Journal: Scientific Reports
Article Title: MSCSCC-Net: multi-scale contextual spatial-channel correlation network for forgery detection and localization of JPEG-compressed image
doi: 10.1038/s41598-025-97555-6
Figure Lengend Snippet: Overall Architecture of the proposed MSCSCC-Net. Deep feature extraction module extracted different scales features from input JPEG-compressed image and then used for forgery detection and localization. The detection head determines if the image is forged based on the prediction score. As we move from Mask 4 to Mask 1, the precision of forgery localization increases. For instance, Mask 1 corrects Mask 4’s prediction that confuses the forged area with the copied one.
Article Snippet: To add JPEG-compressed artifacts, we use the above test datasets to generate JPEG-compressed images by the MATLAB JPEG encoder, and each image’s quality factor ranges uniformly from 10 to 100 in increments of 10. (2) Metrics: In accordance with earlier research , we compute the PSNR, SSIM, and PSNR-B for a quantitative evaluation of the restored image to compare JPEG artifact removal performance.
Techniques: Extraction