ForgNetwork: An adaptive forgery detection and localization framework using stationary wavelet transform and dilated adaptive VGG16 with optimization strategy

Abstract

Recently, the ability to alter images has become remarkably accessible, allowing for effortless manipulation of an image’s semantic content using widely available editing tools and techniques. These techniques are also called journaling tools, and it is used to effectively detect the changes in images. In the image processing technique, the deviations in images are determined by using typically square, slide regular, and artifacts techniques. But, the determination of these changes in images is a tedious procedure. To determine the forgery activities, a suitable method should be developed with the help of wavelet transform and deep learning techniques. Initially, images are gathered from various online sources. These collected images are then processed using Adaptive SWT, where the parameters of SWT are optimized using a HTS-BESO. The adaptive SWT splits the entire image into patches for each sub-band. Following this, the DA-VGG16Net framework is employed to extract deep features from these split patches. The parameters of DA-VGG16Net are also optimized using HTS-BESO. Finally, feature matching is conducted using multi-similarity checking to recognize and localize forgeries within the images. The experimental results are compared to various existing forgery detection models to ensure the efficiency of the model by considering various performance measures