A survey on deep learning-based image forgery detection

Abstract

Image is known as one of the communication tools between humans. With the development and availability of digital devices such as cameras and cell phones, taking images has become easy anywhere. Images are used in many medical, forensic medicine, and judiciary applications. Sometimes images are used as evidence, so the authenticity and reliability of digital images are increasingly important. Some people manipulate images by adding or deleting parts of an image, which makes the image invalid. Therefore, image forgery detection and localization are important. The development of image editing tools has made this issue an important problem in the field of computer vision. In recent years, many different algorithms have been proposed to detect forgery in the image and pixel levels. All these algorithms are categorized into two main methods: traditional and deep-learning methods. The deep learning method is one of the important branches of artificial intelligence science. This method has become one of the most popular methods in most computer vision problems due to the automatic identification and prediction process and robustness against geometric transformations and post-processing operations. In this study, a comprehensive review of image forgery types, benchmark datasets, evaluation metrics in forgery detection, traditional forgery detection methods, discovering the weaknesses and limitations of traditional methods, forgery detection with deep learning methods, and the performance of this method is presented. According to the expansion of deep-learning methods and their successful performance in most computer vision problems, our main focus in this study is forgery detection based on deep-learning methods. This survey can be helpful for a researcher to obtain a deep background in the forgery detection field.