Document image tampering detection faces significant challenges due to the subtle and spatially dispersed nature of tampering traces, which are often confined to localized regions within tampered text. While existing methods leverage frequency domain information to reveal hidden artifacts, they fail to fully exploit the rich frequency spectrum and lack effective mechanisms for aggregating scattered tampering evidence across extended text regions. To overcome these limitations, we propose the Text Aggregation and multi-Frequency Enhancement Network (TAFE-Net). Specifically, to capture more subtle tampering traces, we design a Multi-Frequency Feature Extractor that comprehensively utilizes various proven effective frequency information. In addition, the Visual-Frequency Integration Module and Direction-aware Frequency Decoupling Enhancement module are introduced to aggregate text features in both horizontal and vertical directions within the frequency domain, from coarse to fine granularity, addressing the incomplete detection of tampered text caused by dispersed tampering traces. Experiments on the DocTamper and RTM datasets demonstrate that our approach establishes new state-of-the-art results and maintains superior robustness against various degradations.

The spread of tampered text poses a critical challenge to information security. Previous methods for tampered text detection (TTD) primarily relied on visual artifacts as clues, while overlooking potential semantic inconsistencies introduced during text manipulation. To address this limitation, we propose TVSIP (Tampered text Visual-Semantic InterPreter), a novel framework leveraging Multimodal Large Language Models (MLLMs) to integrate both visual and semantic clues for comprehensive tampered text analysis and verification. TVSIP consists of a Locator and an Interpreter. The Locator combines the visual detection ability of existing expert models with the semantic comprehension capabilities of MLLMs to create precise tampering masks. Subsequently, the Interpreter provides comprehensive descriptions and explanations based on identified tampered regions. To train and evaluate TVSIP, we construct the TextDDLE benchmark using GPT-4o. Extensive experiments demonstrate that TVSIP outperforms expert models in pixel-level localization and advanced MLLMs in interpretability. Furthermore, it maintains robustness against image degradation and exhibits strong generalization ability on out-of-domain datasets. Our work highlights the crucial role of semantic inconsistencies in TTD and establishes a more reliable verification system for ensuring document authenticity in the digital age.