Texture Feature Based Copy-move Forgery And Sharpening Detection Of Digital Image Forensics

Digital images are editable.They can be modified optionally via image processing and editing software even by ordinary people.With the coming challenging of their authenticity and objectivity,whether a digital image is tampered or not has become a hotspot of the researches on digital image examination.Based on passing researches,the current thesis further investigated the copy-move forgery,USM(Unsharp Masking)sharping of the digital image.Local texture features ware employed as a research point and the present testing methods were improved to suit the urgent needs of rapid test of tampered image,thereby raise the intelligence level of the rapid test of evidence at crime scene comprehensively.For the problems on inferior test accuracy and unsatisfactory robustness of most forensic digital image analysis methods,this thesis achieved to proposes a detection method for copymove forgery and attempts to detect tampering in a single texture area.At the same time,a novel test method based on local texture features was employed to attain better test accuracy,while there were few researches on USM sharping analysis at home.There were three parts in this thesis.To begin with,Harris algorithm was used to extract the key points of the digital image and one neighborhood was established surrounding each key point.Local Configuration Pattern(LCP)was employed to extract the characteristics inside the neighborhood and the matching points were determined by analyzing the feature vectors.Random Sample Consensus operator was used for eliminating the wrongfully established matching point and determining the copy-move area.The second part illustrated the modification of traditional Harris operator.The thresholds were adjusted to extract the key points evenly and densely in the whole image,thereby the whole image could be divided into two regions: normal texture region and single texture region.After that,the traditional Harris operator were used to extract the key points in normal texture region,while the dense Harris operator in single texture region.In the third part,a Local Binary Patterns(LBP)based test method was proposed according to the effect that the texture characteristics of an image undergone during the process of the USM sharping.LBP was applied to extract overall characteristics of the image,and the results feature vectors were put in the Support Vector Machines(SVM)for a training model.Thus,whether the USM sharping had happened on an image could be determined by binary classification,and explored the optimal detection pattern.Based on the procedure above,a graphical user interface was created and it formed a digital image forensic system.The study was meaningful in several aspect.Firstly,copy-move forgery region could be recognized accurately by LCP,with good robustness to rotation,brightness,and other operations,and homologous and heterologous multizone tampering can also be detected.Secondly,the dense Harris operator could extract key points in single texture region,which combined with LCP made it possible for us to test the tampering in these regions.Thirdly,the LBP extraction,combined with SVM machine learning,could identify a USM sharpening digital image efficiently.Among all the detection patterns,the rotation invariant pattern was outstanding in accuracy and stability.For weak sharpening images,the test accuracy was higher than 90%,and for strong sharpening ones,it reached 100%.Finally,a digital image forensics system is established,which can detect copy-move forgery and image sharpening through a graphical user interface.This thesis effectively improves the positioning accuracy and robustness of copy-move forgery.And both homologous and heterologous multizone tampering could be located precisely.The detection problem of single texture region is solved,in the meantime,compared with block-based image detection,the division method had reduced the time needed substantially,and the method might provide some assistance for the image segmentation in computer vision.Image sharpening detection method were better than former methods in passing researches in accuracy,especially for the weak sharpening images.It was the digital image forensics system established that combined the theory and the practice,which suited the practical needs in public security field,and a new automatic detection method is provided for image copy-move forgery and sharpening forensics.