CN103903271A - Image forensics method for natural image and compressed and tampered image based on DWT - Google Patents
Image forensics method for natural image and compressed and tampered image based on DWT Download PDFInfo
- Publication number
- CN103903271A CN103903271A CN201410144039.0A CN201410144039A CN103903271A CN 103903271 A CN103903271 A CN 103903271A CN 201410144039 A CN201410144039 A CN 201410144039A CN 103903271 A CN103903271 A CN 103903271A
- Authority
- CN
- China
- Prior art keywords
- image
- coefficient
- tampered
- carry out
- histogram
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Landscapes
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention provides an image forensics method for a natural image and a compressed and tampered image based on DWT. According to the method, the natural image and the compressed image based on the DWT can be effectively distinguished, meanwhile, good distinguishability is achieved on certain specific image tampering carrying out compression trace elimination on the compressed image, the joint probability histogram of a wavelet transform coefficient of the natural image and the tampered image is calculated through the method, the histogram is normalized, then Hough transform is carried out, the mean value, variance value, skewness value and kurtosis value of a Hough transform coefficient matrix are extracted as characteristic values of a support vector machine, and a training set is formed by the characteristic values. A classification model is generated by the support vector machine through the training set in a training mode, unknown characteristic value samples are classified through the model, and whether compression or anti-compression forensics processing is carried out on an image or not is judged. The method is stable in performance, easy and convenient to implement, efficient, high in accuracy and suitable for forensics detection of the natural image and the tampered image in other aspects.
Description
Technical field
The present invention relates to the technical field of image forensics, be specifically related to a kind of evidence collecting method for natural image and the image based on DWT compression tampered image.
Background technology
Along with the fast development of current multimedia technology and network technology, make transmission and shared multimedia become more convenient, people can such as, obtain for corresponding multimedia (picture, video, audio frequency etc.) fast by network and multimedia technology, not accommodating doubting, we live in the world of a vision purpose, and seeing is believing is our traditional idea cognition.But us are huge is also accompanied by for multimedia and distorts the digital technology of present stage bringing simultaneously easily, wherein distorting for image most importantly.The false of picture material is brought certain harm to people's life possibly, and for example South China Tiger event, has caused tremendous influence to entire society.
Traditional picture material protection is all to design according to external mode, for example, in image, add watermark or fingerprint.But under many circumstances, external protection scheme can not well be implemented, therefore need to study for the intrinsic fingerprint characteristic of image itself, thereby promote effectively carrying out of evidence obtaining work.But always there is two-sidedness in science, discuss out after the evidence obtaining detection for a kind of distorted image some researcher, have researcher and cover accordingly and attack processing for this detection, existing evidence obtaining detection method was lost efficacy.For example, in the compression evidence obtaining for image, there is researcher to compress according to the characteristics of image after compression and non-compression classification, thereby guarantee the authenticity of image.Meanwhile, just there is compression counter collect evidence of researcher for these methods, existing compression evidence collecting method was lost efficacy.Therefore, seek fingerprint characteristic intrinsic in natural image and just become particularly important.
In digital evidence obtaining, collecting evidence and compress anti-evidence obtaining for the compression of image mainly concentrates on jpeg image.Because in society, it is all jpeg format that people use the coded system of maximum cameras.In the detection for jpeg image compression algorithm, if Farid etc. is (referring to H.Farid.Digital ballistics from jpeg quantization:A follow up study.Dept.Comp.Sci..Dartmouth College, Tech.Rep.TR2008-638,2008.) adopt different JPEG quantization tables according to different camera lenss, image quantization to be detected is contrasted from the quantization table of the different camera lenses in the database existing, and is which kind of camera to take the information forming by thereby detect this image.Fan etc. are (referring to Z.Fan and R.deQueiroz.Identification of bitmap compression history:JPEG detection and quantizer estimation.IEEE Trans.Image Process.vol.12, no.2, pp.230 – 235, Feb.2003.) adopt the continuity of coefficient histogram of original image DCT and the DCT coefficient histogram of compression image afterwards to exist the character at interval to compress evidence obtaining.For the anti-forensic technologies of compression, Stamm etc. are (referring to Matthew C.Stamm, K.J.Ray Liu.Anti-Forensics of Digital Image Compression.IEEE TRANSACTIONS ON INFORMATION FORENSICS ANDSECURITY.VOL.6, NO.3, 2011.) for the interval problem between the DCT coefficient histogram after JPEG compression, use laplace model to carry out modeling for histogram, then shake interpolation for compartment, thereby fill up the interval between coefficient histogram, make Lin etc. (referring to W.S.Lin, S.K.Tjoa, H.V.Zhao, and K.J.R.Liu.Digital image source coder forensics via intrinsic fingerprints.IEEE Trans.Inf.Forensics Security.vol.4, no.3, pp.460 – 475, etc. Sep.2009.) the evidence obtaining algorithm proposing lost efficacy.
But, along with the fast development of digital technology, JPEG2000 and SPIHT(Set Partitioning in HierarchicalTrees, multistage tree set partitioning algorithm) these image compression algorithms based on DWT are also widely used.For the forensic technologies of DWT compressed image, Lin etc. are (referring to W.S.Lin, S.K.Tjoa, H.V.Zhao, and K.J.R.Liu.Digital image source coder forensics via intrinsic fingerprints.IEEE Trans.Inf.Forensics Security.vol.4, no.3, pp.460 – 475, Sep.2009.) adopt the detection of collecting evidence of interval between the coefficient histogram after conversion.Simultaneously, Farid etc. are (referring to Hany Farid and Siwei Lyu.Higher-order Wavelet Statistics and their Application to Digital Forensics.IEEE Workshop on Statistical in Computer Vision (in conjunction with CVPR), 2003.) after proposing to use image wavelet transform, linear relationship between father node between coefficient and child nodes and neighbor node and uncle's node is carried out natural image and compressed image, or the thought that the evidence obtaining between natural image and computer composograph detects.For anti-forensic technologies, Stamm etc. are (referring to M.C.Stamm and K.J.R.Liu.Wavelet-based image compression anti-forensics.in Proc.IEEE Int.Conf.Image Process.pp.1737 – 1740,2010.) the anti-forensic technologies for SPIHT compression algorithm is proposed, Main Basis laplace model carries out the filling of pectination coefficient histogram.
The method that simultaneously adopts the characteristic of the feature invariance based on natural image to carry out image forensics has also obtained research widely.As
deng (referring to Jan
jessica Fridrich.Calibration Revisited.MM & Sec'09Proceedings of the11th ACM workshop on Multimedia and security.pp.63-74, 2009.) propose based on natural image after carrying out the cutting of fraction pixel, the technology that the less characteristic of variation of entirety statistical property is collected evidence, and Valenzise etc. is (referring to Giuseppe Valenzise, Marco Tagliasacchi, Stefano Tubaro.Revealing the Traces of JPEG Compression Anti-Forensics.IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY.VOL.8, NO.2, 2013.) the evidence obtaining algorithm of the anti-forensic technologies that the variation based on total variance between different compressibilitys (Total Variation) proposing realizes.
In the time that compression of images is collected evidence, be easy to be made by some way this characteristic to be covered by other researchers for the mode that in specific compression algorithm, distinctive characteristic is collected evidence, thereby the algorithm that makes to collect evidence lost efficacy.And for researcher's anti-evidence obtaining algorithm, researcher can further find out again evidence obtaining algorithm, make whole evidence obtaining and anti-evidence obtaining enter a kind of competitive stage of vicious cycle.Therefore, adopting the peculiar characteristic of natural image to carry out image forensics, is a kind of trend and the study hotspot of future development.For the characteristic of wavelet transformation, seek the fingerprint characteristic of natural image after wavelet transformation, for the compression of image or the anti-evidence obtaining of compression, it is all had for different researchers and well can distinguish row, can effectively guarantee the authenticity of image, the game of simultaneously as far as possible forgoing between evidence obtaining and anti-evidence obtaining, for the authenticity of general image detects the more unified method of finding.
Summary of the invention
The technical problem to be solved in the present invention is: the deficiency that overcomes existing forensic technologies, provide a kind of for natural image with based on DWT(Discrete Wavelet Transform, wavelet transform) evidence collecting method of image of compression tampered image, the method is used the incidence relation between DWT different decomposition classification coefficient, obtain the difference of this incidence relation between natural image and tampered image, use Hough transformation is further processed, obtain difference characteristic value, realize the high-quality evidence obtaining for natural image and tampered image.
The technical solution adopted for the present invention to solve the technical problems: a kind of evidence collecting method for natural image and the image based on DWT compression tampered image, the method can be collected evidence for the image based on DWT compression, or collecting evidence of the image that can process for the anti-evidence obtaining of process, it is characterized in that: the joint probability histogram that calculates the wavelet conversion coefficient of natural image and tampered image, be normalized for this histogram, then carry out Hough transformation, extract the average of Hough transformation matrix of coefficients, variance yields, skewness value and kurtosis value are as the eigenwert of support vector machine, this eigenwert composition training set closes, support vector machine is used this training set training to generate disaggregated model, adopts this model to classify to unknown eigenwert sample, judges whether image is through distorting, and described process is distorted as processing through overcompression or through back-pressure contracting evidence obtaining, the eigenwert of extracting described support vector machine specifically comprises the following steps:
The first step, carries out wavelet transform to image, obtains corresponding wavelet coefficient, to rounding of the discrete wavelet coefficient operation obtaining, according to the characteristic of wavelet transformation, be N rank by picture breakdown, wherein in n rank, wherein n=1,2,, N – 1, according to wave filter service condition difference, it can be divided into L
nlow frequency sub-band, H
nhorizontal subband, D
nvertical subband and V
ndiagonal angle subband, then L
nlow frequency sub-band can continue to use wave filter to carry out filtering, obtains the L of n+1 level
n+1low frequency sub-band, H
n+1low frequency sub-band, D
n+1low frequency sub-band and V
n+1low frequency sub-band;
Second step, for the H after wavelet transformation, the subband of V and tri-directions of D carries out respectively joint probability distribution histogram calculation; In H direction, suppose at H
nin a coefficient H
n(r, s), wherein (r, s) represents positional information, defines its " father " coefficient to be
coefficient H so
n(r, s) is just corresponding " child " coefficient; According to this kind of corresponded manner, corresponding " father " coefficient of " child " coefficient, the relation between this in order to make " child " and " father " node is more obvious, adopts and take 10 the end of as, all coefficients is taken the logarithm; Then ask joint probability histogram for " child " and " father " coefficient; Histogrammic the solving of joint probability adopts formula (1) to carry out,
At Η
hin subscript H represents is horizontal direction, what wherein δ represented is target function, definition δ (x, y)=1, and if only if x=0 and y=0; Wherein [] represents to round, M
hbe illustrated in all " father " and " child " numbers of mapping relations one to one in H direction,
with
be illustrated in the value of " father " and " child " in i level DWT decomposition; Simultaneously for Η
vand Η
dalso adopt same account form; For
with
value, the range of definition be 100 ,-99 ..., 100}; Wherein k, l is in integer set Ζ;
The 3rd step, for the joint probability histogram obtaining, is normalized operation, adopts formula (2) to carry out,
Adopt this kind of mode, joint probability histogram is normalized to 0,1 ..., 255}, can become the span of a gray level image, then normalized joint probability histogram is used as to gray level image and processes;
The 4th step, adopts formula (3) to carry out Hough transformation operation for the probability histogram that obtains normalized three directions afterwards, is transformed in the coefficient domain of Hough transformation,
ρ＝xcosθ+ysinθ (3)
Wherein ρ represents the distance of initial point to image cathetus, and θ represents the corner dimension perpendicular to the vertical line of image cathetus and x axle;
The 5th step, averages for the coefficient domain of three different directions that obtain, variance, and skewness and kurtosis, and by four characteristic value combinations of three directions, form the proper vector of one 12 dimension.
Further, adopt SVM(Support Vector Machine, support vector machine) carry out the model training of data, the proper vector of natural image and tampered image is trained, obtain training pattern.
Further, use training pattern to carry out data training the feature value vector of the image of unknown classification, obtain training type.
Principle of the present invention is:
A kind of evidence collecting method for natural image and the image based on DWT compression tampered image of the present invention, comprise that image characteristics extraction, model training and model use three processes, image characteristics extraction need to be carried out corresponding feature extraction to natural image and tampered image; Leaching process comprises the wavelet transformation to image, then obtain the joint probability histogram between coefficient according to corresponding computing formula, carrying out part for joint probability histogram blocks, obtain having most the data of expression power, histogram is normalized simultaneously, finally carry out Hough transformation, obtain conversion coefficient afterwards, carry out the calculating of average, variance, skewness and kurtosis value for coefficient, the eigenwert of three different directions of associating, obtains one 12 proper vector of tieing up; For the proper vector training set of the natural image that comprises same number number and tampered image, adopt the mode of cross validation, obtain the most optimized parameter of svm classifier device, thereby obtain training pattern; For the proper vector of unknown classification, carry out the classification of training pattern, then obtain corresponding type output, obtain the differentiation classification of image;
The process of image characteristics extraction is:
(1) image is carried out to DWT wavelet transformation;
(2) for the coefficient obtaining after DWT conversion, get take 10 as low logarithm, and be multiplied by 10, the logarithm that is 0 coefficient is set as to 0;
(3) the system of logarithm matrix number to the processing obtaining with step (2), adopt " father " coefficient shown in Fig. 1 and the relation between " child " coefficient, obtain respectively DWT and convert father's node of rear three directions and the mapping relations pair one to one of child nodes, and limit the value of father's node and child nodes { 100,-99,, in 100};
(4) adopt formula (1) to carry out the histogrammic calculating of joint probability;
(5) adopt formula (2) to be normalized for the joint probability histogram that obtains, normalized to integer 0,1 ..., in the scope of 255};
(6) for the joint probability histogram obtaining, carry out rim detection;
(7) formula for data acquisition (2) step (6) being obtained carries out Hough transformation, obtains corresponding coefficient domain;
(8) coefficient domain obtaining is carried out to the calculating of average, variance, skewness and kurtosis value;
(9) eigenwert of three directions of associating, generates one 12 proper vector of tieing up;
Model training process is as follows:
(1) adopt SVM to carry out classification design;
(2) select the model of C-SVM as classification;
(3) data in training set are carried out to the model training of cross validation according to the natural image mode identical with tampered image data volume, then obtain the parameter of optimum C-SVM;
Unknown category classification process is as follows:
(1) image feature data of unknown classification is carried out to data requirement;
(2) adopt the model of training to classify, obtain classification results;
The advantage that the present invention compared with prior art had is:
(1) the method applied in the present invention is to distinguish based on the peculiar characteristic of natural image, and can be good at forgoing evidence obtaining and the anti-game between evidence obtaining, for later image forensics provides a kind of algorithm of popularity more that has.
(2) the present invention is simple for the feature extraction of image, and the dimension of the last proper vector generating is low, during for large-scale data, can have better time high efficiency.
(3) unique point of the present invention is high for natural image stability, has very strong robustness.
Accompanying drawing explanation
Fig. 1 is the relation between " father " and " child " node involved in the present invention;
Fig. 2 is the general frame process flow diagram in the present invention;
Fig. 3 is the compressed image evidence obtaining process flow diagram in the present invention;
Fig. 4 is the anti-evidence obtaining compressed image evidence obtaining process flow diagram in the present invention;
Fig. 5 is anti-evidence obtaining compressed image processing process flow diagram involved in the present invention.
Embodiment
Further illustrate the present invention below in conjunction with accompanying drawing and specific embodiment.
Before carrying out experimental implementation of the present invention, need to generate and process for experiment image used.The natural image that this experiment adopts is UCID-v2(G.Schaefer and M.Stich, " UCID-An uncompressed colour image database; " in Proc.SPIE:Storage and Retrieval Methods and Applications for Multimedia, 2004, vol.5307, pp.472 – 480) in 1338 images, for these 1338 images, compressibility from 0.5 to 8 be need to obtain, and 0.5 the compressed image of 16 kinds and the anti-image forensic of ad hoc approach are spaced apart.First be the acquisition process of the compressed image based on SPIHT and JPEG2000, its acquisition process is as follows:
Step 1: arrange for SPIHT image compression algorithm code, the compression algorithm of JPEG2000 adopts the algorithm having existed in Matlab to carry out image processing simultaneously;
Step 2: for different images, carry out compressibility from 0.5 to 8, and be spaced apart 0.5 16 in the generation of compressed image, thereby obtain the image of the spiht algorithm compression of the different compressibilitys of 1338 × 16, and the compressed image of the JPEG2000 of 1338 × 16 different compressibilitys.
Step 3: for different compressibilitys and different compression algorithms, carry out different names and deposit operation for image.
Again, need to obtain the image through back-pressure contracting evidence obtaining algorithm process, the acquisition process of this image as shown in Figure 5, its acquisition process (take the image of SPIHT compression algorithm compression as example) as follows:
Step 1: for the image between different compressibilitys carries out 9/7DWT conversion through SPIHT compression algorithm, obtain the frequency domain information of respective image;
Step 2: in frequency domain, the wavelet coefficient histogram of different brackets and different sub-band carries out Laplace model parameter estimation;
Step 3: according to the model parameter obtaining, carry out the Jitter Calculation at the corresponding coefficient place of different brackets different sub-band;
Step 4: add shake to corresponding coefficient place, thereby obtain through the wavelet coefficient after processing;
Step 5: carry out anti-9/7DWT processing for the wavelet coefficient of processing;
Step 6: carry out image forensics operation, obtain the image of back-pressure compression algorithm processing, generate 1338 × 16 images through the processing of back-pressure compression algorithm herein.
Then, carry out experimental implementation of the present invention.As shown in Figure 2, whole evidence obtaining process comprises that image characteristics extraction, model training and model use three processes to detail flowchart of the present invention, and wherein of paramount importance part is exactly characteristic extraction procedure, as the characteristic extraction procedure being marked in Fig. 2; First leaching process needs image to carry out wavelet transformation, because lossy compression method in the standard of JPEG2000 is to adopt 9/7 wavelet transformation, therefore also adopt this small echo to carry out wavelet transform herein, the operation of then carrying out the calculating of probability histogram and further processing according to different directions; After proper vector has been asked for, need to carry out the model training of SVM, adopt the mode of cross validation to realize parameter estimation, finally carry out discriminant classification for unknown images.
In the present invention, for evidence obtaining be compressed image and the compressed image processed through anti-evidence obtaining, just as shown in Figure 3 and Figure 4, its flow process is the same, is the image difference of processing, is collectively referred to as tampered image herein carries out the explanation of whole implementation operating process by two.Whole implementation step is as follows:
The process of image characteristics extraction is:
Step 1: the image by natural image and after distorting carries out DWT wavelet transformation, adopts 9/7 small echo to carry out wavelet transformation herein;
Step 2: get take 10 as low logarithm for the coefficient obtaining, and be multiplied by 10 and then expand the otherness of natural image and tampered image, the logarithm that is 0 coefficient is set as to 0;
Step 3: to the system of logarithm matrix number obtaining with step 2, for three different directions in DWT conversion, according to the relation between the father shown in Fig. 1 and child nodes, find one one-to-one correspondence pair, and the value that limits father's node and child nodes is { 100 ,-99,, in 100};
Step 4: the joint probability histogram that calculates different directions according to the form of formula (1);
Step 5: adopt formula (2) to be normalized for the joint probability histogram that obtains, normalized to integer 0,1 ..., in the scope of 255};
Step 6: carry out rim detection for the joint probability histogram obtaining, use Sobel operator to carry out the calculating of rim detection herein;
Step 7: the formula for data acquisition (2) that step 5 is obtained carries out Hough transformation, obtains corresponding coefficient domain;
Step 8: the coefficient domain obtaining is carried out to the calculating of average, variance, skewness and kurtosis value;
Step 9: the eigenwert of three directions of associating, generates one 12 proper vector of tieing up;
Model training process is as follows:
Step 1: adopt SVM to carry out classification design, use the libsvm having designed to train herein;
Step 2: select the model of C-SVM as classification;
Step 3: the data in training set are carried out to the model training of cross validation according to the natural image mode identical with tampered image data volume, then obtain the parameter of optimum C-SVM, can adopt the algorithm of the training optimum solution having existed herein;
Unknown category classification process is as follows:
Step 1: the image feature data of unknown classification is carried out to the normalization of libsvm data layout;
Step 2: adopt the model of training to classify, obtain classification results;
Experiment effect of the present invention, this place adopts AUC(Area under curve, area under curve) describe for the effect of experiment, AUC more approaches 1, and illustrative experiment effect is better.Experiment effect of the present invention can reach more than 0.97.
Table 1 has shown and the present invention is directed to SPIHT, the AUC value that JPEG2000 compressed image is collected evidence, and for the AUC value of the evidence obtaining of SPIHT back-pressure contracting evidence obtaining algorithm process image.
Table 1 image forensics AUC value table
Claims (3)
1. the evidence collecting method for natural image and the image based on DWT compression tampered image, the method can be collected evidence for the image based on DWT compression, or collecting evidence of the image that can process for the anti-evidence obtaining of process, it is characterized in that: the joint probability histogram that calculates the wavelet conversion coefficient of natural image and tampered image, be normalized for this histogram, then carry out Hough transformation, extract the average of Hough transformation matrix of coefficients, variance yields, skewness value and kurtosis value are as the eigenwert of support vector machine, and this eigenwert composition training set closes; Support vector machine is used this training set training to generate disaggregated model, adopts this model to classify to unknown eigenwert sample, judges whether image is through distorting, and described process is distorted as processing through overcompression or through back-pressure contracting evidence obtaining; The eigenwert of extracting described support vector machine specifically comprises the following steps:
The first step, carries out wavelet transform to image, obtains corresponding wavelet coefficient, to rounding of the discrete wavelet coefficient operation obtaining, according to the characteristic of wavelet transformation, be N rank by picture breakdown, wherein in n rank, wherein n=1,2,, N – 1, according to wave filter service condition difference, it can be divided into L
nlow frequency sub-band, H
nhorizontal subband, D
nvertical subband and V
ndiagonal angle subband, then L
nlow frequency sub-band can continue to use wave filter to carry out filtering, obtains
n+1the L of level
n+1low frequency sub-band, H
n+1low frequency sub-band, D
n+1low frequency sub-band and V
n+1low frequency sub-band;
Second step, for the H after wavelet transformation, the subband of V and tri-directions of D carries out respectively joint probability distribution histogram calculation; In H direction, suppose at H
nin a coefficient H
n(r, s), wherein (r, s) represents positional information, defines its " father " coefficient to be
coefficient H so
n(r, s) is just corresponding " child " coefficient; According to this kind of corresponded manner, corresponding " father " coefficient of " child " coefficient, the relation between this in order to make " child " and " father " node is more obvious, adopts and take 10 the end of as, all coefficients is taken the logarithm; Then ask joint probability histogram for " child " and " father " coefficient; Histogrammic the solving of joint probability adopts formula (1) to carry out,
At Η
hin subscript H represents is horizontal direction, what wherein δ represented is target function, definition δ (x, y)=1, and if only if x=0 and y=0; Wherein [] represents to round, M
hbe illustrated in all " father " and " child " numbers of mapping relations one to one in H direction,
with
be illustrated in the value of " father " and " child " in i level DWT decomposition; Simultaneously for Η
vand Η
dalso adopt same account form; For
with
value, the range of definition be 100 ,-99 ..., 100}; Wherein k, l is in integer set Ζ;
The 3rd step, for the joint probability histogram obtaining, is normalized operation, adopts formula (2) to carry out,
Adopt this kind of mode, joint probability histogram is normalized to 0,1 ..., 255}, can become the span of a gray level image, then normalized joint probability histogram is used as to gray level image and processes;
The 4th step, adopts formula (3) to carry out Hough transformation operation for the probability histogram that obtains normalized three directions afterwards, is transformed in the coefficient domain of Hough transformation,
ρ＝xcosθ+ysinθ (3)
Wherein ρ represents the distance of initial point to image cathetus, and θ represents the corner dimension perpendicular to the vertical line of image cathetus and x axle;
The 5th step, averages for the coefficient domain of three different directions that obtain, variance, and skewness and kurtosis, and by four characteristic value combinations of three directions, form the proper vector of one 12 dimension.
2. a kind of evidence collecting method for natural image and the image based on DWT compression tampered image according to claim 1, it is characterized in that, adopt SVM(Support Vector Machine, support vector machine) carry out the model training of data, the proper vector of natural image and tampered image is trained, obtain training pattern.
3. a kind of evidence collecting method for natural image and the image based on DWT compression tampered image according to claim 1, is characterized in that, uses training pattern to carry out data training the feature value vector of the image of unknown classification, obtains training type.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410144039.0A CN103903271B (en) | 2014-04-11 | 2014-04-11 | Image forensics method for natural image and compressed and tampered image based on DWT |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410144039.0A CN103903271B (en) | 2014-04-11 | 2014-04-11 | Image forensics method for natural image and compressed and tampered image based on DWT |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103903271A true CN103903271A (en) | 2014-07-02 |
| CN103903271B CN103903271B (en) | 2017-01-18 |
Family
ID=50994578
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410144039.0A Expired - Fee Related CN103903271B (en) | 2014-04-11 | 2014-04-11 | Image forensics method for natural image and compressed and tampered image based on DWT |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103903271B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105513059A (en) * | 2015-12-01 | 2016-04-20 | 北京航空航天大学 | Urban multi-mode media data consistency detection method |
| CN106056639A (en) * | 2016-05-24 | 2016-10-26 | 大连理工大学 | Steganography method based on camera source anti-forensics |
| CN107025647A (en) * | 2017-03-09 | 2017-08-08 | 中国科学院自动化研究所 | Distorted image evidence collecting method and device |
| CN110555792A (en) * | 2019-08-16 | 2019-12-10 | 广东外语外贸大学南国商学院 | Image tampering blind detection method based on normalized histogram comprehensive feature vector |
| CN113259665A (en) * | 2020-02-07 | 2021-08-13 | 华为技术有限公司 | Image processing method and related equipment |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102184537A (en) * | 2011-04-22 | 2011-09-14 | 西安理工大学 | Image region tamper detection method based on wavelet transform and principal component analysis |
| CN102609947A (en) * | 2012-02-10 | 2012-07-25 | 浙江理工大学 | Forgery detection method for spliced and distorted digital photos |
| CN102968793A (en) * | 2012-11-20 | 2013-03-13 | 百年金海安防科技有限公司 | Method for identifying natural image and computer generated image based on DCT (Discrete Cosine Transformation)-domain statistic characteristics |
| CN103034995A (en) * | 2012-11-18 | 2013-04-10 | 百年金海安防科技有限公司 | Splicing image detecting algorithm based on statistical properties of Benford rule |
-
2014
- 2014-04-11 CN CN201410144039.0A patent/CN103903271B/en not_active Expired - Fee Related
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102184537A (en) * | 2011-04-22 | 2011-09-14 | 西安理工大学 | Image region tamper detection method based on wavelet transform and principal component analysis |
| CN102609947A (en) * | 2012-02-10 | 2012-07-25 | 浙江理工大学 | Forgery detection method for spliced and distorted digital photos |
| CN103034995A (en) * | 2012-11-18 | 2013-04-10 | 百年金海安防科技有限公司 | Splicing image detecting algorithm based on statistical properties of Benford rule |
| CN102968793A (en) * | 2012-11-20 | 2013-03-13 | 百年金海安防科技有限公司 | Method for identifying natural image and computer generated image based on DCT (Discrete Cosine Transformation)-domain statistic characteristics |
Non-Patent Citations (1)
| Title |
|---|
| KWANG-FU LI ET AL: "Image Tamper Detection and Recovery System Based on Discrete Wavelet Transformation", 《COMMUNICATIONS, COMPUTERS AND SIGNAL PROCESSING, 2001》 * |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105513059A (en) * | 2015-12-01 | 2016-04-20 | 北京航空航天大学 | Urban multi-mode media data consistency detection method |
| CN106056639A (en) * | 2016-05-24 | 2016-10-26 | 大连理工大学 | Steganography method based on camera source anti-forensics |
| CN106056639B (en) * | 2016-05-24 | 2019-05-14 | 大连理工大学 | It is a kind of based on the anti-hidden decryption method collected evidence of camera source |
| CN107025647A (en) * | 2017-03-09 | 2017-08-08 | 中国科学院自动化研究所 | Distorted image evidence collecting method and device |
| CN107025647B (en) * | 2017-03-09 | 2020-02-28 | 中国科学院自动化研究所 | Image tampering evidence obtaining method and device |
| CN110555792A (en) * | 2019-08-16 | 2019-12-10 | 广东外语外贸大学南国商学院 | Image tampering blind detection method based on normalized histogram comprehensive feature vector |
| CN110555792B (en) * | 2019-08-16 | 2022-05-17 | 广东外语外贸大学南国商学院 | Image tampering blind detection method based on normalized histogram comprehensive feature vector |
| CN113259665A (en) * | 2020-02-07 | 2021-08-13 | 华为技术有限公司 | Image processing method and related equipment |
| CN113259665B (en) * | 2020-02-07 | 2022-08-09 | 华为技术有限公司 | Image processing method and related equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103903271B (en) | 2017-01-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Monga et al. | Perceptual image hashing via feature points: performance evaluation and tradeoffs | |
| Gul et al. | SVD-based universal spatial domain image steganalysis | |
| CN103903271A (en) | Image forensics method for natural image and compressed and tampered image based on DWT | |
| WO2017080196A1 (en) | Video classification method and device based on human face image | |
| CN104778457A (en) | Video face identification algorithm on basis of multi-instance learning | |
| Stamm et al. | Wavelet-based image compression anti-forensics | |
| CN104661037B (en) | The detection method and system that compression image quantization table is distorted | |
| CN110457996B (en) | Video moving object tampering evidence obtaining method based on VGG-11 convolutional neural network | |
| CN105631473A (en) | Camera source identification method in finite labeled sample condition | |
| CN104392207B (en) | A kind of feature coding method recognized for digital image content | |
| Mukhedkar et al. | Fast face recognition based on Wavelet Transform on PCA | |
| CN102547477B (en) | Video fingerprint method based on contourlet transformation model | |
| CN103561274A (en) | Video time domain tamper detection method for removing moving object shot by static camera lens | |
| CN104778689A (en) | Image digest method based on mean secondary image and locality preserving projection | |
| Luo et al. | Image universal steganalysis based on best wavelet packet decomposition | |
| Barbhuiya et al. | Wavelet tranformations & its major applications in digital image processing | |
| Chhikara et al. | A review on digital image steganalysis techniques categorised by features extracted | |
| Atta et al. | An efficient face recognition system based on embedded DCT pyramid | |
| Si et al. | A full‐reference stereoscopic image quality assessment index based on stable aggregation of monocular and binocular visual features | |
| Li et al. | A novel blind detector for additive noise steganography in JPEG decompressed images | |
| Wang et al. | Exploring DCT coefficient quantization effect for image tampering localization | |
| Muthuramalingam et al. | Stego anomaly detection in images exploiting the curvelet higher order statistics using evolutionary support vector machine | |
| Du et al. | Image hashing for tamper detection with multiview embedding and perceptual saliency | |
| Sandeep et al. | Perceptual video hashing based on the Achlioptas's random projections | |
| CN110210489A (en) | Median filtering evidence obtaining recognition methods based on arma modeling sum number to histogram |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20170118 Termination date: 20170411 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |