Research On JPEG Re-compression Detection And Anti-detection Based On Header File Modificatio

With the wide application of Joint Image Experts Group(JPEG)compression in image processing and computer vision,it is an important issue to detect the double JPEG compression images in digital image forensics.Many related detection methods have been proposed.When detecting recompressed images,most detection methods assume that the quantization table in the JPEG header is safe.However,there is a possibility that the header file could be tampered with.Once the quantization table in the header file is tampered with,the method will fail.In the case of double JPEG compression forensics,if the quantization table in the header file is tampered with,it will be difficult for the detector to distinguish between double compressed and single compressed images.But header file tampering can not only bring negative impact,but also improve the forensic performance of the algorithm.In thesis,we investigate the double compression detection and anti-detection of JPEG images by modifying the header file in JPEG images.The main research contents are as follows.(1)In the task of double JPEG compression anti-detection,the way of modifying the quantization table in the header file is improved.In thesis,we combine the idea of GAN and introduce the adversarial loss.The quantization table is modified in the way of gradient descent.The algorithm in thesis considers two main metrics,that is higher visual quality and better anti-detection effect.By adaptively modifying the quantization table through the neural network,it is experimentally demonstrated that the algorithm in this paper can make the generated images can deceive the classifier and still have great visual quality with only few modifications.(2)Most current algorithms consider the case where the first and second compression quality factors are known and design the corresponding detector for each compression case.This improves the detection accuracy,but it is hard for the detector to know the first compression quality factor in practice.Therefore,according to this phenomenon,in the task of double JPEG compression detection,the paper investigates a method to improve the detection accuracy by extracting robust features by modifying the quantization table in the case where the first compression quality factor is unknown.The histogram features of DCT coefficients after modifing quantization table are extracted and given to the neural network for training.The quantization table is adjusted at the same time during the training process.The experimental results show that this algorithm has good performance in datasets such as uncompressed color image database(UCID).