Research Of Detection Method For Image Source And Forgery

While the wide application of digital image brings unquestionable advantages topeople’s lives, it also results in many non-ignorable problems. For example, with thepopularity of all kinds of image editing softwares, people can modify imagesarbitrarily, which has brought challenge to the primitiveness and authenticity ofimages. If such image manipulation techniques are applied to the fields of science,justice, journalism and so on, it will cause devastating consequences. As a result, atechnique that can be used to identify the authenticity of images is urgently needed,which is called digital image forensics. Digital image forensics refers to the techniqueand method that can be used to detect image forgery and verify the origin andauthenticity of an image by analyzing the characteristics of this image. With the rapidgrowth of internet technique, the research on digital image forensics has become moreand more significant. Though much work has been done in this field, digital imageforensics is still in its initial stage, and there are many problems to be investigated. Inthis thesis, the image source identification and image forgery detection algorithmshave been invesigated, and the following research work has been accomplished.(1) An improved algorithm of source camera identification based on sensorpattern noise. Via analysing and studying the existing source camera identificationalgorithms, it can be found that the extraction of sensor pattern noise is crucial to thecamera identification algorithm, but it is susceptible to the influences of lowfrequency defects and high frequency image texture. In order to decrease the aboveinterference, some improvements have been done based on the research of Lukas inthis thesis. Firstly, a Wallis pre-filtering is introduced to suppress the low frequentdefects and meanwhile enhance the sensor pattern noise. Then the image areas withstrong edges or complex texture are removed according to the Sobel edge detectionalgorithm. Finally, the source camera identification is implemented in accordancewith the correlation of sensor pattern noise. Experimental results show that themodified algorithm can improve the identification performance effectively.(2) Image forgery detection algorithm based on the inconsistency of backgroundnoise of an image. Different kinds of noise can be introduced unnoticeably during theimaging process. When an image is manipulated by splicing or adding noise, theconsistency of background noise in this image will be destroyed, which can bedetected via the difference of the noise variance. In this thesis, a local noise varianceestimation algorithm is proposed based on the method proposed by Daniel Zoran,which can be used to detect the image forgery from the level of estimated noisevariance. Experiments have proved the availability of the proposed algorithm.