Digital Image Forgery Detection Using Cyclic Symmetry Convolutional Neural Network

Digital Image Forgery (DIF) detection involves identifying instances where a portion of image is copied and placed in different areas within same image to create a seemingly authentic but altered version.However, the detection of small duplicated regions is challenging, especially when noise is present in the image.This issue becomes more significant when the model is trained on noisy data as it negatively affects its accuracy.This research proposes an Elite Opposition-based Learning with Black Widow Spider Optimization (EBWSO) for feature selection and Cyclic Symmetry Convolutional Neural Network (CSCNN) for detection to enhance accuracy in image forgery detection.Pre-processing techniques such as Single Image Super-Resolution (SISR) and Histogram Equalization (HE) are used for image enhancement.The VGG16 and ResNet50 are used for feature extraction in digital images, done through identifying the key features such as edges and shapes.The EBWSO technique is utilized for feature selection by updating the relevant features and balancing exploration and exploitation.Detection is carried out using the CSCNN technique, which is designed to be rotation-invariant and aims to enhance detection accuracy.The EBWSO-CSCNN model accurately classifies image forgery when compared to the existing techniques such as Stacked Sparse Denoising Autoencoder (SSDAE) and Simple Linear Iterative Cluster (SLIC) algorithm.The proposed method achieves a better accuracy of 99.15% on MICC-F220, 98.10% on MICC-F600, 99.25% on MICC-F2000, and 98.95% on the CASIA 2.0 dataset.