The Research Based Of Statistical Feature On Detection Of Digital Image Passive Forensics

Contrary with Active Forensics, Passive Forensics refers to the digital imagecontent original, integrity and authenticity of the digital image forensics technologywith no embedded the digital signature symbols in digital image. Passive digital imageforensics algorithms have some shortcomings. The reason is that they could not findthe feature extraction algorithm to seize the digital image of the essentialcharacteristics of the image, such algorithm have low robust and accuracy, and narrowscope of application. This article is based on the theory of natural images and passivedigital image forensics, to analyze and study on passive digital image forensics.(1)Analyzed theoretical model of network function interpolation method anddigital image feature extraction. Based on the previous studies of network functioninterpolation built a digital image feature extraction algorithm for the detection ofcomputer-generated images and spliced images. Compared with Algorithm whichbased on higher order statistics and geometric variables, the algorithm has a higherefficiency, accuracy, robustness.(2)Analyzed theoretical of digital image feature extraction with a large number offeature extraction algorithms to form a theoretical framework. According to thePredictable theory of image, researchers have proposed a lot of algorithm to estimatethe image; however, some image estimation method has not yet been introduced intothe passive digital image forensics field. The theoretical basis of this paper includestheory of Net function interpolation method and digital image feature extractiontheory.(3)In order to classify in passive digital image forensics field and build concept ofnatural images, this article has modified the definition of natural images particularlyrefers to images which content have features of non-artificial, non-random, primitive,authenticity and integrity.(4)We have collected image quality evaluation parameters which are the differenceof pixel, correlation, Spectral distance, the human visual system, edge features from alllevel wavelet sub-bands and those predicted. We have had all features of the passive forensics algorithm with the wavelet coefficients at all levels of the mean, variance,skewness, and kurtosis.