Research On Image Splicing Detection Based On Improved Markov Features

With the rapid development of digital technology,image acquisition equipment and Internet technology are rapidly popularized,so that digital images play an increasingly important role in daily life.However,with the popularization of image processing software,image tampering operations have become very simple and difficult for human eyes to perceive.Malicious tampering images appearing in the fields of news,economy,science and technology,justice,politics and military are seriously threatening social justice and social stability.Therefore,research on digital image detection technology is extremely urgent.The detection technology is generally divided into active detection and passive detection.The current passive detection technology is only for specific tampering operations.Tampering operations can be divided into three categories: original tampering,integrity tampering,and authenticity tampering(content tampering and post-processing tampering).Splicing is a kind of content tampering.According to the imaging process of digital images(acquisition,coding,editing),passive detection technology is divided into two categories: detection technology based on inconsistency of characteristics introduced in imaging process and detection technology based on image statistical characteristics.By analyzing the shortcomings in the current research,a detection technique based on improved Markov features is proposed.The details are as follows:(1)Detection technology based on improved Markov featuresIn view of the current problems of Markov feature-based detection technology: the loss of information caused by converting color images to grayscale images or extracting features in only one color channel;high time complexity due to the complex feature extraction process;and the redundant information caused by high feature dimensions,this paper proposes an image splicing detection method based on improved Markov features.Combining the existing problems of Markov feature-based detection technology(insufficient use of color information,high time complexity for feature extraction and redundant information exists in high feature dimensions)and current research hotspots(feature improvements,feature dimensionality reduction and color channel selection),an image splicing detection method based on improved Markov features is proposed.First,extract the three channel features in the YCbCr(or RGB)color space to reduce information loss and fully retain the information of each channel.Second,add inter-block features to extract more comprehensive splicing trace information.Thirdly,the transition probability matrix in the horizontal and vertical directions is extracted in the BDCT domain to highlight the image edge information and select an efficient direction,thereby reducing the complexity of features and improving the efficiency of the algorithm.Fourth,sum the transition probability matrix under each color channel to reduce the feature dimension and reduce the redundant information to highlight the splicing trace.Finally,support vector machines are used to train three classifiers and perform comprehensive decision classification to check the performance of the algorithm.(2)Experimental verification and research conclusionIn order to verify the effectiveness and applicability of the improved algorithm,comparative experiments are conducted on the two color spaces and three public datasets,and compared with other methods.Comprehensive consideration of accuracy and time complexity to determine the optimal parameters and final experimental results: the detection accuracy on the Columbia uncompressed image splicing detection evaluation dataset,CASIA V1.0 and CASIA V2.0 datasets are 95.81%,98.22% and 97.09%,respectively.This method is better than the current main method.At the same time,through the two sets of comparative experiments between intra-block features and inter-block improved features,unsummed features and summed improved features,the effect of inter-block features and summing operations on improving detection performance is verified.Adding inter-block features,the detection accuracy of the Columbia University dataset and the CASIA dataset are generally improved by 4 and 2 percentage points,respectively.The summation operation improves the detection accuracy by about 1.5 percentage points.Finally,the detection accuracy on the IFS-TC competition dataset reached 92.23%,further verifying the practicability of the algorithm.The research found that: 1)Improved intra-block and inter-block features can extract more comprehensive information;summing operations can reduce feature dimensions and highlight splicing traces;generally,the highest accuracy is achieved when the threshold is T = 3.2)YCbCr color space is more suitable for extracting stitching traces.3)Because of the influence of the data set image processing method,storage format,size and number,the experimental results on different datasets are different.