Research On Blind Detection Technology For Digital Image Tampering

Digital image has become one of the most important forms in the field of modern mutil-media because of its intuition, real-time and easiness to deal with. With the increasing development of digital technology and image processing software, ordinary users also can tamper images easily. Accordingly, the phenomenon of digital image forging occurs frequently in all areas of society.In the meantime, the authenticity and integrity of digital image has been questioned and digital image detection technology has become the hot spot in the current security field.At present the more popular image format is the JPEG compressed format. The model of digital image tampering is divided into three types which are uncompressing, simple compression and double compression based on whether the storage format is JPEG. Double compression also can be categorized into two classes called align double compression and the unaligned compression according to the DCT grid offset of the first and second compression. According to the four models, this paper puts forward a detection method for digital image tampering based on the inconsistency of noise estimation and quantitative table as follows:Firstly, this paper proposes a detection method of digital image tampering based on the background noise estimation. Taking advantage of the image skewness statistic characteristic estimates the noise standard deviation block by block. Differential evaluation method is proposed for gaining the optimal value in background noise estimation algorithm, avoiding the appearance of local minimum value. This method has broke the rule that the original signal must be known in high order statistic algorithm and it efficiently avoids the interference which the image detail parts may cause on the noise standard deviation and improve accuracy of the noise estimation.Secondly, this paper proposes a detection method for digital image tampering with single image compression based on quantitative table inconsistency. Do research on the DCT histogram characteristics of simple compression, improve the estimation algorithm of original quantitative table, and then use the estimated original quantitative table to realize the digital image forensics for tampering. The algorithm also includes the generation of the binary image. The experimental results show that the improved algorithm can estimate the quantization step length of direct current (DC) component, improve the accuracy of quantitative table estimation and locate the accurate tampering area.Thirdly, this paper proposes a detection method for digital image tampering with align double compression based on quantitative table inconsistency. Take advantage of the DCT histogram characteristics to estimate the original quantitative step length, do research on the effect of align double compression and then realize the detection for tampering area according to the relation of the two quantitative tables. The experimental results show that if the quantitative step in the first compression is greater than the second, the algorithm is hardly enacted on the background details.Fourthly, this paper proposes a detection method for digital image tampering with unaligned double compression based on quantitative table inconsistency. First, ascertain the first compression grid through estimating the grid offset between the two compressions, and gain the quantitative step length of the first compression through the periodicity of DCT coefficient histogram, and ascertain the tampering area through histogram posterior probability at last. The experimental results show that if quality factor D-value between the two compressions is over25,the correct detection rate is more the90percent and the algorithm can detect smaller size image.Fifthly, this paper proposes a system framework of blind forensics for digital image tampering. The algorithms in the system can realize auto-detection and location for tampering area under four conditions.