Study On Applications Of Digital Forensics For Source Camera Identification And Video Forgery Detection

Digital image source identification is an important branch of digital image forensics,which uses the trances left by hardware or digital signal processing as a basis for distinctionbetween different devices. Using characteristics of color filter array (CFA) interpolation toidentify image source is one of the most important methods of digital image sourceidentification. Most existing CFA interpolation based source image classification methodscharacterize inter-pixel relationship with a linear model, where the estimated interpolationcoefficients are used as features for image source classification. However, the CFAinterpolation algorithm will also introduce correlations between different color channels, andthese correlations can be used as an effective complement to the correlations between pixelsfor image source identification. In recent years, along with the emergence of cheap,high-quality digital video camcorders and functional video editing software, malicioustampering of video is becoming increasingly easy. The rapid emergence of digital videowebsites also makes the needs of malicious tampering detection of video increasingly urgent.In this study, we carry out the research work through digital image source identificationand digital video forgery detection. Based on the analysis and simulation of typical imagesource identification algorithm for, we propose our improved algorithm, and then apply it tothe forgery detection of digital video. The main work can be summarized as follows:1. Simulate and analyze the method based on correlations of color channels, whichextracts two variance maps by estimating the variances of each component of the green-to-redand green-to-blue spectrum differences as the basis to distinguish different interpolationalgorithms, then disadvantages are pointed out. Through analyzing the characteristics ofvariance maps, improved algorithm uses shape and texture features of variance map ratherthan variance map itself. It treats R and B channels differently, and removes unnecessaryestimation process using EM algorithm. Then24-dimensional shape and texture featuresextracted from variance maps are input to an Adaboost classifier to classify the image source.The algorithm is simple and fast without considering complex CFA models and differentinterpolation algorithms. Compared with similar algorithms, our algorithm achieves higher accuracy and better robustness for images shot by commercial cameras.2. Simulate and analyze two classical papers about video tamper detection. One istaking advantage of specific static and temporal perturbations introduced bydouble-compression of MPEG video as evidence of tampering; the other describes twotechniques for detecting traces of tampering in deinterlaced and interlaced video. Fordeinterlaced video, quantify the correlations introduced by the camera or softwaredeinterlacing algorithms and show how tampering can disturb these correlations. Forinterlaced video, we show that the motion between fields of a single frame and across fieldsof neighboring frames should be equal, and propose an efficient way to measure thesemotions and show how tampering can disturb this relationship. In addition, we apply the CFAinterpolation based digital image forensics technology to video forgery detection, and obtain agood result.