Research On Natural Image Blind Identifying Algoirthm Based On Fractal Dimension

With the information age’s coming, digital image technology has developed rapidly, most of the scenes images as well as literary and artistic works have been stored in digital form, they are stored in the form of a digital image, digital image transmission becomes very convenient with the rapid development of Internet technology. When people are enjoying the convenience of these digital images, many negative effects come. For example, people take advantage of some tampering images, but these tampered images are the results of illegal copies, however, the purpose are commercial gain, the condition like this is everywhere. For some personal interests and preferences, by using computer graphic software and a variety of image-editing software, people can easily draw out a variety of vivid scenes images that are difficult to identify the authenticity with the naked eye. So many people beginn to question the so-called "personally seen", therefore, identifying the authenticity of the digital image has become a growing topic of people’s concerning, and it should be taken seriously.The authenticity of digital image includes the following two aspects:the authenticity of digital image sources and the authenticity of the digital image content. The authenticity of digital image sources mainly refers to the authenticity of the ownership and copyright of the digital image, means whether the digital image is after the illegal copying and piracy, and so on. The authenticity of the digital image content mainly refers whether the digital image is subjected to a malicious tampering. The tampering mainly includes image compression, image copying and pasting, image magnification. Digital image forensics mainly includes three aspects:digital signature technology, digital watermarking technology and blind digital image identification techniques. Blind digital image identification technology is different from the digital signature and digital watermarking technology, and applications of blind digital image identification techniques are the most broad, blind digital image identification technology does not require any prior knowledge and it can reach a final detecting purpose only by testing images. Blind digital image identification technique includes the copy-paste detection, resampling detection, image source detection and so on. Image source detection includes natural image detection and computer generated image detection. By using the blind digital image identification technology, it can be more convenient to know the authenticity of the digital images, and there are a variety of detection methods of blind digital image identification, therefore, blind digital image identification technology has played more and more important role in people’s life.Sources of digital image forensics mainly determine the source of the image, that image is from cameras, or drawn by the computer software. Because the image forming process of natural images and the computer generated images is inconsistent, so we can find the differences between the two types of images in the image forming process. Compared to computer generated images, natural images have inherent pattern noise that the computer generated image does not have, and the difference in continuity and texture changes between the natural images and the computer generated images is an important factor. Therefore, we can get a lot of statistical information from the image’s surface texture and continuity between pixels, and finally we can determine the source of the image.The efficiency of traditional using fractal dimension algorithm to identify natural images is low, this paper presents blind identification algorithm based on local fractal dimension and wavelet domain features. There are differences in the statistical characteristics between natural images and computer generated images, and then a feature combination technology is proposed to detect natural images and computer generated images. The features are divided into two aspects; one is the characteristics in the transform domain, namely the sub-band coefficients of the wavelet domain and its higher-order statistics of linear prediction errors. The statistical features from surface texture and the self-similarity of natural images and computer generated images are as another type of features. The support vector machine (SVM) classifier is used to identify the computer generated images and natural images. The experimental results show that the detection accuracy of the computer generated images is rate up to96.5%, so it can prove the effectiveness of the algorithm. In view of only using the first-order fractal dimension to identify the natural image is not efficient. This paper also presents a algorithm that takes advantage of the characteristics of the fractal dimension from first-order and high-order, due to the differences in the image surface texture and image color between natural images and computer generated images, then propose a method to identify computer generated images and natural images based on multiple fractal dimension. In this algorithm, the image color space is from the RGB to HSV, we extract fractal dimension on HSV color space; extract fractal dimension of density gradient image; extract fractal dimension of prediction error matrix as well as global of fractal dimension. In this paper, the fractal dimension that is extracted forms a group of feature vectors, the SVM classifier is used to classify the images from multiple angles. Experimental results show that the detecting correct rate of computer generated images of blind identification algorithm can reach96%, it can prove the effectiveness of the proposed algorithm.