FIDD-1500: Fake Image Detection Dataset

One of the most concerning issues nowadays is image forgery or manipulation, which has serious implications for people’s safety and privacy on the internet, the Web, online news portals, and various social media platforms. Regrettably, the proliferation of modified images has emerged as a significant issue in the dissemination of misinformation through the sharing of images across diverse domains. The assessment of automatic Image Forgery Detection (IFD) approaches in real-world benchmarks appears to be constrained by the scarcity of standard datasets, despite the existence of numerous available IFD algorithms. Furthermore, the underlying motives behind the act of manipulation remain ambiguous. The objective of this research is to tackle these concerns by introducing a new Fake Image Detection Dataset, known as FIDD-1500. This dataset is designed to assess the effectiveness and applicability of IFD algorithms. The dataset contains 300 authentic images, and along with each authentic image, there is a set of four manipulated images that correspond to it. The distinguishing characteristic of this dataset lies in its ground truth annotations, which encompass both technical and social features. The research’s experimental phase entails the utilization of Error Level Analysis (ELA) on the FIDD-1500 dataset to evaluate its efficacy in accurately identifying manipulated areas. The utilization of the FIDD-1500 dataset in conjunction with the ELA algorithm establishes a fundamental basis for forthcoming research endeavors within this domain, facilitating the advancement of more refined and precise methodologies for identifying instances of image manipulation.