Blind Identification Algorithm Of Photorealistic Computer Graphics Based On Local Texture Features

Along with the continuous upgrade of digital cameras and the rapid development ofimage processing technology, digital images appear frequently in people’s field of visionbecause of its advantages of intuitive and being clear at a glance. They are widely used innews reports, judicial forensics, advertising and film production, etc. At the same time, imagemanipulation technology has been improved. Make certain processing in video can moresatisfy people’s visual enjoyment, but if applying the tampered images to news reports andjudicial forensics, it will threaten people’s right to know about the truth and even will let thecriminal at large. The event emerges in endlessly which some criminals use tampered imagesinstead of the natural images. It leads to the fact that "it may not be the truth with the image".Thus, digital image forensics technology has become a hot research topic.The main task of digital image forensics technology is to judge whether the image isnatural images or has it ever been tampered. If it has been tampered, judge the means. It is amultidisciplinary relates to computer graphics, image processing, computer vision, machinelearning, and information security, etc. According to whether the pretreatment of informationis needed, digital image forensics technology includes digital image active forensics anddigital image passive forensics technology. Digital image active forensics technology isdivided into image forensics technology based on digital watermarking and image forensicsbased on digital signature technology. Corresponding to the active forensics technology ispassive digital image forensics technology, which is also known as blind digital imageforensics.First, this paper briefly introduces the technology of digital image forensics. Then,introduce the imaging principle of photographic images and photorealistic computer graphicsin details, and summarize the differences of these two types of images in many aspects. At last,propose the improved algorithms on the basis of the research and analysis of the existingblind identification algorithms of photorealistic computer graphics. Furthermore, introducethe Columbia photographic images and photorealistic computer graphics dataset in details.Photorealistic computer graphics are images which use image processing software to simulate the imaging process of photographic images. It is a more novel tampering meansthan copy-paste and splice. With the rapid development of image processing technology,photorealistic computer graphics has become more and more realistic, which is hard todistinguish with photographic images with the naked eyes. But the imaging process ofphotorealistic computer graphics and photographic images are different, which leads to thedifference of statistical distribution of these two types of images inevitably. According to thedifference of the statistical characteristics, it can identify the natural images and photorealisticcomputer graphics effectively.Considering the classification features, which the existing blind identification algorithmsof photorealistic computer graphics select, have high dimensions and low detection rates, thispaper proposes a blind identification algorithm of photorealistic computer graphics based onweber local descriptor. First, convert the original image from RGB color space to HSV colorspace. Second, to extract weber local descriptor features from three single channel image ofHSV color space, mainly including differential excitation histogram features, Sobelorientation histogram features and Sobel gradient histogram features. In the end, send theextracted classification features into the SVM classifier for training and testing. Theexperimental results show that the algorithm reduces the dimension of features and achieveshigh detection rate at the same time.In view of the fact that the dimension of the classification features the existingphotorealistic computer graphics algorithm based on local binary pattern has adopted is toohigh, this article puts forward with the blind identification algorithm of photorealisticcomputer graphics based on local ternary count. To identify the photographic images andphotorealistic computer graphics by extracting local ternary count features from the HSVcolor space of images. Firstly, convert the original image from RGB color space to HSV colorspace. Secondly, to extract local ternary count features from three single channel image ofHSV color space. Finally, send all classification features into SVM classifier to train and test.The experiments indicate that this algorithm ensure the detection rate and reduces theclassification feature dimension effectively at the same time.