| With the development of computer and multimedia technology, digital images have become the very popular media information, and it has ushered in the era of digital images. In everyday life, non-professionals can edit and modify images to generate the images they want by image-editing software, which gives rise to different visual enjoyment and makes people's lives more colorful. However, on the other hand, the digital images are widely used in military, judicial, news, scientific research and medical fields, then their integrality, authenticity and reliability have more and more attention. Digital images give rise to the visual feast, at the same time, but also create an opportunity for vandals and even criminals. In order to achieve ulterior motives, vandals can create newspaper revelations, false military intelligence and judicial evidence. Moreover, the development of Internet provides a convenient conditions for the dissemination of forged images, making the impact of forged images is not just limited to a certain extent, but around the corners of the world. There is a big security risk in digital images, therefore, it becomes more and more urgent and important to search for an effective method for identifying the authenticity of the images.At present, there are three kinds of relatively technology used to identify the authenticity of digital images: Image identification technology based on digital watermarking, Image identification technology based on digital signature, and Blind identification technology for image authenticity. Image identification technology based on digital watermarking is means which embeds the copyright information as a watermark to digital images using redundancy and randomness of information in digital images, and the image authenticity is determined by the digital watermarking. Image identification technology based on digital signature is means which extracts message digests from the original image as digital signature for identification of image authenticity, and the image authenticity is determined by the digital signature. Blind identification technology for image authenticity is a new kind of analysis idea for authenticity of image content, which identifies the image authenticity by analyzing the abnormal changes of the statistical property, resulted from image tampering. Both of image identification technology based on digital watermarking and digital signature request the provider of the image make pre-processing for the image in advance, however, blind identification technology for image authenticity does not require any pre-embedded information but identifies the image itself directly, having a stronger practicality, and it has gradually become a hot topic of scientific research.Means of image tampering was described firstly in this paper, and we emphatically researched composite tampering which is used most commonly. The existing methods to identify composite tampering were analyzed, and the improved algorithm was presented for detecting composite tampering both in an image and between different images.Composite tampering within the same image is equivalent to the copy-paste operation within the same image, which copies part of the image and pastes it another area of the image to hide the important goals. Such operations often cause correlation between the original image block and the tampered image block, by analyzing the correlation, composite tampering within the same image can be detected.For composite tampering in an image, the existing algorithms mostly can only identify the tampering between regions which are only translation transformation to each other, if there is rotation or scale transformation in regions they mostly can not detect the tampering. Based on this, an improved algorithm was presented which can identify the composite tampering regions though there is rotation or scale transformation. Firstly, the image to be detected graded in accordance with gray, an image block was made up of adjacent pixels with the same gray-level, for each image block calculate the proportion of its actual gray value, that is calculate gray-structure for each image block. Secondly, the suspicious regions can be located according to gray-level and gray-structure of the image blocks. If two image blocks have the same gray-level and the similarity of their gray-structure meets the threshold requirements, they will be set to be a pair of suspicious regions. Then, phase correlation based on the Log-polar coordinate transformation will be done to suspicious regions. If the peak of cross-power spectral density is greater than the relevant threshold value, the rotation angle and zoom scale can be calculated according to the location of the peak. And then, suspicious regions with similar rotation angle and zoom scale are classified as a class. The rotation angle and zoom scale of all the composite tampering regions can be calculated by the lines of the centers of all suspicious regions in a class. A new image will be generated by inverse transform to original image according to the rotation angle and zoom scale, phase correlation are done between the new image and the original image, so their relative displacement can be gotten. Finally, the composite tampering region can be located based on the rotation angle, zoom scale and relative displacement, and the tampering region is marked as white. Use morphological operations on the preliminary detect results to get the final results. The experimental results showed that the algorithm can effectively detect the tampering region though it was rotated or scaled.Composite tampering between different images is equivalent to splicing operation of different images, which usually copies objects or background of an image and pastes them to a region of another image to forge the image attempting to characterize some false appearance. Such operations often undermine the statistical properties of natural images, that is, there are some data inconsistencies in these false images composite from different images tampering, by analyzing the statistical properties of images and data consistency, composite tampering between different images can be detected. If two images are obtained by different image acquisition device, their data will have different characteristics. Different lenses have different optical distortion, different cameras using different color filter array and different interpolation algorithms, so, there are abnormal data features in images composite by two or more images from different image acquisition devices. Different images obtained by the same image acquisition device, although they have the same data characteristics, because of different imaging environment, there will be different characteristics. So, there are some inconsistencies in images composite by two or more images from the same image acquisition device. Based on this, the image data features and the characteristic consistency are often evidence to detect composite tampering between different images.For composite tampering between different images, the image authenticity can be determined by the consistency of light source direction between different regions. For the existing algorithms with shortcoming of high time complexity, we presented an improved algorithm—Blind identification for image authenticity based on Lambert illumination model. This method detects image counterfeit marks by the damage to consistency of light source direction resulted from composite tampering. Hestenes-Powell multiplier method was used to calculate the light source direction for infinite light source images and Levenberg-Marquart least square method was used to calculate the light source direction for local light source images, and the image authenticity was determined based on the consistency in the light source direction. Experimental results showed that the algorithm can effectively detect composite tampering for images with clear lighting conditions, and compared to the original algorithm, the improved one has higher correct detection rate and smaller time complexity. |
PDF Full Text Request