Blind Identification Algorithm Of Photorealistic Computer Graphics Based On LTCP Features

With the advent of the information age, along with the development of electronic technology and the popularity of digital cameras, taking digital images becomes extremely simple and digital image is widely applied to life and learning. Digital image act as various roles as a semantide, showing scenes of life, recorded news, forensic and so on. However, while the development of digital image technology, digital image processing software technology also will be developed so that the image processing becomes easier. In order to meet the individual needs, arbitrary tampering led to frequent incidents of image tampering. Therefore, the image tamper detection which can ensure the authenticity of the image content is of great of has important meaning.Digital image forensics aims at detecting possible image forgeries and determining that the image is authentic or not. Generally speaking, there are two image two different kinds of technologies in digital image forensics, referred to as active and passive. Active forensics technology needs some information to be inserted into digital images in advance, and the digital watermarking and digital fingerprint are the main forms. Without the addition of any preprocessing information of image forensics technology is called passive forensics technology, also known as blind digital image forensics. Blind digital image forensics technology mainly considers whether the image is the original image, the image content is complete, and how images are generated. Thus, the digital image passive forensics can be roughly divided into three categories: Detection based on the authenticity of the image content which can determine an image from after the initial acquisition whether through tampering; Forensics detection based on image source which can judge sources by image acquisition equipment; Forensic detection based on the analysis of image steganography which can determine the integrity of the image.Aiming at photorealistic computer graphics in the source image forensics, these images are detected by extracting features. First of all, research background and significance of the identification of blind digital image forensics technology is introduced and the overseas and domestic research status digital image forensics technology is also analyzed. Then, a brief introduction of the image color space, LIBSVM classifier and photorealistic computer graphics dataset are given in this paper. Finally, analyze the imaging mechanism, summarize the differences which can be reflected by the LTCP histogram. We can put forward two, put forward two kinds of blind identification algorithm for on the basis of improving the existing algorithm of photorealistic computer graphic.The existing computer-generated image detection method about photorealistic computer graphic, by analyzing the differences in the texture of the image, using the histograms based on LTCP features and co-occurrence matrix features based on coherence of adjacent pixels to reflect this difference, we made two blind identification algorithms based on LTCP and co-occurrence matrix to detect photorealistic computer graphics. First, convert the original image to HSV color space, separating image channel in the HSV color space respectively. Secondly, for each channel, extract LTCP features and co-occurrence matrix features based on coherence of adjacent pixels, use SVM classifier to train the fusion features and establish the classification prediction model. Finally, detect whether it is photorealistic computer graphic according to the classification model. The experimental results show that this algorithm has obvious improvement in detection results, with some research prospects.For the existing image identification algorithms based on texture features, they don’t take full advantage of scale and directional information of the image, and the differences of t exture aren’t fully reflected with high dimension, low detection rate. A blind identification algorithm is presented based on SPT and LTCP by conducting steerable pyramid transform and extracting texture feature differences from the multi-scale and multi direction. First, convert the original image to color space, separating image channel in the HSV color space respectively. SPT transformation is conducted on each channel and sub-bands are obtained in an image. Secondly, extract LTCP features are extracted in image sub-bands, SVM classifier is used to train and establish the classification model. Finally, detect whether it is photorealistic computer graphic according to the classification model. The experimental results show that this algorithm has obvious improvement in detection results, and the comprehensive detection rate reached 97%.