Research On Forensics Of Forgery Images Base On Features Extraction

As the digital image processing software is widely used in the daily life and work,people benefit from the wealth of digital image information resources, but at the sametime, also suffer from serious hazards brought by the forgery image. Thus, theauthenticity and integrity of the digital image need to be recognized urgently. Blindforensics of digital image, which is an important branch of the information security, cansolve the above problem effectively. The digital image forgeries consist of copy-moveforgery and splicing forgery. Nowadays, all the previous algorithms mainly utilizeimage features to recognize the copy-move image and classify the splicing image.However, they have low precision rate and high false positive rate. In this paper, theeffective features of the forgery images are analyzed and extracted, then the newdetection methods are proposed for the forensics of the two kinds of digital imageforgeries.The main work of this paper is as follows:①The concept of digital image forgery and the background knowledge of digitalimage forensics are introduced. The classification of the blind forensics of image isanalyzed. Meanwhile, features extraction and support vector machine, which areutilized to detect the forgery images, are briefly presented.②Adetection scheme for copy-move forgery images by using undecimated dyadicwavelet transform (DyWT) and Zernike moments is proposed. This method mainlydetects the copy-move forgery image which undergoes post-processing (such as rotation,Gaussian noise, blurring and JPEG compression). In this method, Zernike moments,which have the rotation invariance, are extracted in the low frequency component ofDyWT for detecting the forgery images. Furthermore, the high frequency component ofDyWT is used to filter the detected result for reducing the false positive rate.Experiments show that the proposed algorithm can detect the copy-move forgeryimages effectively. Meanwhile, the new detection method is better than the previousmethods with respect to the precision rate and false positive rate.③A detection scheme for splicing images by using statistic features is proposed.According to the analysis of the variance caused by image splicing operation, four kindsof image features are extracted to detect forgery images. The four kinds of imagefeatures consist of approximate run-length features, texture features, frequency features and edge features. Moreover, support vector machine is utilized to classify the realimages and forgery images. Experiments show that the proposed algorithm can extracteffective image features to improve the accuracy of recognizing the splicing images.