METHOD FOR GENERATING A SET OF FAKE IMAGES FOR TRAINING A “COPY-MOVE” TYPE FAKE DETECTION SYSTEM УДК 004.056+004.932

Abstract

In today's world, image forgery is very common. The most common method of falsifying images is «copy-move». The advantage of «copy-move» over other modification methods is that an area from the same image is inserted, that is, it has the same brightness and contrast as other objects. Also, these images can be subject to various post-processing techniques and various manipulations, such as JPEG compression, brightening or straightening, which can reduce traces that make detection difficult. In order to train copy-move counterfeit detection systems or evaluate the quality of their work, special data sets are needed. This paper presents a method for adaptively generating a set of fake images for training a copy-move type forgery detection system based on the BusterNet neural network model [1]. The method of adaptive generation of a set of images proposed in the work demonstrated a high level of quality of the created fakes. During its work, a wide range of pre- and post-processing methods was used, which made it possible to make the generated set resistant to detection by modern neural network methods. Its use will make it possible in the future to effectively carry out further training of new methods for detecting «copy-move» counterfeits.