Research On Forensic Algorithm Of Digital Image Contrast Enhancement Operation

With the continuous advancement of science and technology,the application of electronic equipment has become more extensive,images have become an important resource in the digital information age.However,with the popularization of editing software,the authenticity of images is at great risk.In recent years,incidents of fraudulent images in political current affairs,military defense,academic research and other fields have been common.In order to combat image tampering by criminals,digital image forensics technology has gradually developed.Operation forensics is a major branch of image forensics.Among them,contrast enhancement operation is the most common and efficient image editing method,it is also often used by forgers to eliminate the sense of visual violation in forged images,so as to achieve the purpose of concealing tampering traces.It can be seen that whether the contrast enhancement traces in the image can be detected is of great practical significance to the authenticity judgment of the image.Under such a research background,this paper studies and analyzes the forensic issues of digital image contrast enhancement operations in different research scenarios from the perspective of accuracy,robustness and practicability,mainly covering the following three tasks:(1)A contrast-enhanced image forensics scheme based on data enhancement preprocessing is proposed.In this work,an algorithm framework combining data enhancement preprocessing and convolutional neural network is proposed for the recognition scene of conventional image contrast enhancement operations.The scheme proposes that using the high-level gray-level co-occurrence matrix of image as preprocessing to enhance the trace features of the contrast enhancement operation,then combine the high-performance classical convolutional neural network for high-dimensional feature extraction and training.Finally,the influence of data enhancement preprocessing method and classification network on the performance of contrast enhancement forensics is explored and studied.Experimental results show that the proposed scheme can accurately detect contrast enhancement operations and has good generalization.(2)A contrast-enhanced image forensics scheme based on JPEG post-processing is proposed.In the JPEG post-processing recognition scene,the existing contrast enhancement operation forensics scheme has poor detection performance and insufficient robustness to JPEG compression.In response to this problem,this work proposed a forensics scheme based on Xception network and designed two pre-processing methods for JPEG compression post-processing.First,we use the gray-level co-occurrence matrix of the image as the input of the network,which can effectively suppress the image content while retaining the traces of contrast enhancement operations.Secondly,we use the constrained convolutional layer to extract the high-frequency detail trace features in the gray-level co-occurrence matrix,which increases differences between classes,then finally using the Xception network to optimize the further training of the features.The final experimental results show that compared with the existing methods,the detection performance of this scheme has been effectively improved under a series of different intensity JPEG compressions.(3)A contrast-enhanced image forensics scheme based on multiple operation chains is proposed.This work is based on a more realistic and complex operation chain recognition scene and proposes a contrast enhancement forensics method based on multi-domain fusion.This method uses the features of the DCT frequency domain and the gray-level co-occurrence matrix domain of the image as the input of the network,using the Res Net network structure for high-dimensional feature learning,and adding the attention mechanism module to the network structure to further improve the network performance.Experimental results show that compared with existing methods,the proposed scheme can improve the accuracy of contrast enhancement detection in multiple operation chains effectively.