Picture detection method, device and electronic equipment

By reconstructing the error map and extracting spatial, edge, and frequency domain features in parallel, and combining it with global average pooling, the problem of insufficient multi-dimensional artifact capture capability of traditional image detection methods is solved, thereby improving the accuracy and reliability of detection.

CN122156936APending Publication Date: 2026-06-05BEIJING UNIV OF POSTS & TELECOMM
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
BEIJING UNIV OF POSTS & TELECOMM
Filing Date
2026-01-12
Publication Date
2026-06-05

AI Technical Summary

Technical Problem

Traditional image detection methods rely on fixed feature extraction strategies, which cannot effectively capture artifacts in multiple feature domains of different image generation models, resulting in low detection accuracy and insufficient generalization ability, especially in the identification of highly realistic forged images, where the results are unreliable.

Method used

By reconstructing the preprocessed image to generate a reconstruction error map, and combining spatial domain, edge domain and frequency domain features, multi-dimensional feature fusion and global average pooling are adopted to achieve multi-dimensional representation and information interaction of the image, thereby improving the accuracy and reliability of classification decisions.

Benefits of technology

It effectively suppresses noise interference from single feature sources, reduces the probability of false positives and false negatives, and enhances the robustness and accuracy of the detection system.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122156936A_ABST
    Figure CN122156936A_ABST
Patent Text Reader

Abstract

The application relates to the technical field of computer vision, in particular to a picture detection method and device and electronic equipment, wherein the method comprises the following steps: performing center cropping preprocessing on an input image; reconstructing the image and calculating a reconstruction error map; respectively extracting spatial domain features, edge domain features and frequency domain features from the reconstruction error map; intelligently fusing the multi-dimensional features by using a hierarchical cross-attention fusion mechanism; and inputting the fused features into a classifier for true-false binary classification. The application fully mines discriminative information in the reconstruction error by decoupling multi-dimensional features, and intelligently fuses the features by using the hierarchical cross-attention mechanism, thereby enhancing the detection accuracy and generalization ability of the model.
Need to check novelty before this filing date? Find Prior Art