DB-DSCN: Image Tampering Detection Using Dual Branch Deep Stacked Convolutional Network

During the new era of technology with unusual image forging equipment and procedures, digital imaging has become basic. Digital images cannot even be used as evidence anywhere, since it is widely recognized that they may be faked. In order to assist in relieving such derelictions, the problem is examined in an incomprehensible manner. In the digital era, copy-move and splicing of images to produce a fabricated one are commonplace. While the latter entails combining two images to drastically alter the original and produce a new, forged image, copy-move entails copying a portion of the image and pasting it onto another portion of the image. Therefore, to address the limitation of the existing model, a novel hybrid Deep Learning (DL) model is developed. To enhance image details before detecting a tampered image, high pass filtering is employed. The high pass filter might reveal information that has emerged due to the tampered image. To extract the visual semantic features and statistical features, Attention-based Residual Network (AttResNet) and Grey Level Co-occurrence Matrix (GLCM) are utilized. An Attention-based Feature Fusion (AttFF) module is used to fuse the extracted features. A Dual Branch Deep Stacked Convolutional Network (DB-DSCN) is employed to classify the tampered image. The experimental results of the proposed model achieved an accuracy of 98.26%, a precision of 96.28%, a recall of 95.98% and an F1-score of 96%. The proposed model seems to be a superior model for detecting tampered images compared to existing models. It acquired higher accuracy than other existing models. The performance of the proposed model is evaluated in terms of accuracy, precision, recall and F1-score, respectively.