Image tampering detection method based on feature cascade fusion
To address the issue of ineffective handling features at different scales in the field of image tampering detec-tion,a feature cascade fusion detection network is proposed.Firstly,the network utilizes a feature cascade fusion module combined with a U-shaped segmentation network structure,effectively integrating features of various scales.By merging shallow,bottleneck,and deep features at each network block,the network compensates for the deficiency of deep semantic information,suppresses background interference,and enhances the detection capability of the shallow network,achieving precise localization of tampered regions.Experimental results indicate that compared to existing image tampering detection methods,this feature cascade fusion detection network exhibits higher accuracy and stability.It achieved a 3％improvement in F-measure on the CASIA dataset and a 4％increase on the COLUMB dataset,providing evidence of its effectiveness in image tampering detection tasks.
NETL
NSTL
万方数据
维普
