A Study On Medical Image Origin And Forgery Identification

The rapid development of Internet technology and informatization construction of medical industry,like generation and development of the Electronic Medical Record and Picture Archiving and Communication System significant benefits for medical treatment service.Hospital Information System shared in a wide variety of daily medical practice,and efficient retrieval huge amounts of medical information being achieved through digital storage.Currently,most countries attach great importance to regional medical collaborative service development which provides new approaches to the health resources and medical information sharing between hospitals.But at the same time,the medical data is as much important as these security issues in terms of confidentiality,availability,reliability and traceability are to be solved.The introduction of digital signature and digital watermarking can provides protection for medical images,but to be efficient,these initiative schemes need to be implemented inside the image acquisition device,this may limit their use in real practice.To go further,a question is how to verify the origin and forensics of a given image only from its pixels' gray values,with no any prior knowledge about the image.Most of these blind forensics techniques have been proposed in the case of general public devices,e.g.digital cameras,optical scanners or smartphone.In this dissertation,based on analyzing the principle of medical image acquisition system,the sensor pattern noise and imaging algorithm features extracted as fingerprint of image to identify the origin and forensics of a medical image.The main contributions are summarized as follows:(1)Digital Radiography image origin identification algorithm based on noise featuresFirst,based on the imperfection of Digital Radiography(DR)acquisition systems,the Digital Radiography Pattern Noise(DRPN)which can serve as a fingerprint so as to identify the origin of DR images.Taking the different of DR acquisition systems with digital cameras,it is impossible to get a "pure" noise image without any object,the Contourlet filtering were proposed for DRPN extraction,which provides higher identification precision benefit by minimize the impact ofthe edge and texture information of image.On another side,in consideration of all DR images are "raw" image without any compression,in addition to DRPN,scattering noise and thermal noise take up much part of noise.By studying the relationship in-between the image pixel value and the noise intensity,parametric statistical model was built,then the parameters of this model as fingerprint can be used for DR origin identification.But this approach is only effective for different DR model identification,cannot be a specific DR identification for use.So combining this fingerprint with DRPN to identify the specific DR,and the identify accuracy rate has increased at the same time.(2)Computed Tomography image origin identification algorithm based on Original Sensor Pattern NoiseBased on the principles of a Computed Tomography(CT)scanner acquisition chain,X-Rays are emitted and read out by an X-Ray detector array after having passed through the patient,then provided to a tomographic reconstruction algorithm in order to build the final CT tomographic image.The Sensor Pattern Noise can be used as CT fingerprint is directly related to the X-Ray detector array.Then,the Original Sensor Pattern Noise(OSPN)are obtained by the inverse transform of three dimensional CT reconstruction for each type of sensor pattern noise.So as to discriminate acquisition systems based on OSPN.The proposed origin identification algorithm based on OSPN provides higher identification precision compared with other existing methods based on sensor pattern noise.(3)Computed Tomography Image Origin Identification algorithm based on Original Sensor Pattern Noise and 3D Image Reconstruction Algorithm fingerprintsThe principles of a CT scanner acquisition chain illustrated that the reconstruction algorithm is not only reconstruct CT image but also reconstruct the OSPN.Based on the way OSPN is modified by its reconstruction algorithm,the 3D image reconstruction algorithm fingerprint can be acquired.And all the reconstruction algorithms are manufacturer dependent and kept secret.Resort to the simulated experiments conducted on 5 basic reconstruction algorithms,there are some periodicity and strong similaritiesappeared in reconstructed noise,as a consequence,the correlation feature vector was proposed aim at putting in evidence of such periodicity and similarities of reconstructed OSPN.From simulated experiments,this set of features as reconstruction algorithm fingerprint should be able to discriminate CT scanner according to their reconstruction algorithm.Experiments conducted on real CT images show it is possible to identify the origin of one CT image with high accuracy.But this reconstruction algorithm fingerprint is only effective for different 3D reconstruction algorithm,cannot be a specific CT identification for use.So we propose combining this fingerprint with OSPN to solve this problem,and the identify accuracy rate has increased at the same time.(4)Medical image forensics algorithm based on sensor pattern noiseSensor pattern noise proposed for medical image forensics which as image fingerprint used for image origin identification.Based on detecting the tempering part of image benefit by detect the fingerprint correlation between under investigation image and reference one.Specifically,this based on correlation image forensics is a regional image origin identification,the forgery detection algorithm tests for the presence of the fingerprint in each block separately and then fuses all local decisions is come from the reference device or not so as to identify the block is tempering or not.The experimental results have shown that the proposed algorithms not only for forensics but also give out the tampering part location,all the fingerprint tampering can be identified include a region was copied from another part of the image or an entirely different image and so on.