Decoding Image Integrity: A Comprehensive Analysis of YOLOv8’s Performance for Detecting Copy-Move Forgery

Copy-move forgery represents a prevalent type of image manipulation wherein a segment of an image is duplicated and inserted into another area within the same image. Its primary intent is to deceive viewers or alter the image’s content. Detecting such alterations is pivotal in upholding the integrity and authenticity of digital images. In this research, we undertook the task of training the YOLOv8 image classification model using our extensive image forensics dataset. Additionally, we leveraged Gradio to develop a user-friendly graphical interface, streamlining the detection process. The proposed methodology unfolds in two distinct phases: the training of the YOLOv8 model and the creation of a user-friendly interface using Gradio. To validate our model’s performance, we conducted training and validation on a combined dataset created from MICC-F220, MICC-F600, and MICC-F2000 datasets. Furthermore, we assessed the model’s generalization capability by subjecting it to the CoMoFoD v2 dataset. The experimental results substantiate the model’s efficacy, revealing an impressive accuracy of 97% with a mere 0.030 error rate, achieved within a concise span of 10 training epochs. These outcomes underscore the practical utility and robustness of our proposed framework.