DL-CMFD: Deep Learning-Based Copy-Move Forgery Detection Using Parallel Feature-Extractor
Layeba Faheem, Soumya Mukherjee, Mohammad S. Obaidat, Arup Kumar Pal, SK Hafizul Islam, Balqies Sadoun · 2023
Due to readily available software, anyone can easily manipulate the image and create a piece of false information. Among different types of image forgery, copy-move image forgery is widely practiced due to its sophistication and difficulty distinguishing it from the original image. We proposed a new deep learning-based copy-move forgery detection (DL-CMFD) technique to identify the forged image by successfully localizing the forged area to address this challenge. To attain this, the scale-invariant features are extracted from the forged image using the parallel feature extraction technique utilizing a batch normalization-based convolution neural network with the aid of the ReLU activation function. In the next step, to identify the identical features, self-correlation has been calculated, and further, percentile pooling is used to find the top k-percent features. Lastly, a masked image is produced as the output. The proposed scheme has been deployed on the different available datasets like CoMoFD and COCO to testify to the efficiency and accuracy. The proposed scheme is found to be robust because it can detect the forged area accurately when deployed on post-processed forged images. We made various statistical parametric evaluations to evaluate the efficacy and conformity of the proposed copy-move forgery detection scheme, and a comparison analysis is also made.