copy- right Protection of Videos. Cham: Springer International Publishing; 2019, p. 63–80. https://doi.org/10.1007/978-3-319-92837-1_4

    2019 doi

  8. [16]

    IET Image Processing 2013;7:432–41

    Zareian M, Tohidypour HR.Robust quantisation index modulation-based approach for image watermarking. IET Image Processing 2013;7:432–41. https://doi.org/10.1049/iet-ipr.2013.0048

    2013

  9. [17]

    From Fourier Analysis to Wavelets, Cham: Springer International Publishing; 2015, p

    Gomes J, Velho L.The Fast Wavelet Transform. From Fourier Analysis to Wavelets, Cham: Springer International Publishing; 2015, p. 89–100. https://doi.org/10.1007/978-3-319-22075-8_7. 26

    2015 doi

  10. [18]

    Vis Comput 2024;40:345–72

    Arya AS, Saha A, Mukhopadhyay S.ADMM optimizer for integrating wavelet-patch and group-based sparse representation for image inpaint- ing. Vis Comput 2024;40:345–72. https://doi.org/10.1007/s00371-023- 02786-1

    2024 doi

  11. [19]

    SN COMPUT SCI 2022;3:107

    Vaidya SP, Kishore VR.Adaptive Medical Image Watermarking Sys- tem For E-Health Care Applications. SN COMPUT SCI 2022;3:107. https://doi.org/10.1007/s42979-021-00995-w

    2022 doi

  12. [20]

    Digital Signal Processing2024;146:104352.https://doi.org/10.1016/j.dsp.2023.104352

    [1] Kim D-W, Kim J-K, Piao Z, et al.Transient detection-based adap- tive audio watermarking using attack-aware optimization. Digital Signal Processing2024;146:104352.https://doi.org/10.1016/j.dsp.2023.104352

    2023

  13. [21]

    Mahendran D.Digital Watermarking using Lifting Wavelet Transform of Crowd Sourced ImagesIJET 2018;7:357

    Kumar Shrivastava S, S.K. Mahendran D.Digital Watermarking using Lifting Wavelet Transform of Crowd Sourced ImagesIJET 2018;7:357. https://doi.org/10.14419/ijet.v7i2.20.16733

    2018 doi

  14. [22]

    https://doi.org/10.1016/j.dsp.2021.103062

    Li Z, Zhang H, Liu X, et al.Blind and safety-enhanced dual watermarking algorithm with chaotic system encryption based on RHFM and DWT-DCTDigital Signal Processing 2021;115:103062. https://doi.org/10.1016/j.dsp.2021.103062. 27

    2021

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.