<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="wang-etal-2026-token">
    <titleInfo>
        <title>Token-to-Pixel Semantic Guidance for Document Image Tampering Localization</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Tiancheng</namePart>
        <namePart type="family">Wang</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Haoyu</namePart>
        <namePart type="family">Huang</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Zichao</namePart>
        <namePart type="family">Feng</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Linlin</namePart>
        <namePart type="family">Yang</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Sheng</namePart>
        <namePart type="family">Xu</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Baochang</namePart>
        <namePart type="family">Zhang</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <originInfo>
        <dateIssued>2026-10</dateIssued>
    </originInfo>
    <typeOfResource>text</typeOfResource>
    <relatedItem type="host">
        <titleInfo>
            <title>Findings of the Association for Computational Linguistics: EMNLP 2026</title>
        </titleInfo>
        <name type="personal">
            <namePart type="given">Hung-Yi</namePart>
            <namePart type="family">Lee</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Sunipa</namePart>
            <namePart type="family">Dev</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Jessy</namePart>
            <namePart type="family">Li</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Desmond</namePart>
            <namePart type="family">Elliott</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <originInfo>
            <publisher>Association for Computational Linguistics</publisher>
            <place>
                <placeTerm type="text">Budapest, Hungary</placeTerm>
            </place>
        </originInfo>
        <genre authority="marcgt">conference publication</genre>
        <identifier type="isbn">979-8-89176-462-0</identifier>
    </relatedItem>
    <abstract>Document image tampering poses a serious threat to information authenticity in security-critical documents. Existing document tampering detection methods mainly rely on RGB-domain and frequency-domain artifacts, following a paradigm similar to natural image forensics, and often overlook a key property of document images: dense and semantically rich textual content. In this work, we propose TSPG, a plug-and-play Textual Semantic Prior Guidance module for document image tampering detection. TSPG uses a pretrained language model with a lightweight token classification head to estimate token-level tampering-aware semantic scores from OCR-recognized text. It then projects these scores back to the image space through OCR bounding boxes to construct a spatially aligned Textual Semantic Prior (TSP) map. To inject aforementioned prior into existing detectors, we design a Gated Semantic Residual Modulation (GSRM) module that adaptively modulates multi-scale detector features through residual gated guidance. TSPG can be inserted into existing document tampering detectors without modifying their original backbone or decoder. Experiments on DocTamper with two representative baselines, DTD and FFDN, show that TSPG consistently improves both models, demonstrating the complementary value of language-derived textual semantics for document tampering localization.</abstract>
    <identifier type="citekey">wang-etal-2026-token</identifier>
    <location>
        <url>https://aclanthology.org/2026.findings-emnlp.1859/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>39686</start>
            <end>39696</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T Token-to-Pixel Semantic Guidance for Document Image Tampering Localization
%A Wang, Tiancheng
%A Huang, Haoyu
%A Feng, Zichao
%A Yang, Linlin
%A Xu, Sheng
%A Zhang, Baochang
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Findings of the Association for Computational Linguistics: EMNLP 2026
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-462-0
%F wang-etal-2026-token
%X Document image tampering poses a serious threat to information authenticity in security-critical documents. Existing document tampering detection methods mainly rely on RGB-domain and frequency-domain artifacts, following a paradigm similar to natural image forensics, and often overlook a key property of document images: dense and semantically rich textual content. In this work, we propose TSPG, a plug-and-play Textual Semantic Prior Guidance module for document image tampering detection. TSPG uses a pretrained language model with a lightweight token classification head to estimate token-level tampering-aware semantic scores from OCR-recognized text. It then projects these scores back to the image space through OCR bounding boxes to construct a spatially aligned Textual Semantic Prior (TSP) map. To inject aforementioned prior into existing detectors, we design a Gated Semantic Residual Modulation (GSRM) module that adaptively modulates multi-scale detector features through residual gated guidance. TSPG can be inserted into existing document tampering detectors without modifying their original backbone or decoder. Experiments on DocTamper with two representative baselines, DTD and FFDN, show that TSPG consistently improves both models, demonstrating the complementary value of language-derived textual semantics for document tampering localization.
%U https://aclanthology.org/2026.findings-emnlp.1859/
%P 39686-39696

Markdown (Informal)

[Token-to-Pixel Semantic Guidance for Document Image Tampering Localization](https://aclanthology.org/2026.findings-emnlp.1859/) (Wang et al., Findings 2026)

• Token-to-Pixel Semantic Guidance for Document Image Tampering Localization (Wang et al., Findings 2026)

ACL

• Tiancheng Wang, Haoyu Huang, Zichao Feng, Linlin Yang, Sheng Xu, and Baochang Zhang. 2026. Token-to-Pixel Semantic Guidance for Document Image Tampering Localization. In Findings of the Association for Computational Linguistics: EMNLP 2026, pages 39686–39696, Budapest, Hungary. Association for Computational Linguistics.