CN103413336A - Grid non-aligned double-JPEG-compression detecting method and device - Google Patents

Grid non-aligned double-JPEG-compression detecting method and device Download PDF

Info

Publication number
CN103413336A
CN103413336A CN2013103302691A CN201310330269A CN103413336A CN 103413336 A CN103413336 A CN 103413336A CN 2013103302691 A CN2013103302691 A CN 2013103302691A CN 201310330269 A CN201310330269 A CN 201310330269A CN 103413336 A CN103413336 A CN 103413336A
Authority
CN
China
Prior art keywords
matrix
row
picture element
vector
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
Application number
CN2013103302691A
Other languages
Chinese (zh)
Other versions
CN103413336B (en
Inventor
王军龙
杨建权
朱国普
黄晓霞
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shenzhen Institute of Advanced Technology of CAS
Original Assignee
Shenzhen Institute of Advanced Technology of CAS
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Shenzhen Institute of Advanced Technology of CAS filed Critical Shenzhen Institute of Advanced Technology of CAS
Priority to CN201310330269.1A priority Critical patent/CN103413336B/en
Publication of CN103413336A publication Critical patent/CN103413336A/en
Application granted granted Critical
Publication of CN103413336B publication Critical patent/CN103413336B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Landscapes

  • Compression Or Coding Systems Of Tv Signals (AREA)
  • Compression Of Band Width Or Redundancy In Fax (AREA)

Abstract

The invention discloses a grid non-aligned double-JPEG-compression detecting method and device. The grid non-aligned double-JPEG-compression detecting method and device achieve detection of grid non-aligned double-JPEG-compression with low computation complexity. The method comprises the steps of acquiring the quantization table of a JPEG image to be detected; acquiring blocking effect histograms; establishing an original feature vector FV with values of the first T items of each original horizontal blocking effect histogram and each original longitudinal blocking effect histogram, establishing a calibration feature vector FVdel with values of the first T items of each calibration horizontal blocking effect histogram and each calibration longitudinal blocking effect histogram, determining a difference vector of FV and FVdel or the absolute value of each item of the difference vector, and establishing a final feature vector FVfin of the JPEG image to be detected with the difference vector or the absolute value of each item of the difference vector; selecting a support vector machine classifier corresponding to the quantization table according to the quantization table; inputting FVfin into the selected support vector machine classifier to judge whether the JPEG image to be detected is a grid non-aligned double-JPEG-compressed image. The grid non-aligned double-JPEG-compression detecting method and device have the advantages of being low in computation complexity and high in detection accuracy.

Description

Detection method and the device of the dual JPEG compression of a kind of grid non-alignment
Technical field
The present invention relates to image processing field, be specifically related to detection method and the device of the dual JPEG compression of a kind of grid non-alignment.
Background technology
JPEG (joint photographic experts group) (Joint Photographic Expert Group, JPEG) standard is a kind of Static Picture Compression standard of being formulated by JPEG, the compression of image being carried out according to this standard is called as the JPEG compression, and the image obtained by the JPEG compression is called jpeg image.JPEG compression can obtain higher compressibility in the situation that exceed loss image perception quality, thus be widely used in picture transmission, share and store.
The JPEG compression process comprises the following steps: at first a given image to be compressed is divided into the picture element matrix of image 8 * 8 block of pixels of non-overlapping copies; Then each block of pixels is carried out to discrete cosine transform (Discrete Cosine Transform, DCT) conversion, obtain the DCT matrix of coefficients; Then according to predefined quantization table (corresponding quality factor), each coefficient in the DCT matrix of coefficients is quantized; Finally the quantization DCT coefficient matrix is carried out to the entropy coding, and generate jpeg file.The jpeg decompression compression process is the inverse process of JPEG compression, that is: a given jpeg file, at first read the quantization table of its file header, and the code stream in file is carried out to the entropy decoding, obtains the quantization DCT coefficient matrix; Then according to quantization table, the quantization DCT coefficient matrix is carried out to inverse quantization, obtain inverse quantization DCT matrix of coefficients; Then inverse quantization DCT matrix of coefficients is carried out to piecemeal DCT inverse transformation, obtain the picture element matrix of floating type; Finally the floating type picture element matrix of gained is blocked and rounds operation, output integer picture element matrix.
Joint Photographic Experts Group is the Standard of image compression be most widely used at present, and for example, the personal digital camera is substantially all supported jpeg format, and acquiescence is preserved the image photographed with jpeg format mostly.In other words, the image by camera output has often just experienced a JPEG compression.When people edit at the jpeg image to photographing, may be only interested in a certain zone of image, and by the area-of-interest cutting out, save as jpeg format again so as Internet Transmission with share.During due to cutting, people often can deliberately not keep the piecemeal network of JPEG, therefore 63/64 probability is arranged is non-alignment to the piecemeal grid of JPEG compression (carry out camera inside) for the first time and JPEG compression for the second time (carrying out while preserving after cutting), that is to say, the jpeg image obtained by recompression after above-mentioned cutting is the dual compressed of a width grid non-alignment on very large probability.In addition, people may be by certain Target Segmentation in a width jpeg image out, paste on an other width carrier image to produce a splicing scene, perhaps with certain zone of a width jpeg image, cover its another one zone with the certain objects of erasing, these operations also probably produce the dual jpeg compressed image of grid non-alignment.
As can be known from foregoing description, along with the development of multimedia treatment technology, be no matter image processing terminal (for example, individual's camera) or image editing software, its function is become stronger day by day, and is easy to learn and use, thereby makes people become easy to operations such as the editor of image and modifications.In case image important, that process is distorted spreads widely on network, this may cause severe social influence and great economic loss.Therefore whether, how to differentiate that the authenticity of a width jpeg image more and more comes into one's own, detecting jpeg image is to differentiate a kind of important means of jpeg image authenticity through grid non-alignment dual compression.
Whether prior art provides a kind ofly lives through the method for the dual JPEG compression of grid non-alignment based on number of cycles figure (Integer Periodicity Map, IPM) for detection of jpeg image.This detection method is based on a kind of like this fact: if image to be detected was once crossed grid non-alignment dual compression, and so in 8 * 8 block of pixels, the reference position of 8 * 8 piecemeals in the time of must having a location of pixels to be JPEG compression for the first time.Therefore, if from this reference position, intercept successively 8 * 8 block of pixels, then each block of pixels is carried out to dct transform, calculate the histogram of DCT coefficient DC component, the gained histogram will present periodic peaks.Travel through each pixel of whole 8 * 8 piecemeals, calculate and to using the periodicity measurement of each location of pixels resulting DCT coefficient histogram during as reference position, form the periodicity measurement matrix of 8 * 8, i.e. IPM.By the homogeneity of calculating 63 IPM values, estimate and adjudicate: if homogeneity is estimated, be less than predefined threshold value, judge that image to be detected does not pass through grid non-alignment dual compression, otherwise, judge that it is through grid non-alignment dual compression.
The detection method that above-mentioned prior art provides could measure the periodicity that the DCT coefficient distributes, so computation complexity is high after need to carrying out 64 block DCT transforms to every image to be detected, and real-time is poor; Periodicity measurement is to noise-sensitive, and therefore in the situation that noise is larger, Detection accuracy is lower.
Summary of the invention
The embodiment of the present invention provides the detection method of the dual JPEG compression of a kind of grid non-alignment, realizes the detection to the dual JPEG compression of grid non-alignment with lower computation complexity.
The detection method of the dual JPEG compression of a kind of grid non-alignment, comprising: the quantization table QT that obtains jpeg image I to be detected; Obtain the blocking effect histogram, comprise following: obtain picture element matrix M after jpeg image I to be detected is decompressed, calculate respectively 8 original lateral blocks effect histograms and 8 original longitudinal block effect histograms of described picture element matrix M, and obtain picture element matrix M after picture element matrix M procession is cut out del, to M delCarry out successively after compression and decompression obtaining the picture element matrix M ' after decompress(ion) del, calculate respectively the picture element matrix M ' after described decompress(ion) del8 calibration lateral blocks effect histograms and 8 calibration longitudinal block effect histograms; The value formation length of getting the front T item of described each original lateral blocks effect histogram and original longitudinal block effect histogram is the original feature vector FV of 2 * 8 * T, and the value formation length of getting described each calibration lateral blocks effect histogram and the front T item of calibration longitudinal block effect histogram is the alignment features vector FV of 2 * 8 * T del, ask for described original feature vector FV and described alignment features vector FV delDifference vector or the every absolute value of described difference vector, with described difference vector or the every absolute value of described difference vector, form the final proper vector FV of described jpeg image to be detected finAccording to described quantization table QT, select the support vector machine classifier corresponding with described quantization table QT; By described final proper vector FV finThe support vector machine classifier of inputting described selection be take and adjudicated described jpeg image to be detected and whether be the dual jpeg compressed image of grid non-alignment.
A kind of method of extracting the jpeg image proper vector, described method comprises: obtain the blocking effect histogram, comprise following: obtain picture element matrix M after jpeg image I to be detected is decompressed, calculate respectively 8 original lateral blocks effect histograms and 8 original longitudinal block effect histograms of described picture element matrix M, and obtain picture element matrix M after picture element matrix M procession is cut out del, to M delCarry out successively after compression and decompression obtaining the picture element matrix M ' after decompress(ion) del, calculate respectively the picture element matrix M ' after described decompress(ion) del8 calibration lateral blocks effect histograms and 8 calibration longitudinal block effect histograms; The value formation length of getting the front T item of described each original lateral blocks effect histogram and original longitudinal block effect histogram is the original feature vector FV of 2 * 8 * T, and the value formation length of getting described each calibration lateral blocks effect histogram and the front T item of calibration longitudinal block effect histogram is the alignment features vector FV of 2 * 8 * T del, ask for described original feature vector FV and described alignment features vector FV delDifference vector or the every absolute value of described difference vector, with described difference vector or the every absolute value of described difference vector, form the final proper vector FV of described jpeg image to be detected fin.
The pick-up unit of the dual JPEG compression of a kind of grid non-alignment, described device comprises:
The quantization table acquisition module, be used to obtaining the quantization table QT of jpeg image I to be detected;
Blocking effect histogram acquisition module, be used to obtaining the blocking effect histogram, described blocking effect histogram acquisition module comprises following the first acquisition module and the second acquisition module, wherein:
Described the first acquisition module, obtain picture element matrix M after jpeg image I to be detected is decompressed, calculate respectively 8 original lateral blocks effect histograms and 8 original longitudinal block effect histograms of described picture element matrix M;
Described the second acquisition module, obtain picture element matrix M after picture element matrix M procession is cut out del, to M delCarry out successively after compression and decompression obtaining the picture element matrix M ' after decompress(ion) del, calculate respectively the picture element matrix M ' after described decompress(ion) del8 calibration lateral blocks effect histograms and 8 calibration longitudinal block effect histograms;
Proper vector builds module, for the value of getting the front T item of described each original lateral blocks effect histogram and original longitudinal block effect histogram, forming length is the original feature vector FV of 2 * 8 * T, and the value of getting described each calibration lateral blocks effect histogram and the front T item of calibration longitudinal block effect histogram forms the alignment features vector FV that length is 2 * 8 * T del, ask for described original feature vector FV and described alignment features vector FV delDifference vector or the every absolute value of described difference vector, with described difference vector or the every absolute value of described difference vector, form the final proper vector FV of described jpeg image to be detected fin
Select module, for according to described quantization table QT, select the support vector machine classifier corresponding with described quantization table QT;
Judging module, for by described final proper vector FV finThe support vector machine classifier of inputting described selection be take and adjudicated described jpeg image to be detected and whether be the dual jpeg compressed image of grid non-alignment.
A kind of device that extracts the jpeg image proper vector, described device comprises: blocking effect histogram acquisition module, be used to obtaining the blocking effect histogram, described blocking effect histogram acquisition module comprises following the first acquisition module and the second acquisition module, wherein:
Described the first acquisition module, obtain picture element matrix M after jpeg image I to be detected is decompressed, calculate respectively 8 original lateral blocks effect histograms and 8 original longitudinal block effect histograms of described picture element matrix M;
Described the second acquisition module, obtain picture element matrix M after picture element matrix M procession is cut out del, to M delCarry out successively after compression and decompression obtaining the picture element matrix M ' after decompress(ion) del, calculate respectively the picture element matrix M ' after described decompress(ion) del8 calibration lateral blocks effect histograms and 8 calibration longitudinal block effect histograms;
Proper vector builds module, for the value of getting the front T item of described each original lateral blocks effect histogram and original longitudinal block effect histogram, forming length is the original feature vector FV of 2 * 8 * T, and the value of getting described each calibration lateral blocks effect histogram and the front T item of calibration longitudinal block effect histogram forms the alignment features vector FV that length is 2 * 8 * T del, ask for described original feature vector FV and described alignment features vector FV delDifference vector or the every absolute value of described difference vector, with described difference vector or the every absolute value of described difference vector, form the final proper vector FV of described jpeg image to be detected fin.
Embodiment is as can be known from the invention described above, when the second order difference of calculating pixel value, only need on horizontal and vertical both direction, respectively calculate 8 blocking effect histograms is lateral blocks effect histogram and longitudinal block effect histogram, and whether based on these blocking effect histograms, obtain for adjudicating jpeg image to be detected is the proper vector of the dual jpeg compressed image of grid non-alignment.The detection method provided with prior art needs respectively 64 possible grid offset to be added up and compared, and the method that the embodiment of the present invention provides only needs to add up 16 times; On the other hand, the method provided due to the embodiment of the present invention is directly on picture element matrix, to carry out when extracting proper vector, avoided 64 possible grid offset are respectively carried out to a large amount of calculating that once piece discrete cosine transform brings, therefore reduced computation complexity, real-time is better.
The accompanying drawing explanation
Fig. 1 is the basic procedure schematic diagram of the detection of the dual JPEG compression of the grid non-alignment of the embodiment of the present invention;
Fig. 2 is that the part rows to picture element matrix M of the embodiment of the present invention intercepts the matrix schematic diagram that rear gained is new;
Fig. 3 is that the partial row to picture element matrix M of the embodiment of the present invention intercepts the matrix schematic diagram that rear gained is new;
Fig. 4 be the embodiment of the present invention to cutting apart gained (n-8)/8 a column split submatrix, vertically be arranged in order the matrix schematic diagram that gained is new;
Fig. 5 is cutting apart submatrix and laterally be arranged in order the matrix schematic diagram that gained is new cutting apart gained (m-8)/8 row of the embodiment of the present invention;
Fig. 6 is the basic procedure schematic diagram of the method for the extraction jpeg image proper vector that provides of the embodiment of the present invention;
Fig. 7 is the pick-up unit logical organization schematic diagram of the dual JPEG compression of the grid non-alignment that provides of the embodiment of the present invention;
Fig. 8 is the pick-up unit logical organization schematic diagram of the dual JPEG compression of the grid non-alignment that provides of another embodiment of the present invention;
Fig. 9 is the device logical organization schematic diagram of the extraction jpeg image proper vector that provides of the embodiment of the present invention;
Figure 10 is the device logical organization schematic diagram of the extraction jpeg image proper vector that provides of another embodiment of the present invention.
Embodiment
The embodiment of the present invention provides the detection method of the dual JPEG compression of a kind of grid non-alignment, comprising: the quantization table QT that obtains jpeg image I to be detected; Obtain the blocking effect histogram, comprise following: obtain picture element matrix M after jpeg image I to be detected is decompressed, calculate respectively 8 original lateral blocks effect histograms and 8 original longitudinal block effect histograms of described picture element matrix M, and obtain picture element matrix M after picture element matrix M procession is cut out del, to M delCarry out successively after compression and decompression obtaining the picture element matrix M ' after decompress(ion) del, calculate respectively the picture element matrix M ' after described decompress(ion) del8 calibration lateral blocks effect histograms and 8 calibration longitudinal block effect histograms; The value formation length of getting the front T item of described each original lateral blocks effect histogram and original longitudinal block effect histogram is the original feature vector FV of 2 * 8 * T, and the value formation length of getting described each calibration lateral blocks effect histogram and the front T item of calibration longitudinal block effect histogram is the alignment features vector FV of 2 * 8 * T del, ask for described original feature vector FV and described alignment features vector FV delDifference vector or the every absolute value of described difference vector, with described difference vector or the every absolute value of described difference vector, form the final proper vector FV of described jpeg image to be detected finAccording to described quantization table QT, select the support vector machine classifier corresponding with described quantization table QT; By described final proper vector FV finThe support vector machine classifier of inputting described selection be take and adjudicated described jpeg image to be detected and whether be the dual jpeg compressed image of grid non-alignment.The embodiment of the present invention also provides the pick-up unit of the dual JPEG compression of corresponding grid non-alignment.Below be elaborated respectively.
The basic procedure of the detection of the dual JPEG compression of the grid non-alignment of the embodiment of the present invention can, with reference to figure 1, mainly comprise step:
S101, obtain the quantization table QT of jpeg image I to be detected.
S102, obtain the blocking effect histogram.
When image adopts the JPEG compression, owing to using 8 * 8 block DCT transforms and quantification manner, can on jpeg image, stay the blocking effect vestige.For the jpeg image of single compression, the blocking effect vestige only occurs at the boundary of each 8 * 8 piecemeal, and for the jpeg image of non-alignment dual compression, the blocking effect vestige also can be observed in the inside of 8 * 8 piecemeals.Therefore, portray in the piece image piece and the blocking effect vestige of block boundary by complete, can judge whether a jpeg image to be detected experiences grid non-alignment dual compression.Particularly, in embodiments of the present invention, the blocking effect vestige is described by a proper vector, and at first the extraction of this proper vector needs to obtain the blocking effect histogram.
In embodiments of the present invention, obtain the blocking effect histogram by the following method formula realize:
After jpeg image I to be detected is decompressed, obtain picture element matrix M, calculate respectively 8 original lateral blocks effect histograms and 8 original longitudinal block effect histograms of described picture element matrix M; And
After being cut out, picture element matrix M procession obtains picture element matrix M del, to M delCarry out successively after compression and decompression obtaining the picture element matrix M ' after decompress(ion) del, calculate respectively the picture element matrix M ' after described decompress(ion) del8 calibration lateral blocks effect histograms and 8 calibration longitudinal block effect histograms.Particularly, delete that the 1st of picture element matrix M walks to that i is capable, the 1st row are to j row and capable, the last 8-j row of last 8-i (, 1≤i≤8,1≤j≤8, and i gets 8 when different with j) herein, obtain the picture element matrix M after cutting del, re-use quantization table QT to picture element matrix M delCarry out the JPEG compression, obtain one with reference to jpeg image I del.Finally, to reference jpeg image I delDecompress, obtain picture element matrix M ' del.
No matter be above-mentioned which kind of mode, only need to respectively calculate the blocking effect that 8 histograms may exist for portraying jpeg image on horizontal and vertical both direction.With respect to prior art, need respectively 64 possible grid offset to be added up, the present invention only needs to add up 16 times, and calculated amount significantly descends.
Below take picture element matrix M and be example as m * n rank matrix, m and n are 8 multiple, 8 original lateral blocks effect histograms and 8 histogrammic processes of original longitudinal block effect of calculating pixel matrix M respectively are described, particularly, comprise the following steps S1021 to step S1026:
S1021, the 1st row to n-2 row, the 2nd row to the n-1 row and the 3rd row to n row that intercept respectively described picture element matrix M obtain matrix A, B and C, intercept respectively the 1st of described picture element matrix M and walk to that n-2 is capable, the 2nd to walk to n-1 capable and the 3rd walk to capable matrix E, F and the G of obtaining of n.As shown in Figure 2, matrix E, F and G are as shown in Figure 3 for matrix A, B and C.
S1022, compute matrix M r=︱ A-2 * B+C ︱, M c=︱ E-2 * F+G ︱.
S1023, puncture table M rLast 6 row obtain the rank matrix M of m * (n-8) 1r, puncture table M cLast 6 row obtain (m-8) * n rank matrix M 1c.
S1024, by the rank matrix M of m * (n-8) 1rEvery 8 column splits, become submatrix, can obtain altogether (n-8)/8 a column split submatrix, by (m-8) * n rank matrix M 1cEvery 8 row, be divided into submatrix, can obtain altogether (m-8)/8 a horizontal submatrix of cutting apart.
S1025, vertically be arranged in order the column split submatrix and obtain (m(n-8)/8) * 8 rank matrix M ' r, row is cut apart to submatrix and laterally is arranged in order one 8 * ((m-8) n/8) rank matrix M of acquisition ' c.M ' rAs shown in Figure 4, matrix M ' cAs shown in Figure 5.
S1026, compute matrix M ' rThe normalization histogram of each row Obtain longitudinal block effect histogram, compute matrix M ' cThe normalization histogram of every delegation h c k ( t ) = Σ i = 1 ( m - 8 ) n / 8 δ ( t - M c ′ ( k , i ) ) Obtain lateral blocks effect histogram, wherein, Σ t h r k ( t ) = 1 , Σ t h c k ( t ) = 1 , k = 1,2 , . . . , 8 .
S103, the value formation length of getting the front T item of each original lateral blocks effect histogram and original longitudinal block effect histogram is the original feature vector FV of 2 * 8 * T, and value formation length of getting each calibration lateral blocks effect histogram and the front T item of calibration longitudinal block effect histogram is the alignment features vector FV of 2 * 8 * T del, ask for described original feature vector FV and described alignment features vector FV delDifference vector or the every absolute value of described difference vector, with described difference vector or the every absolute value of described difference vector, form the final proper vector FV of described jpeg image to be detected fin.
It should be noted that, consider the image of single JPEG compression and the image of the dual JPEG compression of grid non-alignment, the histogrammic difference of its blocking effect mainly concentrates on front end, therefore, each lateral blocks effect histogram and the longitudinal block effect histogram that obtain for above-described embodiment, the value that can get its front T item forms a proper vector, can simplify thus the calculating of proper vector.For example, for the original feature vector obtained, be designated as
Figure BDA00003602256900085
Wherein, k=1,2 ..., 8, t=1,2 ..., T, thus the length of original feature vector FV is 2 * 8 * T.The value constitutive characteristic vector FV of T item before getting each calibration lateral blocks effect histogram and calibration longitudinal block effect histogram del(when its length is 2 * 8 * T), as a preferred embodiment of the invention, can ask for original feature vector FV and alignment features vector FV delDifference vector or the every absolute value of described difference vector, with described difference vector or the every absolute value of described difference vector, form the final proper vector FV of described jpeg image to be detected fin.Obviously, according to above-described embodiment, FV herein finLength be still 2 * 8 * T.
S104, according to the quantization table QT of jpeg image I to be detected, select the support vector machine classifier corresponding with this quantization table QT.
So-called support vector machine (Support Vector Machine, SVM) is a kind of machine learning method be most widely used at present.The cardinal principle of SVM is: one, by maximizing class interval, find optimum interphase; Two, by core, shine upon sample vector is transformed to higher dimensional space by lower dimensional space.Based on above-mentioned principle, SVM trains sorter by using through the sample of mark, makes it to have the ability that the new samples classification is judged.In order to adjudicate, the method for the embodiment of the present invention also comprises structure svm classifier device.The quantization table of the compression of JPEG for the first time is QT1 if remember, the quantization table of JPEG compression for the second time is QT2, and the structure svm classifier device process that provides of the embodiment of the present invention is as follows:
A) collect the image construction training plan image set S of uncompressed, in note S, the image number is N;
B) use quantization table QT2 to carry out the JPEG compression to the every width image in S, remember the image construction single compressed image collection S after these compressions s, the image number is still N;
C) set the training area R of QT1 QT1, note R QT1Element number be w, the traversal R QT1In each value, the every width image in S is carried out to the JPEG compression, can obtain altogether wN and open the single compressed image;
D) every compressed image wN being opened in the single compressed image carries out decompress(ion), obtain its picture element matrix, picture element matrix to gained carries out random cropping: the 1st～u that dismisses picture element matrix is capable, 1st～v is listed as and capable, the last 8～v row of last 8～u (1≤u≤8,1≤v≤8, u, v are 8 while being random value and difference).Then, the picture element matrix of service property (quality) factor Q T2 after to these cuttings carries out the JPEG compression again, obtains the jpeg image collection S through grid non-alignment dual compression d, S dMiddle image number is wN;
E) adopt in accompanying drawing 1 step S102 to the proper vector extracting method of step S103 example, calculating chart image set S sAnd S dIn the proper vector of every width image, the proper vector of gained, for training the svm classifier device, is obtained to the svm classifier device SVM corresponding with QT2 QF2.
S105, the proper vector FV that step S103 is obtained finThe support vector machine sorter that input step S104 selects, take and adjudicate jpeg image to be detected and whether be the dual jpeg compressed image of grid non-alignment.
Due to svm classifier device in step S104, in advance through training, therefore, need only the proper vector FV that step S103 is obtained finWhether wherein, just can adjudicate jpeg image to be detected is the dual jpeg compressed image of grid non-alignment in input.
The detection method that the dual JPEG of grid non-alignment provided from the invention described above embodiment compresses is as can be known, when the second order difference of calculating pixel value, only need on horizontal and vertical both direction, respectively calculate 8 blocking effect histograms is lateral blocks effect histogram and longitudinal block effect histogram, and whether based on these blocking effect histograms, obtain for adjudicating jpeg image to be detected is the proper vector of the dual jpeg compressed image of grid non-alignment.The detection method provided with prior art needs respectively 64 possible grid offset to be added up and compared, the method that the embodiment of the present invention provides only needs to add up 16 times, on the other hand, the method provided due to the embodiment of the present invention is directly on picture element matrix, to carry out when extracting proper vector, avoided 64 possible grid offset are respectively carried out to a large amount of calculating that once piece discrete cosine transform brings, therefore reduced computation complexity, real-time is better.
For the method that illustrates that better the embodiment of the present invention provides, below provide in the method for accompanying drawing 1 example when the value of parameters is following data, in different quantization table QT1 and quantization table QT2 situation, the accuracy rate of the detection method of the dual JPEG compression of grid non-alignment:
In m * n picture element matrix M, m=512, n=512, T=3, i=4, j=4, while building the svm classifier device, the number N=500 of image in image set S, the training area R of quality Q T1 QF1As shown in QT1 hurdle in following table 1.In this jpeg image judging process to be detected, used the image of other 500 512 * 512 to test.
Under above-mentioned parameter, the accuracy rate of the detection method of the dual JPEG compression of the grid non-alignment of accompanying drawing 1 example is as shown in table 1 below:
Figure BDA00003602256900101
Figure BDA00003602256900111
Table 1
In 5 kinds of situations of QT1=QT2, the average detected accuracy rate of the detection method of the dual JPEG compression of the grid non-alignment that the embodiment of the present invention provides is 83.8%, in 10 kinds of situations of QT1<QT2, the average detected accuracy rate of the detection method of the dual JPEG compression of the grid non-alignment that the embodiment of the present invention provides is 95.2%.
Referring to accompanying drawing 6, is the basic procedure of the method for the extraction jpeg image proper vector that provides of the embodiment of the present invention, mainly comprises step:
S601, obtain the blocking effect histogram.
In embodiments of the present invention, obtain the blocking effect histogram by the following method formula realize:
After jpeg image I to be detected is decompressed, obtain picture element matrix M, calculate respectively 8 original lateral blocks effect histograms and 8 original longitudinal block effect histograms of described picture element matrix M; And
After being cut out, picture element matrix M procession obtains picture element matrix M del, to M delCarry out successively after compression and decompression obtaining the picture element matrix M ' after decompress(ion) del, calculate respectively the picture element matrix M ' after described decompress(ion) del8 calibration lateral blocks effect histograms and 8 calibration longitudinal block effect histograms.Particularly, delete that the 1st of picture element matrix M walks to that i is capable, the 1st row are to j row and capable, the last 8-j row of last 8-i (, 1≤i≤8,1≤j≤8, and i gets 8 when different with j) herein, obtain the picture element matrix M after cutting del, re-use quantization table to picture element matrix M delCarry out the JPEG compression, obtain one with reference to jpeg image I del.Finally, to reference jpeg image I delDecompress, obtain picture element matrix M ' del.
No matter be above-mentioned which kind of mode, only need to respectively calculate the blocking effect that 8 histograms may exist for portraying jpeg image on horizontal and vertical both direction.With respect to prior art, need respectively 64 possible grid offset to be added up, the present invention only needs to add up 16 times, and calculated amount significantly descends.
Below take picture element matrix M and be example as m * n rank matrix, m and n are 8 multiple, 8 original lateral blocks effect histograms and 8 histogrammic processes of original longitudinal block effect of calculating pixel matrix M respectively are described, particularly, comprise the following steps S6021 to step S6026:
S6021, the 1st row to n-2 row, the 2nd row to the n-1 row and the 3rd row to n row that intercept respectively described picture element matrix M obtain matrix A, B and C, intercept respectively the 1st of described picture element matrix M and walk to that n-2 is capable, the 2nd to walk to n-1 capable and the 3rd walk to capable matrix E, F and the G of obtaining of n.As shown in Figure 2, matrix E, F and G are as shown in Figure 3 for matrix A, B and C.
S6022, compute matrix M r=︱ A-2 * B+C ︱, M c=︱ E-2 * F+G ︱.
S6023, puncture table M rLast 6 row obtain the rank matrix M of m * (n-8) 1r, puncture table M cLast 6 row obtain (m-8) * n rank matrix M 1c.
S6024, by m * (n-8) the rank matrix becomes submatrix every 8 column splits, can obtain (n-8)/8 a column split submatrix altogether, and (m-8) * n rank are divided into to submatrix every 8 row, can obtain altogether (m-8)/8 a horizontal submatrix of cutting apart.
S6025, vertically be arranged in order the column split submatrix and obtain (m(n-8)/8) * 8 rank matrix M ' r, row is cut apart to submatrix and laterally is arranged in order one 8 * ((m-8) n/8) rank matrix M of acquisition ' c.M ' rAs shown in Figure 4, matrix M ' cAs shown in Figure 5.
S6026, compute matrix M ' rThe normalization histogram of each row
Figure BDA00003602256900121
Obtain longitudinal block effect histogram, compute matrix M ' cThe normalization histogram of every delegation h c k ( t ) = &Sigma; i = 1 ( m - 8 ) n / 8 &delta; ( t - M c &prime; ( k , i ) ) Obtain lateral blocks effect histogram, wherein, &Sigma; t h r k ( t ) = 1 , &Sigma; t h c k ( t ) = 1 , k = 1,2 , . . . , 8 .
S602, the value formation length of getting the front T item of each original lateral blocks effect histogram and original longitudinal block effect histogram is the original feature vector FV of 2 * 8 * T, and value formation length of getting each calibration lateral blocks effect histogram and the front T item of calibration longitudinal block effect histogram is the alignment features vector FV of 2 * 8 * T del, ask for described original feature vector FV and described alignment features vector FV delDifference vector or the every absolute value of described difference vector, with described difference vector or the every absolute value of described difference vector, form the final proper vector FV of described jpeg image to be detected fin.
It should be noted that, consider the image of single JPEG compression and the image of the dual JPEG compression of grid non-alignment, the histogrammic difference of its blocking effect mainly concentrates on front end, therefore, each lateral blocks effect histogram and the longitudinal block effect histogram that obtain for above-described embodiment, the value that can get its front T item forms a proper vector, can simplify thus the calculating of proper vector.For example, for the original feature vector obtained, be designated as Wherein, k=1,2 ..., 8, t=1,2 ..., T, thus the length of original feature vector FV is 2 * 8 * T.The value constitutive characteristic vector FV of T item before getting each calibration lateral blocks effect histogram and calibration longitudinal block effect histogram del(when its length is 2 * 8 * T), as a preferred embodiment of the invention, can ask for original feature vector FV and alignment features vector FV delDifference vector or the every absolute value of described difference vector, with described difference vector or the every absolute value of described difference vector, form the final proper vector FV of described jpeg image to be detected fin.Obviously, according to above-described embodiment, FV herein finLength be still 2 * 8 * T.
Below the pick-up unit of the dual JPEG compression of grid non-alignment of the embodiment of the present invention of the detection method be used to carrying out the dual JPEG of above-mentioned grid non-alignment compression described, its basic logical structure is with reference to figure 7, comprise that mainly quantization table acquisition module 701, blocking effect histogram acquisition module 702, proper vector build module 703, select module 704 and judging module 705, wherein, blocking effect histogram acquisition module 702 comprises the first acquisition module 7021 and the second acquisition module 7022, and each module is described in detail as follows:
Quantization table acquisition module 701, be used to obtaining the quantization table QT of jpeg image I to be detected;
Blocking effect histogram acquisition module 702, be used to obtaining the blocking effect histogram;
The first acquisition module 7021, obtain picture element matrix M after jpeg image I to be detected is decompressed, calculate respectively 8 original lateral blocks effect histograms and 8 original longitudinal block effect histograms of described picture element matrix M;
The second acquisition module 7022, obtain picture element matrix M after picture element matrix M procession is cut out del, to M delCarry out successively after compression and decompression obtaining the picture element matrix M ' after decompress(ion) del, calculate respectively the picture element matrix M ' after described decompress(ion) del8 calibration lateral blocks effect histograms and 8 calibration longitudinal block effect histograms;
Proper vector builds module 703, for the value of getting the front T item of described each original lateral blocks effect histogram and original longitudinal block effect histogram, forming length is the original feature vector FV of 2 * 8 * T, and the value of getting described each calibration lateral blocks effect histogram and the front T item of calibration longitudinal block effect histogram forms the alignment features vector FV that length is 2 * 8 * T del, ask for described original feature vector FV and described alignment features vector FV delDifference vector or the every absolute value of described difference vector, with described difference vector or the every absolute value of described difference vector, form the final proper vector FV of described jpeg image to be detected fin
Select module 704, for according to described quantization table QT, select the support vector machine classifier corresponding with described quantization table QT;
Judging module 705, for by described final proper vector FV finThe support vector machine classifier of inputting described selection be take and adjudicated described jpeg image to be detected and whether be the dual jpeg compressed image of grid non-alignment.
In the pick-up unit of the dual JPEG compression of the grid non-alignment of accompanying drawing 7 examples, picture element matrix M can be m * n rank matrix, and wherein, m and n are 8 multiple.The first acquisition module 7021 of accompanying drawing 7 examples can comprise interception unit 801, computing unit 802, delete cells 803, cutting unit 804, arrangement units 805 and normalization unit 806, the pick-up unit of the dual JPEG compression of the grid non-alignment that provides of another embodiment of the present invention as shown in Figure 8, wherein:
Interception unit 801, for the 1st row to n-2 row, the 2nd row to the n-1 row and the 3rd row to n row that intercept respectively described picture element matrix M, obtain matrix A, B and C, intercept respectively the 1st of described picture element matrix M and walk to that n-2 is capable, the 2nd to walk to n-1 capable and the 3rd walk to capable matrix E, F and the G of obtaining of n;
Computing unit 802, for compute matrix M r=︱ A-2 * B+C ︱, M c=︱ E-2 * F+G ︱;
Delete cells 803, be used to deleting described matrix M rLast 6 row obtain the rank matrix of m * (n-8), delete described matrix M cLast 6 row obtain (m-8) * n rank matrixes;
Cutting unit 804, be used for described m * (n-8) the rank matrix becomes submatrix every 8 column splits, can obtain altogether (n-8)/8 a column split submatrix, described (m-8) * n rank are divided into to submatrix every 8 row, can obtain altogether (m-8)/8 a horizontal submatrix of cutting apart;
Arrangement units 805, obtain (m(n-8)/8 for described column split submatrix vertically is arranged in order) * 8 rank matrix M ' r, described row is cut apart to submatrix and laterally is arranged in order acquisition 8 * ((m-8) n/8) rank matrix M ' c
Normalization unit 806, be used to calculating described matrix M ' rThe normalization histogram of each row h c k ( t ) = &Sigma; i = 1 ( m - 8 ) n / 8 &delta; ( t - M c &prime; ( k , i ) ) , Obtain 8 longitudinal block effect histograms, calculate described matrix M ' cThe normalization histogram of every delegation
Figure BDA00003602256900142
Obtain lateral blocks effect histogram, wherein, &Sigma; t h r k ( t ) = 1 , &Sigma; t h c k ( t ) = 1 , Described k=1,2 ..., 8.
Below the device of the extraction jpeg image proper vector of the embodiment of the present invention of the method be used to carrying out said extracted jpeg image proper vector is described, its basic logical structure is with reference to figure 9, comprise that mainly blocking effect histogram acquisition module 901 and proper vector build module 902, wherein, blocking effect histogram acquisition module 901 comprises the first acquisition module 9011 and the second acquisition module 9012, and each module is described in detail as follows:
Blocking effect histogram acquisition module 901, be used to obtaining the blocking effect histogram;
The first acquisition module 9011, obtain picture element matrix M after jpeg image I to be detected is decompressed, calculate respectively 8 original lateral blocks effect histograms and 8 original longitudinal block effect histograms of described picture element matrix M;
The second acquisition module 9012, obtain picture element matrix M after picture element matrix M procession is cut out del, to M delCarry out successively after compression and decompression obtaining the picture element matrix M ' after decompress(ion) del, calculate respectively the picture element matrix M ' after described decompress(ion) del8 calibration lateral blocks effect histograms and 8 calibration longitudinal block effect histograms;
Proper vector builds module 902, for the value of getting the front T item of described each original lateral blocks effect histogram and original longitudinal block effect histogram, forming length is the original feature vector FV of 2 * 8 * T, and the value of getting described each calibration lateral blocks effect histogram and the front T item of calibration longitudinal block effect histogram forms the alignment features vector FV that length is 2 * 8 * T del, ask for described original feature vector FV and described alignment features vector FV delDifference vector or the every absolute value of described difference vector, with described difference vector or the every absolute value of described difference vector, form the final proper vector FV of described jpeg image to be detected fin.
In the device of the extraction jpeg image proper vector of accompanying drawing 9 examples, picture element matrix M can be m * n rank matrix, and wherein, m and n are 8 multiple.The first acquisition module 9011 of accompanying drawing 9 examples can comprise interception unit 1001, computing unit 1002, delete cells 1003, cutting unit 1004, arrangement units 1005 and normalization unit 1006, the device of the extraction jpeg image proper vector that provides of another embodiment of the present invention as shown in Figure 10, wherein:
Interception unit 1001, for the 1st row to n-2 row, the 2nd row to the n-1 row and the 3rd row to n row that intercept respectively described picture element matrix M, obtain matrix A, B and C, intercept respectively the 1st of described picture element matrix M and walk to that n-2 is capable, the 2nd to walk to n-1 capable and the 3rd walk to capable matrix E, F and the G of obtaining of n;
Computing unit 1002, for compute matrix M r=︱ A-2 * B+C ︱, M c=︱ E-2 * F+G ︱;
Delete cells 1003, be used to deleting described matrix M rLast 6 row obtain the rank matrix of m * (n-8), delete described matrix M cLast 6 row obtain (m-8) * n rank matrixes;
Cutting unit 1004, be used for described m * (n-8) the rank matrix becomes submatrix every 8 column splits, can obtain altogether (n-8)/8 a column split submatrix, described (m-8) * n rank are divided into to submatrix every 8 row, can obtain altogether (m-8)/8 a horizontal submatrix of cutting apart;
Arrangement units 1005, obtain (m(n-8)/8 for described column split submatrix vertically is arranged in order) * 8 rank matrix M ' r, described row is cut apart to submatrix and laterally is arranged in order acquisition 8 * ((m-8) n/8) rank matrix M ' c
Normalization unit 1006, be used to calculating described matrix M ' rThe normalization histogram of each row Obtain 8 longitudinal block effect histograms, calculate described matrix M ' cThe normalization histogram of every delegation
Figure BDA00003602256900162
Obtain 8 lateral blocks effect histograms, wherein, &Sigma; t h r k ( t ) = 1 , &Sigma; t h c k ( t ) = 1 , Described k=1,2 ..., 8.
One of ordinary skill in the art will appreciate that all or part of step in the whole bag of tricks of above-described embodiment is to come the hardware that instruction is relevant to complete by program, this program can be stored in a computer-readable recording medium, storage medium can comprise: ROM (read-only memory) (ROM, Read Only Memory), random access memory (RAM, Random Access Memory), disk or CD etc.
Detection method and device that the above dual JPEG of grid non-alignment that the embodiment of the present invention is provided compresses are described in detail, applied specific case herein principle of the present invention and embodiment are set forth, the explanation of above embodiment is just be used to helping to understand method of the present invention and core concept thereof; Simultaneously, for one of ordinary skill in the art, according to thought of the present invention, all will change in specific embodiments and applications, in sum, this description should not be construed as limitation of the present invention.

Claims (8)

1. the detection method of the dual JPEG compression of grid non-alignment, is characterized in that, described method comprises:
Obtain the quantization table QT of jpeg image I to be detected;
Obtain the blocking effect histogram, comprise: obtain picture element matrix M after jpeg image I to be detected is decompressed, calculate respectively 8 original lateral blocks effect histograms and 8 original longitudinal block effect histograms of described picture element matrix M, and obtain picture element matrix M after picture element matrix M procession is cut out del, to M delCarry out successively after compression and decompression obtaining the picture element matrix M ' after decompress(ion) del, calculate respectively the picture element matrix M ' after described decompress(ion) del8 calibration lateral blocks effect histograms and 8 calibration longitudinal block effect histograms;
The value formation length of getting the front T item of described each original lateral blocks effect histogram and original longitudinal block effect histogram is the original feature vector FV of 2 * 8 * T, and the value formation length of getting described each calibration lateral blocks effect histogram and the front T item of calibration longitudinal block effect histogram is the alignment features vector FV of 2 * 8 * T del, ask for described original feature vector FV and described alignment features vector FV delDifference vector or the every absolute value of described difference vector, with described difference vector or the every absolute value of described difference vector, form the final proper vector FV of described jpeg image to be detected fin
According to described quantization table QT, select the support vector machine classifier corresponding with described quantization table QT;
By described final proper vector FV finThe support vector machine classifier of inputting described selection be take and adjudicated described jpeg image to be detected and whether be the dual jpeg compressed image of grid non-alignment.
2. method according to claim 1, is characterized in that, described picture element matrix M is m * n rank matrix, and described m and n are 8 multiple;
8 original lateral blocks effect histograms and 8 original longitudinal block effect histograms of the described matrix M of calculating pixel respectively comprise:
The 1st row to n-2 row, the 2nd row to the n-1 row and the 3rd row to n row that intercept respectively described picture element matrix M obtain matrix A, B and C, intercept respectively the 1st of described picture element matrix M and walk to that n-2 is capable, the 2nd to walk to n-1 capable and the 3rd walk to capable matrix E, F and the G of obtaining of n;
Compute matrix M r=︱ A-2 * B+C ︱, M c=︱ E-2 * F+G ︱;
Delete described matrix M rLast 6 row obtain the rank matrix of m * (n-8), delete described matrix M cLast 6 row obtain (m-8) * n rank matrixes;
By described m * (n-8) the rank matrix becomes submatrix every 8 column splits, can obtain altogether (n-8)/8 a column split submatrix, described (m-8) * n rank matrix is divided into to submatrix every 8 row, can obtain altogether (m-8)/8 a horizontal submatrix of cutting apart;
Described column split submatrix vertically is arranged in order and obtains (m(n-8)/8) * 8 rank matrix M ' r, described row is cut apart to submatrix and laterally is arranged in order acquisition 8 * ((m-8) n/8) rank matrix M ' c
Calculate described matrix M ' rThe normalization histogram of each row
Figure FDA00003602256800021
Obtain 8 longitudinal block effect histograms, calculate described matrix M ' cThe normalization histogram of every delegation h c k ( t ) = &Sigma; i = 1 ( m - 8 ) n / 8 &delta; ( t - M c &prime; ( k , i ) ) , Obtain 8 lateral blocks effect histograms, wherein, &Sigma; t h r k ( t ) = 1 ,
Figure FDA00003602256800024
Described k=1,2 ..., 8.
3. a method of extracting the jpeg image proper vector, is characterized in that, described method comprises:
Obtain the blocking effect histogram, comprise: obtain picture element matrix M after jpeg image I to be detected is decompressed, calculate respectively 8 original lateral blocks effect histograms and 8 original longitudinal block effect histograms of described picture element matrix M, and obtain picture element matrix M after picture element matrix M procession is cut out del, to M delCarry out successively after compression and decompression obtaining the picture element matrix M ' after decompress(ion) del, calculate respectively the picture element matrix M ' after described decompress(ion) del8 calibration lateral blocks effect histograms and 8 calibration longitudinal block effect histograms;
The value formation length of getting the front T item of described each original lateral blocks effect histogram and original longitudinal block effect histogram is the original feature vector FV of 2 * 8 * T, and the value formation length of getting described each calibration lateral blocks effect histogram and the front T item of calibration longitudinal block effect histogram is the alignment features vector FV of 2 * 8 * T del, ask for described original feature vector FV and described alignment features vector FV delDifference vector or the every absolute value of described difference vector, with described difference vector or the every absolute value of described difference vector, form the final proper vector FV of described jpeg image to be detected fin.
4. method according to claim 3, is characterized in that, described picture element matrix M is m * n matrix, and described m and n are 8 multiple;
8 original lateral blocks effect histograms and 8 original longitudinal block effect histograms of the described matrix M of calculating pixel respectively comprise:
The 1st row to n-2 row, the 2nd row to the n-1 row and the 3rd row to n row that intercept respectively described picture element matrix M obtain matrix A, B and C, intercept respectively the 1st of described picture element matrix M and walk to that n-2 is capable, the 2nd to walk to n-1 capable and the 3rd walk to capable matrix E, F and the G of obtaining of n;
Compute matrix M r=︱ A-2 * B+C ︱, M c=︱ E-2 * F+G ︱;
Delete described matrix M rLast 6 row obtain the rank matrix of m * (n-8), delete described matrix M cLast 6 row obtain (m-8) * n rank matrixes;
By described m * (n-8) the rank matrix becomes submatrix every 8 column splits, can obtain altogether (n-8)/8 a column split submatrix, described (m-8) * n rank matrix is divided into to submatrix every 8 row, can obtain altogether (m-8)/8 a horizontal submatrix of cutting apart;
Described column split submatrix vertically is arranged in order and obtains (m(n-8)/8) * 8 rank matrix M ' r, described row is cut apart to submatrix and laterally is arranged in order acquisition 8 * ((m-8) n/8) rank matrix M ' c
Calculate described matrix M ' rThe normalization histogram of each row
Figure FDA00003602256800031
Obtain 8 longitudinal block effect histograms, calculate described matrix M ' cThe normalization histogram of every delegation h c k ( t ) = &Sigma; i = 1 ( m - 8 ) n / 8 &delta; ( t - M c &prime; ( k , i ) ) , Obtain 8 lateral blocks effect histograms, wherein, &Sigma; t h r k ( t ) = 1 ,
Figure FDA00003602256800034
Described k=1,2 ..., 8.
5. the pick-up unit of the dual JPEG compression of grid non-alignment, is characterized in that, described device comprises:
The quantization table acquisition module, be used to obtaining the quantization table QT of jpeg image I to be detected;
Blocking effect histogram acquisition module, be used to obtaining the blocking effect histogram, described blocking effect histogram acquisition module comprises following the first acquisition module and the second acquisition module, wherein:
Described the first acquisition module, obtain picture element matrix M after jpeg image I to be detected is decompressed, calculate respectively 8 original lateral blocks effect histograms and 8 original longitudinal block effect histograms of described picture element matrix M;
Described the second acquisition module, obtain picture element matrix M after picture element matrix M procession is cut out del, to M delCarry out successively after compression and decompression obtaining the picture element matrix M ' after decompress(ion) del, calculate respectively the picture element matrix M ' after described decompress(ion) del8 calibration lateral blocks effect histograms and 8 calibration longitudinal block effect histograms;
Proper vector builds module, for the value of getting the front T item of described each original lateral blocks effect histogram and original longitudinal block effect histogram, forming length is the original feature vector FV of 2 * 8 * T, and the value of getting described each calibration lateral blocks effect histogram and the front T item of calibration longitudinal block effect histogram forms the alignment features vector FV that length is 2 * 8 * T del, ask for described original feature vector FV and described alignment features vector FV delDifference vector or the every absolute value of described difference vector, with described difference vector or the every absolute value of described difference vector, form the final proper vector FV of described jpeg image to be detected fin
Select module, for according to described quantization table QT, select the support vector machine classifier corresponding with described quantization table QT;
Judging module, for by described final proper vector FV finThe support vector machine classifier of inputting described selection be take and adjudicated described jpeg image to be detected and whether be the dual jpeg compressed image of grid non-alignment.
6. device according to claim 5, is characterized in that, described picture element matrix M is m * n rank matrix, and described m and n are 8 multiple;
Described the first acquisition module comprises:
Interception unit, for the 1st row to n-2 row, the 2nd row to the n-1 row and the 3rd row to n row that intercept respectively described picture element matrix M, obtain matrix A, B and C, intercept respectively the 1st of described picture element matrix M and walk to that n-2 is capable, the 2nd to walk to n-1 capable and the 3rd walk to capable matrix E, F and the G of obtaining of n;
Computing unit, for compute matrix M r=︱ A-2 * B+C ︱, M c=︱ E-2 * F+G ︱;
Delete cells, be used to deleting described matrix M rLast 6 row obtain the rank matrix of m * (n-8), delete described matrix M cLast 6 row obtain (m-8) * n rank matrixes;
Cutting unit, be used for described m * (n-8) the rank matrix becomes submatrix every 8 column splits, can obtain altogether (n-8)/8 a column split submatrix, described (m-8) * n rank matrix is divided into to submatrix every 8 row, can obtain altogether (m-8)/8 a horizontal submatrix of cutting apart;
Arrangement units, obtain (m(n-8)/8 for described column split submatrix vertically is arranged in order) * 8 rank matrix M ' r, described row is cut apart to submatrix and laterally is arranged in order acquisition 8 * ((m-8) n/8) rank matrix M ' c
The normalization unit, be used to calculating described matrix M ' rThe normalization histogram of each row Obtain 8 longitudinal block effect histograms, calculate described matrix M ' cThe normalization histogram of every delegation
Figure FDA00003602256800042
Obtain 8 lateral blocks effect histograms, wherein,
Figure FDA00003602256800043
Described k=1,2 ..., 8.
7. a device that extracts the jpeg image proper vector, is characterized in that, described device comprises:
Blocking effect histogram acquisition module, be used to obtaining the blocking effect histogram, described blocking effect histogram acquisition module comprises following the first acquisition module and the second acquisition module, wherein:
Described the first acquisition module, obtain picture element matrix M after jpeg image I to be detected is decompressed, calculate respectively 8 original lateral blocks effect histograms and 8 original longitudinal block effect histograms of described picture element matrix M;
Described the second acquisition module, obtain picture element matrix M after picture element matrix M procession is cut out del, to M delCarry out successively after compression and decompression obtaining the picture element matrix M ' after decompress(ion) del, calculate respectively the picture element matrix M ' after described decompress(ion) del8 calibration lateral blocks effect histograms and 8 calibration longitudinal block effect histograms;
Proper vector builds module, for the value of getting the front T item of described each original lateral blocks effect histogram and original longitudinal block effect histogram, forming length is the original feature vector FV of 2 * 8 * T, and the value of getting described each calibration lateral blocks effect histogram and the front T item of calibration longitudinal block effect histogram forms the alignment features vector FV that length is 2 * 8 * T del, ask for described original feature vector FV and described alignment features vector FV delDifference vector or the every absolute value of described difference vector, with described difference vector or the every absolute value of described difference vector, form the final proper vector FV of described jpeg image to be detected fin.
8. device according to claim 7, is characterized in that, described picture element matrix M is m * n rank matrix, and described m and n are 8 multiple;
Described the first acquisition module comprises:
Interception unit, for the 1st row to n-2 row, the 2nd row to the n-1 row and the 3rd row to n row that intercept respectively described picture element matrix M, obtain matrix A, B and C, intercept respectively the 1st of described picture element matrix M and walk to that n-2 is capable, the 2nd to walk to n-1 capable and the 3rd walk to capable matrix E, F and the G of obtaining of n;
Computing unit, for compute matrix M r=︱ A-2 * B+C ︱, M c=︱ E-2 * F+G ︱;
Delete cells, be used to deleting described matrix M rLast 6 row obtain the rank matrix of m * (n-8), delete described matrix M cLast 6 row obtain (m-8) * n rank matrixes;
Cutting unit, be used for described m * (n-8) the rank matrix becomes submatrix every 8 column splits, can obtain altogether (n-8)/8 a column split submatrix, described (m-8) * n rank matrix is divided into to submatrix every 8 row, can obtain altogether (m-8)/8 a horizontal submatrix of cutting apart;
Arrangement units, obtain (m(n-8)/8 for described column split submatrix vertically is arranged in order) * 8 rank matrix M ' r, described row is cut apart to submatrix and laterally is arranged in order acquisition 8 * ((m-8) n/8) rank matrix M ' c
The normalization unit, be used to calculating described matrix M ' rThe normalization histogram of each row
Figure FDA00003602256800061
Obtain 8 longitudinal block effect histograms, calculate described matrix M ' cThe normalization histogram of every delegation
Figure FDA00003602256800062
Obtain 8 lateral blocks effect histograms, wherein, &Sigma; t h r k ( t ) = 1 , &Sigma; t h c k ( t ) = 1 , Described k=1,2 ..., 8.
CN201310330269.1A 2013-07-31 2013-07-31 Detection method and the device of the dual JPEG compression of a kind of grid non-alignment Active CN103413336B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201310330269.1A CN103413336B (en) 2013-07-31 2013-07-31 Detection method and the device of the dual JPEG compression of a kind of grid non-alignment

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201310330269.1A CN103413336B (en) 2013-07-31 2013-07-31 Detection method and the device of the dual JPEG compression of a kind of grid non-alignment

Publications (2)

Publication Number Publication Date
CN103413336A true CN103413336A (en) 2013-11-27
CN103413336B CN103413336B (en) 2016-06-08

Family

ID=49606341

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201310330269.1A Active CN103413336B (en) 2013-07-31 2013-07-31 Detection method and the device of the dual JPEG compression of a kind of grid non-alignment

Country Status (1)

Country Link
CN (1) CN103413336B (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104869425A (en) * 2015-05-13 2015-08-26 信阳师范学院 Compression and decompression method based on texture image similarity
CN107908379A (en) * 2017-10-27 2018-04-13 长安大学 A kind of method for showing super large page numeral printing original
CN108010049A (en) * 2017-11-09 2018-05-08 华南理工大学 Split the method in human hand region in stop-motion animation using full convolutional neural networks
CN111415323A (en) * 2019-01-04 2020-07-14 阿里巴巴集团控股有限公司 Image detection method and device and neural network training method and device

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090257656A1 (en) * 2008-04-14 2009-10-15 Yun-Qing Shi Detecting Double JPEG Compression in Images
CN101989355A (en) * 2009-07-31 2011-03-23 北京邮电大学 Detection method for secondary compression of JPEG image
CN102413328A (en) * 2011-11-11 2012-04-11 中国科学院深圳先进技术研究院 Double compression detection method and system of joint photographic experts group (JPEG) image

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090257656A1 (en) * 2008-04-14 2009-10-15 Yun-Qing Shi Detecting Double JPEG Compression in Images
CN101989355A (en) * 2009-07-31 2011-03-23 北京邮电大学 Detection method for secondary compression of JPEG image
CN102413328A (en) * 2011-11-11 2012-04-11 中国科学院深圳先进技术研究院 Double compression detection method and system of joint photographic experts group (JPEG) image

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
WEIQI LUO.ETC: "A Novel Method for Detecting Cropped and Recompressed Image Block", 《IEEE CONFERENCE ON ACOUSTICS, SPEECH AND SIGNAL PROCESSING》 *

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104869425A (en) * 2015-05-13 2015-08-26 信阳师范学院 Compression and decompression method based on texture image similarity
CN107908379A (en) * 2017-10-27 2018-04-13 长安大学 A kind of method for showing super large page numeral printing original
CN107908379B (en) * 2017-10-27 2020-08-14 长安大学 Method for displaying digital publishing original image of super-large page
CN108010049A (en) * 2017-11-09 2018-05-08 华南理工大学 Split the method in human hand region in stop-motion animation using full convolutional neural networks
CN111415323A (en) * 2019-01-04 2020-07-14 阿里巴巴集团控股有限公司 Image detection method and device and neural network training method and device
CN111415323B (en) * 2019-01-04 2022-05-27 阿里巴巴集团控股有限公司 Image detection method and device and neural network training method and device

Also Published As

Publication number Publication date
CN103413336B (en) 2016-06-08

Similar Documents

Publication Publication Date Title
Zhang et al. Split-brain autoencoders: Unsupervised learning by cross-channel prediction
Shin et al. Block-based noise estimation using adaptive Gaussian filtering
Li et al. Revealing the trace of high-quality JPEG compression through quantization noise analysis
Diana Andrushia et al. Saliency-based image compression using Walsh–Hadamard transform (WHT)
JP2006505075A (en) Nonlinear quantization and similarity matching method for video sequence retrieval with multiple image frames
JP2004350283A (en) Method for segmenting compressed video into 3-dimensional objects
CN112235569B (en) Quick video classification method, system and device based on H264 compressed domain
CN103413336A (en) Grid non-aligned double-JPEG-compression detecting method and device
WO2014186886A1 (en) System and method having transparent composite model for transform coefficients
CN114915786B (en) Asymmetric semantic image compression method for Internet of things scene
US9106925B2 (en) WEAV video compression system
Wang A survey on IQA
Jindal et al. Applicability of fractional transforms in image processing-review, technical challenges and future trends
Lukin et al. Lossy compression of remote sensing images with controllable distortions
CN114651270A (en) Depth loop filtering by time-deformable convolution
Zemliachenko et al. Compression ratio prediction in lossy compression of noisy images
Okarma Colour image quality assessment using the combined full-reference metric
CN115361559A (en) Image encoding method, image decoding method, image encoding device, image decoding device, and storage medium
Petrov et al. Intra frame compression and video restoration based on conditional markov processes theory
Imani et al. Stereoscopic video quality assessment using modified parallax attention module
EP3154023A1 (en) Method and apparatus for de-noising an image using video epitome
Kimishima et al. Frame adaptive rate control scheme for video compressive sensing
WO2015128302A1 (en) Method and apparatus for filtering and analyzing a noise in an image
CN108259891B (en) 3D video quality blind evaluation method based on binocular spatiotemporal intrinsic inference mechanism
Ke et al. A Video Image Compression Method based on Visually Salient Features.

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