Research On Image Forgeries Detection Algorithm Based On Feature Points

The development of computer science and technology not only improves the development speed of image editing software,but also promotes the rapid development of image processing technology.How to detect forged images quickly and efficiently is a longstanding problem that needs to be solved.In recent years,tampered and forged images have been used in scientific research,news report,photographic art and other fields,which leads to the problems of academic fraud,misleading and fraud,causing huge losses to the country and society.In this thesis,according to the existing knowledge of image forgeries detection,the following research works have been done on the problem of image tampering and forgery:(1)The common image forgeries technology and image forgeries detection technology are studied.The research and summary of the common image forgeries technology reveal that this kind of technology develops rapidly and has a wide variety.Copy-move forgeries technology has become one of the common image forgeries methods due to its advantages such as easy access to materials,simple tampering and realistic forgeries images.A lot of targeted research works have been done on its detection method based on full understanding of Copy-move image forgeries technology.(2)An image forgeries detection method based on PST(Polar Sine Transform)feature points is proposed.In order to solve the problem of poor detection effect and slow detection speed of the existing forgeries detection algorithms for image forgeries rotation interference,this thesis takes the Copy-move forgeries of image as the background,starts from the extraction algorithm of image feature points,and proposes an image forgeries detection method based on feature points according to the rotation invariant property of polar sine transform(PST)feature.First,preprocess examination image and extract the global PST feature points to form a feature space.Second,in the feature matching phase,the fast approximate nearest neighbor search algorithm Patch Match is introduced,and the two candidate offsets in the propagation phase are improved to eight candidate offsets to adapt to the rotation invariant feature of this thesis.In addition,the random search strategy is adopted to overcome the disadvantage of long matching time due to the huge feature space and search space.Then,median filtering,fitting error filtering and binarization are performed to remove the mismatches.Finally,the connected region deletion method and morphological operation are used to refine the detection results.A large number of simulation experiments show that the detection accuracy of linear Copy-move forgeries and rotation interference can reach more than 99.0%.Although some difficulties are encountered in the process of detecting JPEG(Joint Photographic Experts Group)compression interference forgeries,tampered areas that are accurately detected and located also achieve 60.0%.(3)An image forgeries detection algorithm based on PST-P feature points is proposed,which improves the PST feature points extraction method.First,the polar resample of the image to be tested is used to calculate the PST features to form the PST-P feature space,and then the same process as the PST feature points detection method is used to detect the suspected forged image.The experiments show that the proposed method not only has a very good detection effect on the linear Copy-move forgeries and rotation interference of the image,but also has robustness to the forgeries interference such as size transformation,noise and JPEG compression.Specifically,the preliminary exploration of the comprehensive interference forgeries suggests that the accuracy can reach 98.0% when the comprehensive forgeries range is small.Finally,it shows a great advantage in detection speed.