Image Tampering Detection Based on Feature Fusion and Local Anomalies
The continuous development of digital image processing techniques has made the task of image tampe-ring detection increasingly challenging.In addition,different image sizes and operations such as sampling,convolution and pooling can also result in the loss of some spatial information and tampered features.In order to minimise the loss of feature information,the model in this paper preserves the low-level semantic information of the image by fusing the corresponding features of up sampling and down sampling and fusing the noise stream and constrained convolution stream with the convolution features of the model;secondly,a local abnormal region detection mechanism is construc-ted based on the segmentation model to focus on the abnormal regions of the image and reduce the complexity of the model;finally,attention is used to enhance the focus on the tampered features to solve the problem of unclear semantic information.The tampering detection method in this paper achieves optimal performance on three benchmark datasets and is tested against the latest methods,with F1 values improving by 15.9%on the CASIA dataset,4.5%on the NIST2016 dataset and 5.7%on the COVERAGE dataset.
万方数据
维普
