Block-level temporal anomaly detection and auxiliary positioning method and system for video tampering

By using block-level temporal anomaly detection and MS-SSIM assisted localization mechanism, the shortcomings of existing video tampering detection methods in type detection and location localization are solved, achieving high accuracy and interpretability in video tampering detection, and is applicable to scenarios such as network content review and judicial evidence collection.

CN122115341APending Publication Date: 2026-05-29SOUTHEAST UNIV +1
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
SOUTHEAST UNIV
Filing Date
2026-01-30
Publication Date
2026-05-29

AI Technical Summary

Technical Problem

Existing video tampering detection methods struggle to simultaneously and accurately detect the type of tampering and pinpoint the specific location where it occurred. Furthermore, deep learning-based methods lack interpretability and the ability to perceive localized tampering behavior.

Method used

A block-level temporal anomaly detection method is adopted, which combines global semantic features and local block-level temporal features for joint modeling. The MS-SSIM assisted localization mechanism is introduced, and temporal modeling is performed through a multi-layer attention structure and a feedforward network. Combined with multi-scale structural similarity analysis, the tampered frame can be located.

Benefits of technology

It improves the accuracy and interpretability of video tampering detection, and can output the location and time range of suspected tampered frames, making it suitable for application scenarios such as network content review and judicial evidence collection.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122115341A_ABST
    Figure CN122115341A_ABST
Patent Text Reader

Abstract

The application discloses a block-level time sequence anomaly detection and auxiliary positioning method and system for video tampering. After preprocessing the video frames, the method extracts global semantic features and local block features; image blocks at the same position in space are aggregated along the time sequence to obtain block-level time sequence features; and local and global features are fused to determine the tampering type. To locate the tampering position, sudden change detection and threshold determination are performed based on the multi-scale structural similarity and its change rate of adjacent frames to obtain a suspected tampering frame interval, which is combined with the classification result to output. The corresponding system includes video input preprocessing, feature extraction, block-level time sequence modeling, feature fusion classification, similarity auxiliary positioning and result output modules. The application takes into account type discrimination and time positioning, improves detection accuracy and interpretability, and is suitable for content review and digital forensics.
Need to check novelty before this filing date? Find Prior Art