Research On Passive-Forensics-Based Image Source Identification And Image Splicing Detection Method

As one of the indispensable information transmission media of human society,digital image has made our life more colorful.However,the crisis of confidence caused by digital images has also come one after another.It has brought negative influence to all aspects of human society.In order to solve the problems of the authenticity of image content,digital image forensic technology came into being and has became a hot research topic in the field of information security gradually.In this paper,we focus on the image passive forensics technology,and study the methods of image sources identification and image splicing detection.The main works are as follows.We propose a statistical-features-based image source forensics method.The proposed method is different from the existing image source forensics methods,it can realize two functions simultaneously:(1)Distinguishing computer generated images from photographs.(2)Camera source identification for photographs.In the proposed method,according to the fact that the imaging process of nature image will inevitably introduce the sensor mode noise in photographs,while the computer-generated images does not produce such noise,and different imaging devices will inevitably introduce specific pattern noise in photographs,thus we extract the sensor pattern noise,and combine with five digital features and average gradient values of the Gray-level co-occurrence matrix of the image space domain and frequency domain to construct feature vector.Then we establish statistical model and use the SVM to distinguish the computer generated images from photographs,and identify camera source for photographs.The experimental results show that the proposed method has satisfactory classification effect and robustness,the detection accuracy is not less than that of the methods of distinguishing computer generated images from photographs,and the methods of camera source identificationConsidering that image splicing/compositing is one of the most common images tampering approach,we propose an image splicing/compositing localization method based on multi-scale geometric analysis.In the proposed method,we firstly segment image to blocksusing the superpixels segmentation algorithm(SLIC).Different scale factors are selected in the segmentation process,and the images are divided into blocks at different scales.We de-noise the segmented image blocks by using dictionary learning algorithm,then extracting PRNU from de-noised image blocks,and calculating PRNU noise variance.On the other hand,we calculate the pixel variance of the image block from selected color channel.Then we construct the filtering algorithm to detect the suspicious image blocks.The final tampering regions are defined as the intersection of the suspicious blocks selected at different scales.The experimental results show that our algorithm can accurately detect and locate the splicing image regions,the localization effect is satisfactory,and the algorithm is robust.