Copy Move Forgery Detection Network Based on Strip Self-correlation Calculation and Tamper Edge Attention
Self-correlation calculation is crucial in identifying highly similar regions in copy move forged images,and is an important technique for copy move forgery detection.Traditional self-correlation methods are limited by their large computational complexity.To solve this problem,strip self-correlation calculation was adopted,which effectively reduced the computational complexity and improved the detection efficiency by processing the feature map into blocks.Meanwhile,edge features play a crucial role in replication,movement,and tampering detection.Traditional methods often extract all edge information indiscriminately through general means such as networks or Sobel operators,but fail to consider the negative impact of irrelevant region edge features on detection accuracy.After introducing the tampering edge attention mechanism,the edge features are weighted by combining similarity information,highlighting the edge features of the tampered area and improving the accuracy of detection.In addition,during the feature fusion process,the downsampling convolution module can capture rich contextual information across scales,providing more useful information for detection.The experimental results show that the network combined with these modules performs well in the accuracy of copying,moving,and tampering detection,outperforming multiple existing methods.
deep learningcopy move forgery detectionstrip self-correlation calculationobject detection
万方数据
维普
