Photorealistic Computer Graphics Identifying Algorithm Based On Wavelet Transform And Fractal Dimension

The popularity of high-quality digital cameras and a variety of simple, powerful image editing software continue to emerge, which makes more and more users edit or modify more simply. Now, even non-professionals can already easily create a digital image they want according to their own ideas, whose effect is difficult to distinguish by human eyes. However, there are some people tampering them maliciously for some ulterior purpose. The effective regulation and testing of image tampering in scientific research, military and other areas has become a key issue to be solved in the current picture information security field.Digital image forensics we use to resist tampering image is an emerging field of study, which mainly includes two major research directions: active and passive forensic evidence. Active forensic technology can be refined for digital watermarking technology and digital signature technology. Both of them need to be pre-processed with some limitations in protected images. However, passive forensics determines whether it’s been tampered with or synthesized only by the images’ own features. Relative to the active forensics, which is called blind digital image identification, it is considered to be more challenging academic subjects. Due to the different generated ways between natural images and computer generated images as well as the large differences in continuity and texture, we can start here and then get statistics to determine the source of images.The main work:1. Describe the differences between natural images and computer-generated images according to the digital imaging principle, natural images imaging are generated through digital camera CCD sensor and then obtained R, G, B Pixel value of the three channels by CFA interpolation. While computer-generated images are produced by a series of analog camera imaging process of image software such as retouching, sharpening, smoothing, and even re-interpolation. Therefore, natural images and computer-generated images differ in texture.2. This paper presents a computer-generated image authentication algorithm based on wavelet transform, and extracts feature value of images with multi-resolution wavelet transform form. Images are wavelet de-composited and then corresponding high-frequency and low-frequency information are available. Each time doing that, a low-frequency image and three high-frequency images are received at last. The former mainly save the low-frequency information of images, while the latter save the details. At the same time the Gaussian noise de-noising algorithms is used to extract image histogram as feature vectors.3. This paper advances a computer-generated image authentication algorithm in view of image gradient and fractal dimension. Image gradient can effectively obtain edge information which is important differences between natural images and computer-generated images. In this paper, the statistical histogram of the gradient magnitude and direction is drawn as the feature vector extraction after image gradient transformation, Fractal dimension can describe the image texture information better, in which divide the image into pieces first and extract characteristic values making use of the box dimension algorithm, and it is improved as well in this article.In this paper, the SVM classifier is chosen to detection and training feature. The experimental results show that the proposed algorithm has a high identification rate, as well as is able to identify the natural images and computer-generated imagery effectively.