Research On Algorithm Of Detecting Non-aligned Double JPEG Compression

With the development of Internet technology and the growing popularity of digitaldevices with cameras, images are playing a more and more important role in theinformation media field. The software on image is increasingly powerful and one caneasily change an image. As a result, the authenticity of the image can’t be guaranteed,once important tampered images are widely put into the network, which will likelyresult in adverse social impact and significant economic losses. Therefore, the imageforensics technology increasingly attracts widespread attention.The feature of high compression ratio makes JPEG compression widely used onthe network, JPEG compression detection is an important research direction of imageforensics. This dissertation is mainly focus on the left block traces of JPEGcompression process. To find the most discriminative features, the composed featurevector are input into the SVM classifier, using images library to train classifiers imagesample, so that it can distinguish whether a JPEG image original or tampered.Deficiencies of the previous algorithm are analyzes. In this dissertation, theseproblems are solved: Introduced the idea of reference image of the original image andimproved the detection accuracy a lot; proposed the concept of histogram of blockingartifacts.8histograms in the horizontal and8histograms in the vertical are computedeach image used to describe the block effect. The computational complexity declinesand time-consuming is getting shorter; Proposed the method of second differencehistogram. It solved the effects to the image blocking artifacts histogram which arecaused by the contents of the image in conventional method vulnerable; established asample image database. Sample image library can be used to train the SVM classifierwhich is prepared to distinguish whether a JPEG image is tampered or not.