Model training method, fake image detection method and device
By constructing a training sample set and adjusting the parameters of the forgery detection model using the triplet loss function, the problem of poor forgery image detection performance in existing technologies is solved. This enables accurate differentiation and identification of forgery regions in highly similar images, improving the robustness and generalization of the model.
CN121767786BActive Publication Date: 2026-06-26北京万方数据股份有限公司 +1
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- 北京万方数据股份有限公司
- Filing Date
- 2026-03-02
- Publication Date
- 2026-06-26
Smart Images
Figure CN121767786B_ABST
Abstract
The method comprises the following steps: acquiring a training sample set; the training sample set comprises a plurality of sample images and a sample label of each sample image; the sample label of each sample image comprises a real forgery identifier corresponding to each pixel point in the sample image; performing at least one training operation on an initial forgery detection model based on the training sample set, and taking the initial forgery detection model meeting a preset training end condition as a trained forgery detection model. The method can forcibly increase the difference between the predicted features of the real region and the predicted features of the forged region, effectively improve the distinguishing effect in the high-similarity image, and be beneficial to improving the detection effect in the high-similarity academic image forgery detection scene.
Need to check novelty before this filing date? Find Prior Art