CN110363207A - Image detection and the generation method of sorter model, device and equipment - Google Patents
Image detection and the generation method of sorter model, device and equipment Download PDFInfo
- Publication number
- CN110363207A CN110363207A CN201810309030.9A CN201810309030A CN110363207A CN 110363207 A CN110363207 A CN 110363207A CN 201810309030 A CN201810309030 A CN 201810309030A CN 110363207 A CN110363207 A CN 110363207A
- Authority
- CN
- China
- Prior art keywords
- image
- color component
- component feature
- detected
- training
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 66
- 238000001514 detection method Methods 0.000 title claims description 14
- 238000012549 training Methods 0.000 claims description 144
- 230000006835 compression Effects 0.000 claims description 140
- 238000007906 compression Methods 0.000 claims description 140
- 239000000284 extract Substances 0.000 claims description 29
- 238000012512 characterization method Methods 0.000 claims description 19
- 230000004069 differentiation Effects 0.000 claims description 15
- 238000000605 extraction Methods 0.000 claims description 10
- 238000013507 mapping Methods 0.000 claims description 7
- 238000010586 diagram Methods 0.000 description 10
- 238000005070 sampling Methods 0.000 description 8
- 238000006243 chemical reaction Methods 0.000 description 6
- 230000000694 effects Effects 0.000 description 6
- 238000004040 coloring Methods 0.000 description 3
- 238000003384 imaging method Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 241001062009 Indigofera Species 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000004590 computer program Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000003825 pressing Methods 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 230000032258 transport Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/56—Extraction of image or video features relating to colour
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
This application discloses a kind of image detecting methods, which comprises at least one color component feature is extracted from image to be detected；According to the color component feature, sorter model is used to differentiate described image to be detected for initial intact compressed format image or be the image through overcompression.Using the above method, solves the demand that primitiveness identification is carried out to image.
Description
Technical field
This application involves field of image recognition, and in particular to a kind of image detecting method, device, electronic equipment and storage are set
It is standby.The application is related to generation method, device, electronic equipment and the storage equipment of a kind of sorter model simultaneously.
Background technique
Image as it is a kind of commonly using medium, in many cases require ensure its authenticity, need to carry out image original
The identification of beginning property.The identification of image primitiveness refers to and differentiates that nondestructive compression type image is initial intact compressed format image or process
The image of compression.For example, some website platforms require the image uploaded to user to carry out the identification of image primitiveness, uploaded to identify
Whether information is credible in image；Alternatively, when image is as court evidence, it is desirable that ensure that image evidence is reliable, need to identify
The primitiveness of image, and differentiate image either with or without being tampered with.
But with the fast development of technology, easily image can be usurped by the image editing software of various kinds
Change, can not with the naked eye discover by the image distorted, resulting morals and legal issue also become more and more.If energy
The primitiveness and specific modification details for enough identifying image, have very important significance for image forensics.
If image further through software is converted to lossless format image after being compressed, the trace of some compressions can be theoretically remained
Mark.With being widely used for lossless format image, the demand for carrying out primitiveness identification to image is increasing.
Summary of the invention
The application provides a kind of image detecting method, to solve to carry out image the demand of primitiveness identification.
The application provides a kind of image detecting method, comprising:
At least one color component feature is extracted from image to be detected；
According to the color component feature, sorter model is used to differentiate that described image to be detected compresses lattice for initial intact
Formula image is the image through overcompression.
It is optionally, described that at least one color component feature is extracted from image to be detected, comprising:
Described image to be detected is transformed into YCbCr color space from RGB color；
At least one described color component feature is extracted from the YCbCr color space of described image to be detected.
Optionally, described to extract at least one described color component from the YCbCr color space of described image to be detected
Feature, comprising:
Y color component feature and Cb color component feature are extracted from the YCbCr color space of described image to be detected,
Alternatively, extracting Y color component feature and Cr color component feature from the YCbCr color space of described image to be detected.
Optionally, further includes:
The Y color component feature and the Cb color component feature are spliced, spliced described image is obtained
It is special to obtain spliced image alternatively, the Y color component feature and the Cr color component feature are spliced for feature
Sign；
It is described according to the color component feature, use sorter model differentiate described image to be detected for original image or
Person is the image through overcompression, comprising: according to spliced characteristics of image, differentiates described image to be detected using sorter model
It for original image or is the image through overcompression.
Optionally, Y color component feature and Cb face are extracted in the YCbCr color space from described image to be detected
Colouring component feature, alternatively, extracting Y color component feature and Cr color from the YCbCr color space of described image to be detected
Component characterization, comprising:
Using in the richness model of airspace Spam11 submodel and minmax24 submodel from the YCbCr of described image to be detected
Y color component feature and Cb color component feature are extracted in color space, alternatively, using Spam11 in the richness model of airspace
Model and minmax24 submodel extract Y color component feature and Cr from the YCbCr color space of described image to be detected
Color component feature.
Optionally, the sorter model includes: for judging whether image to be detected is initial intact compressed format figure
The first classifier of picture；
It is described according to the color component feature, use sorter model to differentiate described image to be detected for initial intact pressure
Contracting format-pattern is the image through overcompression, comprising:
According to the color component feature, described image to be detected is differentiated using the first described classifier；
Initial intact compressed format image is all determined as according to the first all classifier, it is determined that the mapping to be checked
As being initial intact compressed format image；
Differentiate that described image to be detected is not initial intact compressed format image according at least one the first classifier,
Then determine that described image to be detected is the image through overcompression.
Optionally, further includes: when the classification for determining described image to be detected is the image through overcompression, determine to described
The tool of compression used when image to be detected is compressed.
Optionally, the sorter model, further includes: for determining the of tool of compression type that image to be detected uses
Two kinds of classifiers；
Determine the tool of compression used when compressing to described image to be detected, comprising:
According to the color component feature, is differentiated using second of classifier and generate differentiation result；
By differentiation result combination producing one coding of each second of classifier；
Calculate the Hamming distances of the coding of the coding and preset tool of compression；
The compression that will be used when being compressed with the coding shortest tool of compression of Hamming distances as described image to be detected
Tool.
Optionally, further includes: determine the compression parameters used when the compression of described image to be detected.
Optionally, the sorter model, further includes: for determine image to be detected use compression parameters the third
Classifier；
The compression parameters used when determination image to be detected compression, comprising: according to the color component feature,
The differentiation for being carried out compression parameters to image to be detected using the third classifier is used when determining the compression of described image to be detected
Compression parameters.
The application also provides a kind of generation method of sorter model, which comprises
Determine the image for participating in training；
From at least one color component feature of the image zooming-out of the participation training；
According to the color component feature, training sorter model, the sorter model is for detecting image to be detected
It for initial intact compressed format image or is the image through overcompression.
It is optionally, described from least one color component feature of the image zooming-out of the participation training, comprising:
The image for participating in training is transformed into YCbCr color space from RGB color；
At least one described color component feature is extracted from the YCbCr color space of the image for participating in training.
Optionally, the YCbCr color space from the image for participating in training extracts at least one described color
Component characterization, comprising:
Y color component feature and Cb color component are extracted from the YCbCr color space of the image for participating in training
Feature, alternatively, extracting Y color component feature and Cr color point from the YCbCr color space of the image for participating in training
Measure feature.
Optionally, further includes:
The Y color component feature and the Cb component characterization are spliced, spliced described image feature is obtained,
Alternatively, the Y color component feature and the Cr component characterization are spliced, spliced characteristics of image is obtained；
It is described according to the color component feature, training sorter model, comprising: according to spliced characteristics of image, instruction
Practice sorter model.
Optionally, it is described from it is described participate in training image YCbCr color space in extract Y color component feature and
Cb color component feature, alternatively, extracting Y color component feature from the YCbCr color space of the image for participating in training
With Cr color component feature, comprising:
Using in the richness model of airspace Spam11 submodel and minmax24 submodel from the image for participating in training
Y color component feature and Cb color component feature are extracted in YCbCr color space, alternatively, using in the richness model of airspace
Spam11 submodel and minmax24 submodel extract Y color from the YCbCr color space of the image for participating in training
Component characterization and Cr color component feature.
Optionally, the determining image for participating in training, comprising:
Determine that sample set, the sample set include First Kind Graph picture and the second class image, the First Kind Graph picture is original
Nondestructive compression type image, the second class image are the image through overcompression；
It is described from described image zooming-out at least one color component feature for participating in training, comprising: from the First Kind Graph
Picture and the second class image zooming-out at least one color component feature.
Optionally, the sorter model includes: for judging whether image to be detected is initial intact compressed format figure
The first classifier of picture；
It is described according to the color component feature, training sorter model, comprising:
According to the color component feature, the first described classifier of training.
Optionally, the sorter model further include:
For determining second of classifier of the tool of compression type of image to be detected use；
It is described according to the color component feature, training sorter model, further includes:
According to the color component feature, training second of classifier.
Optionally, the sorter model further include:
For determining the third classifier of the compression parameters of image to be detected use；
It is described according to the color component feature, training sorter model, further includes:
According to the color component feature, the third described classifier of training.
In addition the application provides a kind of image detection device, described device includes:
Feature extraction unit, for extracting at least one color component feature from image to be detected；
Image discriminating unit, for differentiating the mapping to be checked using sorter model according to the color component feature
As being initial intact compressed format image or being the image through overcompression.
In addition the application provides a kind of generating means of sorter model, described device includes:
Image determination unit, for determining the image for participating in training；
Feature extraction unit, for from described image zooming-out at least one color component feature for participating in training；
Model training unit, for according to the color component feature, training sorter model, the sorter model use
It is initial intact compressed format image or is the image through overcompression in detection image to be detected.
The application also provides a kind of electronic equipment, comprising:
Processor；And
Memory, for the program of image detecting method, which, which is powered and passes through the processor, runs image inspection
After the program of survey method, following step is executed:
At least one color component feature is extracted from image to be detected；
According to the color component feature, sorter model is used to differentiate that described image to be detected compresses lattice for initial intact
Formula image is the image through overcompression.
In addition the application provides a kind of electronic equipment, comprising:
Processor；And
Memory, the program of the generation method for sorter model, the equipment are powered and pass through the processor and runs
After the program of the generation method of the sorter model, following step is executed:
Determine the image for participating in training；
From at least one color component feature of the image zooming-out of the participation training；
According to the color component feature, training sorter model, the sorter model is for detecting image to be detected
It for initial intact compressed format image or is the image through overcompression.
In addition the application provides a kind of storage equipment,
It is stored with the program for storing image detecting method, which is run by processor, execute following step:
At least one color component feature is extracted from image to be detected；
According to the color component feature, sorter model is used to differentiate that described image to be detected compresses lattice for initial intact
Formula image is the image through overcompression.
In addition the application provides a kind of storage equipment,
It is stored with the program of the generation method for storing sorter model, which is run by processor, executes following
Step:
Determine the image for participating in training；
From at least one color component feature of the image zooming-out of the participation training；
According to the color component feature, training sorter model, the sorter model is for detecting image to be detected
It for initial intact compressed format image or is the image through overcompression.
Compared with prior art, the application has the following advantages:
Image detecting method and device provided by the present application, according to the color component feature extracted from image to be detected,
Sorter model is used to differentiate described image to be detected for initial intact compressed format image or be the image through overcompression, it is full
Foot carries out the demand of primitiveness identification to image.
The generation method and device of sorter model provided by the present application, according to the color from the image zooming-out for participating in training
Component characterization, training sorter model provide sorter model to carry out primitiveness identification to image.
Detailed description of the invention
Fig. 1 is a kind of flow chart for image detecting method that the application first embodiment provides.
Fig. 2 is a kind of color space conversion schematic diagram that the application first embodiment provides.
Fig. 3 is the schematic diagram of the first classifier that the application first embodiment provides and second of classifier.
Fig. 4 is the schematic diagram for the third classifier that the application first embodiment provides.
Fig. 5 is a kind of flow chart of the generation method for sorter model that the application second embodiment provides.
Fig. 6 is a kind of schematic diagram for image detection device that the application 3rd embodiment provides.
Fig. 7 is a kind of schematic diagram of the generating means for sorter model that the application fourth embodiment provides.
Fig. 8 is the schematic diagram for a kind of electronic equipment that the 5th embodiment of the application provides.
Fig. 9 is the schematic diagram for another electronic equipment that the application sixth embodiment provides.
Specific embodiment
Many details are explained in the following description in order to fully understand the application.But the application can be with
Much it is different from other way described herein to implement, those skilled in the art can be without prejudice to the application intension the case where
Under do similar popularization, therefore the application is not limited by following public specific implementation.
The application first embodiment provides a kind of image detecting method.Fig. 1, Fig. 2, Fig. 3, Fig. 4 are please referred to, Fig. 1 is shown
A kind of flow chart for image detecting method that the application first embodiment provides, Fig. 2 shows the application first embodiments to mention
The color space conversion schematic diagram of confession, Fig. 3 show the first classifier that the application first embodiment provides and second point
The schematic diagram of class device, Fig. 4 show the application first embodiment offer the third classifier schematic diagram, below in conjunction with Fig. 1,
Fig. 2, Fig. 3, Fig. 4 are described in detail.
As shown in Figure 1, in step s101, at least one color component feature is extracted from image to be detected.
Described image to be detected can refer to that the nondestructive compression type image of pending image primitiveness identification, image are original
Property identification, refer to differentiate nondestructive compression type image be initial intact compressed format image or the image through overcompression.
Nondestructive compression type refers to the format that will not compress image information that storage image uses when being shot with camera, i.e. RAW
(RAW Image Format) format will not carry out lossy compression to image with this format storage image, because without losing
Image information, but committed memory compare lossy compression format will be more.For example, image suffix is that (Canon's camera generates CR2
RAW file, the entitled CR2 of file suffixes) or ARW (ARW is a kind of exclusive file format of Sony, is RAW file format
Another form) image be nondestructive compression type image.
The color component feature, can refer to YCbCr (YCbCr indicate be DTV (video) color space and
Digital interface, this is international standard) color component feature in color space, i.e. Y color component feature, Cb color point
Measure feature, Cr color component feature, wherein Y indicates luminance component, and Cb indicates chroma blue component, and Cr indicates red color
Component.
At least one color component feature is extracted from image to be detected, can take following methods:
Described image to be detected is transformed into YCbCr color space from RGB color；
At least one described color component feature is extracted from the YCbCr color space of described image to be detected.
Due to that first three in image RGB color can will be led to during compression of images (for example, JPEG compression)
Road color space (R (red), G (green), B (indigo plant)) is transformed into YCbCr color space, later to Y color component feature, Cb color point
Measure feature and Cr color component feature are using different down-sampling factor progress down-samplings, and different tools of compression are when compressing image
The compression quality difference according to selection image is waited, the down-sampling factor taken the compression of each color component feature is different, and one
As be that Y color component feature decimation factor is different, Cb color component feature and Cr color component feature using identical sampling because
Son, existing maximum difference is decimation factor difference between the compress mode of certain tools of compression, therefore first by image to be detected
YCbCr color space (as shown in Figure 2) is transformed by RGB color, then extracts the feature of YCbCr color space again, then
Using the classification of feature decision image to be detected of YCbCr color space, effect, which has, to be obviously improved.
It should be noted that only choosing it since Cb color component feature is identical with Cr color component feature decimation factor
In a color component feature, therefore Y color can be extracted from the YCbCr color space of described image to be detected
Component characterization and Cb color component feature, alternatively, extracting Y color point from the YCbCr color space of described image to be detected
Measure feature and Cr color component feature.
Further, Y color component feature and Cb color component feature can also be spliced, is obtained spliced described
Characteristics of image obtains spliced figure alternatively, the Y color component feature and the Cr color component feature are spliced
As feature, the input for being used as sorter model differentiates that described image to be detected is initial intact compressed format image or is
Image through overcompression.
When extracting feature, there are many characteristics of image that can choose, and the statistical nature effect based on gray level co-occurrence matrixes is preferable,
Specifically, can be using the Spam11 submodel and minmax24 in airspace richness model (Spatial Rich Model, SRM)
Model extracts Y color component feature and Cb color component feature from the YCbCr color space of described image to be detected, or
Person, using in the richness model of airspace Spam11 submodel and minmax24 submodel from the YCbCr color of described image to be detected
Y color component feature and Cr color component feature are extracted in space.
Feature used in SRM has up to 34671 dimensions, it has used 45 filters, wherein incorporating part filter
Finally it is left 39 filters, this 39 filters can be quantified when constructing residual plot with step-length q again, wherein there are 11 filters
Wave device uses step-length q value to generate submodel for { 1,2 } two parameters, and 28 filters use step-length q value for { 1,1.5,2 }
Three parameters generate submodel, and one shares 106 submodels, and each submodel has 325 or 338 dimensional features.The application test
The classifying quality of all submodels chooses the best submodel of two of them effect, the s35_spam11_q2 of specifically 338 dimensions
With two subtemplates of s1_minmax24_q2 of 325 dimensions, 663 dimensional feature altogether, if having extracted two color component spies of Y and Cb
Sign or two color component features of Y and Cr, latter image of merging features have 663*2=1326 dimensional feature.
As shown in Figure 1, in step s 102, according to the color component feature, using sorter model differentiate it is described to
Detection image is initial intact compressed format image or is the image through overcompression.
The initial intact compressed format image refers to by after the shooting of the equipment such as camera, using not pressing image information
The image of the format storage of contracting.
Image through overcompression refers to the nondestructive compression type image crossed by tool of compression lossy compression.Tool of compression can
To refer to compressed software, for example, Photoshop, ACDsee etc..
The sorter model includes: for judging whether image to be detected is the first of initial intact compressed format image
Kind classifier；
According to the color component feature, sorter model is used to differentiate that described image to be detected compresses lattice for initial intact
Formula image is the image through overcompression, comprising:
According to the color component feature, described image to be detected is differentiated using the first described classifier；
Initial intact compressed format image is all determined as according to the first all classifier, it is determined that the mapping to be checked
As being initial intact compressed format image；
Differentiate that described image to be detected is not initial intact compressed format image according at least one the first classifier,
Then determine that described image to be detected is the image through overcompression.
For example, as shown in figure 3, by taking Photoshop, ACDsee, nEO iMAGING these three softwares as an example, it is assumed that every kind soft
Part has { 1,2 } two class, and initial intact compressed format image is replaced with O, and Photoshop, ACDsee and shadow magic use P, A respectively
With G replace, F (O-P1), F (O-P2), F (O-A1), F (O-A2), F (O-G1), two classifier of F (O-G2), for for judge to
Detection image whether be initial intact compressed format image the first classifier.An image to be detected is inputted when test,
Characteristics of image F (the image spy after can be two color component merging features of Y and Cb is extracted after first carrying out color space conversion
Sign), characteristics of image F is input to above-mentioned two classifier, if two all classifiers are all determined as initial intact compressed format
Image, it is determined that described image to be detected is initial intact compressed format image, if any one classifier judgement among the above
Image to be detected is not initial intact compressed format image, it is determined that described image to be detected is the image through overcompression.
In the application first embodiment, two classifiers can use Fisher linear discriminant (Fisher Linear
Discriminant, FLD) integrated classifier, FLD integrated classifier is two classifiers, can combine multiple two classifiers
Complete more classification tasks.
Further, when the classification for determining described image to be detected is image through overcompression, can determine to it is described to
The tool of compression used when detection image is compressed.
At this point, sorter model, can also include: second for determining the tool of compression type of image to be detected use
Kind classifier.
Still continue to use above-mentioned example: as shown in figure 3, F (P1-A1), F (P1-A2), F (P1-G1), F (P1-G2), F (P2-
A1), F (P2-A2), F (P2-G1), F (P2-G2), F (A1-G1), F (A1-G2), F (A2-G1), two classifier of F (A2-G2) are
Which kind of tool of compression second of classifier has used when can differentiate that image to be detected is compressed by these two classifiers.
In the specific implementation, the tool of compression used when compressing to described image to be detected is determined, under may include
State step:
According to the color component feature, is differentiated using second of classifier and generate differentiation result；
By differentiation result combination producing one coding of each second of classifier；
Calculate the Hamming distances of the coding of the coding and preset tool of compression；
The compression that will be used when being compressed with the coding shortest compressed software of Hamming distances as described image to be detected
Tool.
Still continue to use above-mentioned example: if it is determined that needing to pick out to be detected when image to be detected is the image through overcompression
Characteristics of image F is input to all above-mentioned second of classifiers, each classification by the tool of compression used when image is compressed
Device can all export one and differentiate as a result, differentiating that result is "+1 " or " -1 ".These are differentiated into one string encoding of result combination producing, then
Calculate this image to be tested coding and preset various softwares coding Hamming distances (can also with Euclidean away from
From), shortest with the Hamming distances of above-mentioned coding in all tools of compression is the compression work used when image to be detected compression
Tool.
As shown in table 1, it illustrates the corresponding codings of compressed softwares various in preset Fig. 3.
Table 1
| P1-A1 | P1-A2 | P2-A1 | P2-A2 | P1-G1 | P1-G2 | P2-G1 | P2-G2 | A1-G1 | A1-G2 | A2-G1 | A2-G2 | |
| P | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 0 | 0 | 0 | 0 |
| A | -1 | -1 | -1 | -1 | 0 | 0 | 0 | 0 | 1 | 1 | 1 | 1 |
| G | 0 | 0 | 0 | 0 | -1 | -1 | -1 | -1 | -1 | -1 | -1 | -1 |
Reference table 1, such as software P (Photoshop) is in F (P1-A1), F (P1-A2), F (P1-G1), F (P1-G2), F
(P2-A1), F (P2-A2), F (P2-G1), the class that is positive is divided in two classifier of F (P2-G2), then volume of the P in these two classifiers
Code is that 1, P is not engaged in trained F (A1-G1), F (A1-G2), F (A2-G1), two classifier of F (A2-G2), software P corresponding two
Classifier is encoded to 0；G (nEO iMAGING) is in F (P1-G1), F (P1-G2), F (P2-G1), F (P2-G2), F (A1-G1), F
(A1-G2), F (A2-G1), the class that is negative is divided in F (A2-G2) classifier, then G does not have in -1, the G that is encoded to of these two classifiers P
There are participation training F (P1-A1), F (P1-A2), F (P2-A1), two classifier of F (P2-A2), the corresponding two classifiers coding of software P
It is 0.
Hamming distances calculation method, reference table 2 is exemplified below, if image to be detected (being replaced in table 2 with T) uses
After second of classifier is distinguished, the coding of generation is 011111111000, each compares with the coding of P, A, G, such as
The identical then corresponding position in the corresponding position of fruit is 0, and difference is just 1, and it is direct not calculate namely distance if Software Coding is 0
Be 0, to image to be detected differentiate after generate be encoded to 011111111000, P be encoded to 111111110000 and P away from
From being 100000000000, adding up the distance that distance is just 1 and A is 011100000000, and distance is that the distance of 3 and G is
000011111000, distance is 5, due to most short with the distance of P, it is thus determined that image to be detected is compressed by Photoshop
Image.
Table 2
Further, it after determining the tool of compression used when compressing to described image to be detected, can also determine
The compression parameters that described image to be detected uses when compressing.
At this point, sorter model, can also include: the third point for determining the compression parameters of image to be detected use
Class device.
Determine the compression parameters used when the compression of described image to be detected, comprising: according to the color component feature, use
The third classifier carries out the differentiation of compression parameters to image to be detected, determines the compression used when the compression of described image to be detected
Parameter.
Include parameter 0-12, (f0-f1)-(f0-f12) as shown in figure 4, with Photoshop), (f1-f2)-(f1-
F12)) ... above-mentioned two classifier is the third classifier.
Above-mentioned example is still continued to use, determines that image to be detected is the image compressed by Photoshop, calls Photoshop
Characteristics of image F, is input to the third all classifiers, root by compression parameters classifier (the third classifier in i.e. such as 4)
Final result is elected to vote on according to all two classification results.If parameter output is 10, when determining image to be detected compression
The compression parameters used are 10.
So far, the embodiment of the image detecting method provided the application first embodiment is described in detail.This
Apply for that first embodiment according to the color component feature extracted from image to be detected, is differentiated described to be checked using sorter model
Altimetric image is initial intact compressed format image or is the image through overcompression, not only meets and carries out primitiveness mirror to image
Fixed demand, and when sorter model includes second of classifier, it can also determine and described image to be detected is pressed
The tool of compression used when contracting can further determine that the mapping to be checked when sorter model includes the third classifier
As the compression parameters used when compression, to meet a variety of demands identified image.
The application second embodiment provides a kind of generation method of sorter model.Referring to FIG. 5, Fig. 5 shows this
Apply for a kind of flow chart of the generation method for sorter model that second embodiment provides.
As shown in figure 5, determining the image for participating in training in step S501.
The image for participating in training, refers to the image for participating in sorter model training.
The image for participating in training can be a sample set, and the sample set includes First Kind Graph picture and the second class image,
The First Kind Graph picture is initial intact compressed format image, and the second class image is the image through overcompression.
The initial intact compressed format image refers to by after the shooting of the equipment such as camera, using not pressing image information
The image of the format storage of contracting.
Image through overcompression refers to the nondestructive compression type image crossed by tool of compression lossy compression.Tool of compression can
To refer to compressed software, for example, Photoshop, ACDsee etc..
As shown in figure 5, in step S502, from least one color component feature of the image zooming-out of the participation training.
The color component feature, can refer to YCbCr (YCbCr indicate be DTV (video) color space and
Digital interface, this is international standard) color component feature in color space, i.e. Y color component feature, Cb color point
Measure feature, Cr color component feature, wherein Y indicates luminance component, and Cb indicates chroma blue component, and Cr indicates red color
Component.
At least one color component feature is extracted from the image for participating in training, following methods can be taken:
The image for participating in training is transformed into YCbCr color space from RGB color；
At least one described color component feature is extracted from the YCbCr color space of the image for participating in training.
Due to that first three in image RGB color can will be led to during compression of images (for example, JPEG compression)
Road color space (R (red), G (green), B (indigo plant)) is transformed into YCbCr color space, later to Y color component feature, Cb color point
Measure feature and Cr color component feature are using different down-sampling factor progress down-samplings, and different tools of compression are when compressing image
The compression quality difference according to selection image is waited, the down-sampling factor taken the compression of each color component feature is different, and one
As be that Y color component feature decimation factor is different, Cb color component feature and Cr color component feature using identical sampling because
Son, existing maximum difference is decimation factor difference between the compress mode of certain tools of compression, therefore will first participate in training
Image is transformed into YCbCr color space (as shown in Figure 2) by RGB color, then extracts the spy of YCbCr color space again
Sign recycles the feature training sorter model of YCbCr color space, and effect, which has, to be obviously improved.
It should be noted that only choosing it since Cb color component feature is identical with Cr color component feature decimation factor
In a color component feature, therefore can from it is described participate in training image YCbCr color space in extract Y
Color component feature and Cb color component feature, alternatively, being extracted from the YCbCr color space of the image for participating in training
Y color component feature and Cr color component feature.
Further, Y color component feature and Cb color component feature can also be spliced, is obtained spliced described
Characteristics of image obtains spliced figure alternatively, the Y color component feature and the Cr color component feature are spliced
As feature, then according to spliced characteristics of image, training sorter model.
When extracting feature, there are many characteristics of image that can choose, and the statistical nature effect based on gray level co-occurrence matrixes is preferable,
Specifically, can be using the Spam11 submodel and minmax24 in airspace richness model (Spatial Rich Model, SRM)
Model extracts Y color component feature and Cb color component feature from the YCbCr color space of the image for participating in training,
Alternatively, using in the richness model of airspace Spam11 submodel and minmax24 submodel from the image for participating in training
Y color component feature and Cr color component feature are extracted in YCbCr color space.
Feature used in SRM has up to 34671 dimensions, it has used 45 filters, wherein incorporating part filter
Finally it is left 39 filters, this 39 filters can be quantified when constructing residual plot with step-length q again, wherein there are 11 filters
Wave device uses step-length q value to generate submodel for { 1,2 } two parameters, and 28 filters use step-length q value for { 1,1.5,2 }
Three parameters generate submodel, and one shares 106 submodels, and each submodel has 325 or 338 dimensional features.The application test
The classifying quality of all submodels chooses the best submodel of two of them effect, the s35_spam11_q2 of specifically 338 dimensions
With two subtemplates of s1_minmax24_q2 of 325 dimensions, 663 dimensional feature altogether, if having extracted two color component spies of Y and Cb
Sign or two color component features of Y and Cr, latter image of merging features have 663*2=1326 dimensional feature.
As shown in figure 5, in step S503, according to the color component feature, training sorter model, the classifier
Model is initial intact compressed format image or is the image through overcompression for detecting image to be detected.
The sorter model may include: for judging whether image to be detected is initial intact compressed format image
The first classifier；It is described according to the color component feature, training sorter model, comprising: special according to the color component
Sign, the first described classifier of training.
For example, as shown in figure 3, by taking Photoshop, ACDsee, nEO iMAGING these three softwares as an example, it is assumed that every kind soft
Part has { 1,2 } two class, and initial intact compressed format image is replaced with O, and Photoshop, ACDsee and shadow magic use P, A respectively
It is replaced with G, needs to train F (O-P1), F (O-P2), F (O-A1), F (O-A2), F (O-G1), two classifier of F (O-G2), it is above-mentioned
Two classifiers be for judge image to be detected whether be initial intact compressed format image the first classifier.It can be by
A kind of image extracts characteristics of image F (after can be two color component merging features of Y and Cb after first carrying out color space conversion
Characteristics of image), characteristics of image F is then input to the first classifier, the first classifier of training.
Further, the sorter model can also include: the tool of compression type used for determining image to be detected
Second of classifier；It is described according to the color component feature, training sorter model, further includes: according to the color point
Measure feature, training second of classifier.
As shown in figure 3, F (P1-A1), F (P1-A2), F (P1-G1), F (P1-G2), F (P2-A1), F (P2-A2), F (P2-
G1), F (P2-G2), F (A1-G1), F (A1-G2), F (A2-G1), two classifier of F (A2-G2) are second of classifier, Ke Yitong
It crosses these two classifiers and differentiates which kind of tool of compression used when image to be detected is compressed.
In second of classifier of training, the second class image can be divided into multiple second subclass figures according to tool of compression
Picture can will pass through for example, Photoshop has 13 different compression parameters by taking two softwares of Photoshop, ACDsee as an example
The image of Photoshop compression is divided into 13 the second subclass images according to compression parameters.Since ACDsee has 100 different pressures
Contracting parameter can choose 13 compression parameters of ACDsee according to compressibility factor phase approximately principle, choose 13 according to compression parameters
Second subclass image.
It should be noted that training second of classifier when, can according to compressed software type two-by-two subclass intersect into
Row training classifier.For example, the image by Photoshop compression can be divided into the second son of P0-P12 according to compression parameters
Class image will be divided into A0-A12 the second subclass image according to compression parameters by the image of ACDsee compression, then using belonging to
The second subclass image P0-P12 of Photoshop intersects with the second subclass image A0-A12 for belonging to ACDsee is trained second
Kind classifier.For example, (can be with using the color component feature extracted from P1 the second subclass image and A1 the second subclass image
It is the characteristics of image after two color component merging features of Y and Cb) second of classifier (P1-A1) of training.
Further, sorter model can also include: for determine image to be detected use compression parameters the third
Classifier；It is described according to the color component feature, training sorter model, further includes: according to the color component feature, instruction
Practice the third described classifier.
Include parameter 0-12, (f0-f1)-(f0-f12) as shown in figure 4, with Photoshop), (f1-f2)-(f1-
F12)) ... above-mentioned two classifier is the third classifier.In training, intersects two-by-two according to compression parameters and be trained classification
Device, for example, (being also possible to Y using the color component feature extracted from P1 the second subclass image and P2 the second subclass image
With the characteristics of image after two color component merging features of Cb) train the third classifier (f1-f2).
So far, the embodiment of the generation method of the sorter model provided the application second embodiment has carried out in detail
Explanation.The application second embodiment is according to from image zooming-out at least one the color component feature for participating in training, and according to described
Color component feature, training sorter model, provides not only in judging whether image to be detected is initial intact compressed format
The first classifier of image, and provide second of classification for determining the tool of compression type of image to be detected use
Device, also provided is the third classifiers for determining the compression parameters of image to be detected use, to meet to figure
As a variety of demands identified.
Corresponding with a kind of image detecting method of above-mentioned offer, the application 3rd embodiment additionally provides a kind of image
Detection device.
As shown in fig. 6, image detection device includes: feature extraction unit 601, image discriminating unit 602.
Feature extraction unit 601, for extracting at least one color component feature from image to be detected；
Image discriminating unit 602, for being differentiated using sorter model described to be detected according to the color component feature
Image is initial intact compressed format image or is the image through overcompression.
Optionally, the feature extraction unit, comprising:
Color space conversion subelement, it is empty for described image to be detected to be transformed into YCbCr color from RGB color
Between；
Color component extracts subelement, for from the YCbCr color space of described image to be detected extract it is described at least
One color component feature.
Optionally, the color component extracts subelement, is specifically used for:
Y color component feature and Cb color component feature are extracted from the YCbCr color space of described image to be detected,
Alternatively, extracting Y color component feature and Cr color component feature from the YCbCr color space of described image to be detected.
Optionally, described device further include:
Merging features unit is obtained for splicing the Y color component feature and the Cb color component feature
Spliced described image feature obtains alternatively, the Y color component feature and the Cr color component feature are spliced
To spliced characteristics of image；
Described image judgement unit, is specifically used for: according to spliced characteristics of image, differentiated using sorter model described in
Image to be detected is original image or is the image through overcompression.
Optionally, Y color component feature and Cb face are extracted in the YCbCr color space from described image to be detected
Colouring component feature, alternatively, extracting Y color component feature and Cr color from the YCbCr color space of described image to be detected
Component characterization, comprising:
Using in the richness model of airspace Spam11 submodel and minmax24 submodel from the YCbCr of described image to be detected
Y color component feature and Cb color component feature are extracted in color space, alternatively, using Spam11 in the richness model of airspace
Model and minmax24 submodel extract Y color component feature and Cr from the YCbCr color space of described image to be detected
Color component feature.
Optionally, the sorter model includes: for judging whether image to be detected is initial intact compressed format figure
The first classifier of picture；
Described image judgement unit, is specifically used for:
According to the color component feature, described image to be detected is differentiated using the first described classifier；
Initial intact compressed format image is all determined as according to the first all classifier, it is determined that the mapping to be checked
As being initial intact compressed format image；
Differentiate that described image to be detected is not initial intact compressed format image according at least one the first classifier,
Then determine that described image to be detected is the image through overcompression.
Optionally, described device further includes tool of compression determination unit, for when the classification for determining described image to be detected
When for image through overcompression, the tool of compression used when compressing to described image to be detected is determined.
Optionally, the sorter model, further includes: for determining the of tool of compression type that image to be detected uses
Two kinds of classifiers；
The tool of compression determination unit, is specifically used for:
According to the color component feature, is differentiated using second of classifier and generate differentiation result；
By differentiation result combination producing one coding of each second of classifier；
Calculate the Hamming distances of the coding of the coding and preset tool of compression；
The compression that will be used when being compressed with the coding shortest tool of compression of Hamming distances as described image to be detected
Tool.
Optionally, described device further include: compression parameters determination unit makes when for determining the compression of described image to be detected
Compression parameters.
Optionally, the sorter model, further includes: for determine image to be detected use compression parameters the third
Classifier；
Described: compression parameters determination unit is specifically used for: according to the color component feature, using the third classifier
The differentiation that compression parameters are carried out to image to be detected determines the compression parameters used when the compression of described image to be detected.
It should be noted that can be referred to for the detailed description for the image detection device that the application 3rd embodiment provides
To the associated description of the application first embodiment, which is not described herein again.
Corresponding with a kind of generation method of sorter model of above-mentioned offer, the application fourth embodiment additionally provides
A kind of generating means of sorter model.
As shown in fig. 7, the generating means of sorter model include: image determination unit 701, feature extraction unit, 702 moulds
Type training unit 703.
Image determination unit 701, for determining the image for participating in training；
Feature extraction unit 702, for from described image zooming-out at least one color component feature for participating in training；
Model training unit 703, for according to the color component feature, training sorter model, the classifier mould
Type is initial intact compressed format image or is the image through overcompression for detecting image to be detected.
Optionally, which is characterized in that the feature extraction unit, comprising:
Color space conversion subelement, for the image for participating in training to be transformed into YCbCr face from RGB color
The colour space；
Color component extracts subelement, described for extracting from the YCbCr color space of the image for participating in training
At least one color component feature.
Optionally, the color component extracts subelement, is specifically used for:
Y color component feature and Cb color component are extracted from the YCbCr color space of the image for participating in training
Feature, alternatively, extracting Y color component feature and Cr color point from the YCbCr color space of the image for participating in training
Measure feature.
Optionally, described device further include:
Merging features unit is spliced for splicing the Y color component feature and the Cb component characterization
Described image feature afterwards, alternatively, the Y color component feature and the Cr component characterization are spliced, after obtaining splicing
Characteristics of image；
The model training unit, is specifically used for: according to spliced characteristics of image, training sorter model.
Optionally, it is described from it is described participate in training image YCbCr color space in extract Y color component feature and
Cb color component feature, alternatively, extracting Y color component feature from the YCbCr color space of the image for participating in training
With Cr color component feature, comprising:
Using in the richness model of airspace Spam11 submodel and minmax24 submodel from the image for participating in training
Y color component feature and Cb color component feature are extracted in YCbCr color space, alternatively, using in the richness model of airspace
Spam11 submodel and minmax24 submodel extract Y color from the YCbCr color space of the image for participating in training
Component characterization and Cr color component feature.
Optionally, described image determination unit is specifically used for:
Determine that sample set, the sample set include First Kind Graph picture and the second class image, the First Kind Graph picture is original
Nondestructive compression type image, the second class image are the image through overcompression；
It is described from described image zooming-out at least one color component feature for participating in training, comprising: from the First Kind Graph
Picture and the second class image zooming-out at least one color component feature.
Optionally, the sorter model includes: for judging whether image to be detected is initial intact compressed format figure
The first classifier of picture；
The model training unit, is specifically used for:
According to the color component feature, the first described classifier of training.
Optionally, the sorter model further include:
For determining second of classifier of the tool of compression type of image to be detected use；
The model training unit, is also used to:
According to the color component feature, training second of classifier.
Optionally, the sorter model further include:
For determining the third classifier of the compression parameters of image to be detected use；
The model training unit, is also used to:
According to the color component feature, the third described classifier of training.
It should be noted that for the application fourth embodiment provide Image Classifier model generating means it is detailed
Description can be with the associated description of reference pair the application second embodiment, and which is not described herein again.
Corresponding with a kind of image detecting method of above-mentioned offer, the 5th embodiment of the application additionally provides a kind of electronics
Equipment.
As shown in figure 8, electronic equipment includes:
Processor 801；And
Memory 802, for the program of image detecting method, which, which is powered and passes through the processor, runs the image
After the program of detection method, following step is executed:
At least one color component feature is extracted from image to be detected；
According to the color component feature, sorter model is used to differentiate that described image to be detected compresses lattice for initial intact
Formula image is the image through overcompression.
It is optionally, described that at least one color component feature is extracted from image to be detected, comprising:
Described image to be detected is transformed into YCbCr color space from RGB color；
At least one described color component feature is extracted from the YCbCr color space of described image to be detected.
Optionally, described to extract at least one described color component from the YCbCr color space of described image to be detected
Feature, comprising:
Y color component feature and Cb color component feature are extracted from the YCbCr color space of described image to be detected,
Alternatively, extracting Y color component feature and Cr color component feature from the YCbCr color space of described image to be detected.
Optionally, following step is also executed:
The Y color component feature and the Cb color component feature are spliced, spliced described image is obtained
It is special to obtain spliced image alternatively, the Y color component feature and the Cr color component feature are spliced for feature
Sign；
It is described according to the color component feature, use sorter model differentiate described image to be detected for original image or
Person is the image through overcompression, comprising: according to spliced characteristics of image, differentiates described image to be detected using sorter model
It for original image or is the image through overcompression.
Optionally, Y color component feature and Cb face are extracted in the YCbCr color space from described image to be detected
Colouring component feature, alternatively, extracting Y color component feature and Cr color from the YCbCr color space of described image to be detected
Component characterization, comprising:
Using in the richness model of airspace Spam11 submodel and minmax24 submodel from the YCbCr of described image to be detected
Y color component feature and Cb color component feature are extracted in color space, alternatively, using Spam11 in the richness model of airspace
Model and minmax24 submodel extract Y color component feature and Cr from the YCbCr color space of described image to be detected
Color component feature.
Optionally, the sorter model includes: for judging whether image to be detected is initial intact compressed format figure
The first classifier of picture；
It is described according to the color component feature, use sorter model to differentiate described image to be detected for initial intact pressure
Contracting format-pattern is the image through overcompression, comprising:
According to the color component feature, described image to be detected is differentiated using the first described classifier；
Initial intact compressed format image is all determined as according to the first all classifier, it is determined that the mapping to be checked
As being initial intact compressed format image；
Differentiate that described image to be detected is not initial intact compressed format image according at least one the first classifier,
Then determine that described image to be detected is the image through overcompression.
Optionally, following step is also executed: when the classification for determining described image to be detected is the image through overcompression, really
Determine the tool of compression used when compressing to described image to be detected.
Optionally, the sorter model, further includes: for determining the of tool of compression type that image to be detected uses
Two kinds of classifiers；
Determine the tool of compression used when compressing to described image to be detected, comprising:
According to the color component feature, is differentiated using second of classifier and generate differentiation result；
By differentiation result combination producing one coding of each second of classifier；
Calculate the Hamming distances of the coding of the coding and preset tool of compression；
The compression that will be used when being compressed with the coding shortest tool of compression of Hamming distances as described image to be detected
Tool.
Optionally, it also executes following step: determining the compression parameters used when the compression of described image to be detected.
Optionally, the sorter model, further includes: for determine image to be detected use compression parameters the third
Classifier；
The compression parameters used when determination image to be detected compression, comprising: according to the color component feature,
The differentiation for being carried out compression parameters to image to be detected using the third classifier is used when determining the compression of described image to be detected
Compression parameters.
Corresponding with a kind of generation method of sorter model of above-mentioned offer, the application sixth embodiment additionally provides
A kind of electronic equipment.
As shown in figure 9, electronic equipment includes:
Processor 901；And
Memory 902, the program of the generation method for sorter model, the equipment are powered and pass through the processor and transports
After the program of the generation method of the row sorter model, following step is executed:
Determine the image for participating in training；
From at least one color component feature of the image zooming-out of the participation training；
According to the color component feature, training sorter model, the sorter model is for detecting image to be detected
It for initial intact compressed format image or is the image through overcompression.
It is optionally, described from least one color component feature of the image zooming-out of the participation training, comprising:
The image for participating in training is transformed into YCbCr color space from RGB color；
At least one described color component feature is extracted from the YCbCr color space of the image for participating in training.
Optionally, the YCbCr color space from the image for participating in training extracts at least one described color
Component characterization, comprising:
Y color component feature and Cb color component are extracted from the YCbCr color space of the image for participating in training
Feature, alternatively, extracting Y color component feature and Cr color point from the YCbCr color space of the image for participating in training
Measure feature.
Optionally, following step is also executed:
The Y color component feature and the Cb component characterization are spliced, spliced described image feature is obtained,
Alternatively, the Y color component feature and the Cr component characterization are spliced, spliced characteristics of image is obtained；
It is described according to the color component feature, training sorter model, comprising: according to spliced characteristics of image, instruction
Practice sorter model.
Optionally, it is described from it is described participate in training image YCbCr color space in extract Y color component feature and
Cb color component feature, alternatively, extracting Y color component feature from the YCbCr color space of the image for participating in training
With Cr color component feature, comprising:
Using in the richness model of airspace Spam11 submodel and minmax24 submodel from the image for participating in training
Y color component feature and Cb color component feature are extracted in YCbCr color space, alternatively, using in the richness model of airspace
Spam11 submodel and minmax24 submodel extract Y color from the YCbCr color space of the image for participating in training
Component characterization and Cr color component feature.
Optionally, the determining image for participating in training, comprising:
Determine that sample set, the sample set include First Kind Graph picture and the second class image, the First Kind Graph picture is original
Nondestructive compression type image, the second class image are the image through overcompression；
It is described from described image zooming-out at least one color component feature for participating in training, comprising: from the First Kind Graph
Picture and the second class image zooming-out at least one color component feature.
Optionally, the sorter model includes: for judging whether image to be detected is initial intact compressed format figure
The first classifier of picture；
It is described according to the color component feature, training sorter model, comprising:
According to the color component feature, the first described classifier of training.
Optionally, the sorter model further include:
For determining second of classifier of the tool of compression type of image to be detected use；
It is described according to the color component feature, training sorter model, further includes:
According to the color component feature, training second of classifier.
Optionally, the sorter model further include:
For determining the third classifier of the compression parameters of image to be detected use；
It is described according to the color component feature, training sorter model, further includes:
According to the color component feature, the third described classifier of training.
The 7th embodiment of the application provides a kind of storage equipment,
It is stored with the program for storing image detecting method, which is run by processor, execute following step:
At least one color component feature is extracted from image to be detected；
According to the color component feature, sorter model is used to differentiate that described image to be detected compresses lattice for initial intact
Formula image is the image through overcompression.
The 8th embodiment of the application provides another storage equipment,
It is stored with the program of the generation method for storing sorter model, which is run by processor, executes following
Step:
Determine the image for participating in training；
From at least one color component feature of the image zooming-out of the participation training；
According to the color component feature, training sorter model, the sorter model is for detecting image to be detected
It for initial intact compressed format image or is the image through overcompression.
Although the application is disclosed as above with preferred embodiment, it is not for limiting the application, any this field skill
Art personnel are not departing from spirit and scope, can make possible variation and modification, therefore the guarantor of the application
Shield range should be subject to the range that the claim of this application defined.
In a typical configuration, calculating equipment includes one or more processors (CPU), input/output interface, net
Network interface and memory.
Memory may include the non-volatile memory in computer-readable medium, random access memory (RAM) and/or
The forms such as Nonvolatile memory, such as read-only memory (ROM) or flash memory (flash RAM).Memory is computer-readable medium
Example.
Computer-readable medium includes permanent and non-permanent, removable and non-removable media can be by any method
Or technology come realize information store.Information can be computer readable instructions, data structure, the module of program or other data.
The example of the storage medium of computer includes, but are not limited to phase change memory (PRAM), static random access memory (SRAM), moves
State random access memory (DRAM), other kinds of random access memory (RAM), read-only memory (ROM), electric erasable
Programmable read only memory (EEPROM), flash memory or other memory techniques, read-only disc read only memory (CD-ROM) (CD-ROM),
Digital versatile disc (DVD) or other optical storage, magnetic cassettes, tape magnetic disk storage or other magnetic storage devices
Or any other non-transmission medium, can be used for storage can be accessed by a computing device information.As defined in this article, it calculates
Machine readable medium does not include non-temporary computer readable media (transitory media), such as the data-signal and carrier wave of modulation.
It will be understood by those skilled in the art that embodiments herein can provide as method, system or computer program product.
Therefore, complete hardware embodiment, complete software embodiment or embodiment combining software and hardware aspects can be used in the application
Form.It is deposited moreover, the application can be used to can be used in the computer that one or more wherein includes computer usable program code
The shape for the computer program product implemented on storage media (including but not limited to magnetic disk storage, CD-ROM, optical memory etc.)
Formula.
Claims (25)
1. a kind of image detecting method characterized by comprising
At least one color component feature is extracted from image to be detected；
According to the color component feature, sorter model is used to differentiate described image to be detected for initial intact compressed format figure
Picture is the image through overcompression.
2. the method according to claim 1, wherein described extract at least one color point from image to be detected
Measure feature, comprising:
Described image to be detected is transformed into YCbCr color space from RGB color；
At least one described color component feature is extracted from the YCbCr color space of described image to be detected.
3. according to the method described in claim 2, it is characterized in that, the YCbCr color space from described image to be detected
Extract at least one described color component feature, comprising:
Y color component feature and Cb color component feature are extracted from the YCbCr color space of described image to be detected, or
Person extracts Y color component feature and Cr color component feature from the YCbCr color space of described image to be detected.
4. according to the method described in claim 3, it is characterized by further comprising:
The Y color component feature and the Cb color component feature are spliced, spliced described image feature is obtained,
Alternatively, the Y color component feature and the Cr color component feature are spliced, spliced characteristics of image is obtained；
It is described according to the color component feature, use sorter model to differentiate described image to be detected for original image or be
Image through overcompression, comprising: according to spliced characteristics of image, sorter model is used to differentiate described image to be detected for original
Beginning image is the image through overcompression.
5. according to the method described in claim 3, it is characterized in that, the YCbCr color space from described image to be detected
In extract Y color component feature and Cb color component feature, alternatively, from the YCbCr color space of described image to be detected
Extract Y color component feature and Cr color component feature, comprising:
Using in the richness model of airspace Spam11 submodel and minmax24 submodel from the YCbCr color of described image to be detected
Y color component feature and Cb color component feature are extracted in space, alternatively, using the Spam11 submodel in the richness model of airspace
Y color component feature and Cr color are extracted from the YCbCr color space of described image to be detected with minmax24 submodel
Component characterization.
6. the method according to claim 1, wherein the sorter model includes: for judging mapping to be checked
It seem no the first classifier for initial intact compressed format image；
It is described according to the color component feature, use sorter model to differentiate that described image to be detected is initial intact compression lattice
Formula image is the image through overcompression, comprising:
According to the color component feature, described image to be detected is differentiated using the first described classifier；
Initial intact compressed format image is all determined as according to the first all classifier, it is determined that described image to be detected is
Initial intact compressed format image；
Differentiate that described image to be detected is not initial intact compressed format image according at least one the first classifier, then really
Fixed described image to be detected is the image through overcompression.
7. according to the method described in claim 6, it is characterized by further comprising: when the classification for determining described image to be detected is
When image through overcompression, the tool of compression used when compressing to described image to be detected is determined.
8. the method according to the description of claim 7 is characterized in that the sorter model, further includes: to be detected for determining
Second of classifier of the tool of compression type that image uses；
Determine the tool of compression used when compressing to described image to be detected, comprising:
According to the color component feature, is differentiated using second of classifier and generate differentiation result；
By differentiation result combination producing one coding of each second of classifier；
Calculate the Hamming distances of the coding of the coding and preset tool of compression；
The tool of compression that will be used when being compressed with the coding shortest tool of compression of Hamming distances as described image to be detected.
9. method according to claim 7 or 8, which is characterized in that further include: make when determining the compression of described image to be detected
Compression parameters.
10. according to the method described in claim 9, it is characterized in that, the sorter model, further includes: to be checked for determining
The third classifier for the compression parameters that altimetric image uses；
The compression parameters that described determination described image to be detected uses when compressing, comprising: according to the color component feature, use
The third classifier carries out the differentiation of compression parameters to image to be detected, determines the compression used when the compression of described image to be detected
Parameter.
11. a kind of generation method of sorter model characterized by comprising
Determine the image for participating in training；
From at least one color component feature of the image zooming-out of the participation training；
According to the color component feature, training sorter model, the sorter model is for detecting image to be detected as original
Beginning nondestructive compression type image is the image through overcompression.
12. according to the method for claim 11, which is characterized in that described from the image zooming-out at least one for participating in training
A color component feature, comprising:
The image for participating in training is transformed into YCbCr color space from RGB color；
At least one described color component feature is extracted from the YCbCr color space of the image for participating in training.
13. according to the method for claim 12, which is characterized in that the YCbCr face from the image for participating in training
The colour space extracts at least one described color component feature, comprising:
Y color component feature and Cb color component feature are extracted from the YCbCr color space of the image for participating in training,
Alternatively, extracting Y color component feature and Cr color component spy from the YCbCr color space of the image for participating in training
Sign.
14. according to the method for claim 13, which is characterized in that further include:
The Y color component feature and the Cb component characterization are spliced, spliced described image feature is obtained, or
The Y color component feature and the Cr component characterization are spliced, obtain spliced characteristics of image by person；
It is described according to the color component feature, training sorter model, comprising: according to spliced characteristics of image, training point
Class device model.
15. according to the method for claim 13, which is characterized in that the YCbCr face from the image for participating in training
Y color component feature and Cb color component feature are extracted in the colour space, alternatively, from the YCbCr of the image for participating in training
Y color component feature and Cr color component feature are extracted in color space, comprising:
Using in the richness model of airspace Spam11 submodel and minmax24 submodel from it is described participate in training image YCbCr
Y color component feature and Cb color component feature are extracted in color space, alternatively, using Spam11 in the richness model of airspace
Model and minmax24 submodel extract Y color component feature from the YCbCr color space of the image for participating in training
With Cr color component feature.
16. according to the method for claim 11, which is characterized in that the determining image for participating in training, comprising:
Determine that sample set, the sample set include First Kind Graph picture and the second class image, the First Kind Graph picture is initial intact
Compressed format image, the second class image are the image through overcompression；
It is described from described image zooming-out at least one color component feature for participating in training, comprising: from the First Kind Graph picture and
At least one color component feature of second class image zooming-out.
17. according to the method for claim 11, which is characterized in that the sorter model includes: to be detected for judging
Image whether be initial intact compressed format image the first classifier；
It is described according to the color component feature, training sorter model, comprising:
According to the color component feature, the first described classifier of training.
18. according to the method for claim 17, which is characterized in that the sorter model further include:
For determining second of classifier of the tool of compression type of image to be detected use；
It is described according to the color component feature, training sorter model, further includes:
According to the color component feature, training second of classifier.
19. according to the method for claim 18, which is characterized in that the sorter model further include:
For determining the third classifier of the compression parameters of image to be detected use；
It is described according to the color component feature, training sorter model, further includes:
According to the color component feature, the third described classifier of training.
20. a kind of image detection device characterized by comprising
Feature extraction unit, for extracting at least one color component feature from image to be detected；
Image discriminating unit, for according to the color component feature, use sorter model differentiate described image to be detected for
Initial intact compressed format image is the image through overcompression.
21. a kind of generating means of sorter model characterized by comprising
Image determination unit, for determining the image for participating in training；
Feature extraction unit, for from described image zooming-out at least one color component feature for participating in training；
Model training unit, for according to the color component feature, training sorter model, the sorter model to be for examining
Image to be detected is surveyed to be initial intact compressed format image or be the image through overcompression.
22. a kind of electronic equipment characterized by comprising
Processor；And
Memory, for the program of image detecting method, which is powered and runs the image detection side by the processor
After the program of method, following step is executed:
At least one color component feature is extracted from image to be detected；
According to the color component feature, sorter model is used to differentiate described image to be detected for initial intact compressed format figure
Picture is the image through overcompression.
23. a kind of electronic equipment characterized by comprising
Processor；And
Memory, the program of the generation method for sorter model, which, which is powered and passes through the processor, runs this point
After the program of the generation method of class device model, following step is executed:
Determine the image for participating in training；
From at least one color component feature of the image zooming-out of the participation training；
According to the color component feature, training sorter model, the sorter model is for detecting image to be detected as original
Beginning nondestructive compression type image is the image through overcompression.
24. a kind of storage equipment, which is characterized in that
It is stored with the program for storing image detecting method, which is run by processor, execute following step:
At least one color component feature is extracted from image to be detected；
According to the color component feature, sorter model is used to differentiate described image to be detected for initial intact compressed format figure
Picture is the image through overcompression.
25. a kind of storage equipment, which is characterized in that
It is stored with the program of the generation method for storing sorter model, which is run by processor, execute following step:
Determine the image for participating in training；
From at least one color component feature of the image zooming-out of the participation training；
According to the color component feature, training sorter model, the sorter model is for detecting image to be detected as original
Beginning nondestructive compression type image is the image through overcompression.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810309030.9A CN110363207A (en) | 2018-04-09 | 2018-04-09 | Image detection and the generation method of sorter model, device and equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810309030.9A CN110363207A (en) | 2018-04-09 | 2018-04-09 | Image detection and the generation method of sorter model, device and equipment |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110363207A true CN110363207A (en) | 2019-10-22 |
Family
ID=68213504
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810309030.9A Pending CN110363207A (en) | 2018-04-09 | 2018-04-09 | Image detection and the generation method of sorter model, device and equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110363207A (en) |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101674389A (en) * | 2009-09-30 | 2010-03-17 | 大连理工大学 | Method for testing compression history of BMP image based on loss amount of image information |
| CN101894256A (en) * | 2010-07-02 | 2010-11-24 | 西安理工大学 | Iris identification method based on odd-symmetric 2D Log-Gabor filter |
| CN102903100A (en) * | 2012-09-06 | 2013-01-30 | 宁波大学 | Detection method for digital photo image compression history |
| CN102959588A (en) * | 2011-04-28 | 2013-03-06 | 中国科学院自动化研究所 | Method for detecting tampering with color digital image based on chroma of image |
| CN104661037A (en) * | 2013-11-19 | 2015-05-27 | 中国科学院深圳先进技术研究院 | Tampering detection method and system for compressed image quantization table |
| CN105874472A (en) * | 2014-02-21 | 2016-08-17 | 三星电子株式会社 | Multi-band biometric camera system having iris color recognition |
| CN107256385A (en) * | 2017-05-22 | 2017-10-17 | 西安交通大学 | Infrared iris Verification System and method based on 2D Log Gabor Yu composite coding method |
| US20180054630A1 (en) * | 2016-08-19 | 2018-02-22 | Apple Inc. | Compression of image assets |
-
2018
- 2018-04-09 CN CN201810309030.9A patent/CN110363207A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101674389A (en) * | 2009-09-30 | 2010-03-17 | 大连理工大学 | Method for testing compression history of BMP image based on loss amount of image information |
| CN101894256A (en) * | 2010-07-02 | 2010-11-24 | 西安理工大学 | Iris identification method based on odd-symmetric 2D Log-Gabor filter |
| CN102959588A (en) * | 2011-04-28 | 2013-03-06 | 中国科学院自动化研究所 | Method for detecting tampering with color digital image based on chroma of image |
| CN102903100A (en) * | 2012-09-06 | 2013-01-30 | 宁波大学 | Detection method for digital photo image compression history |
| CN104661037A (en) * | 2013-11-19 | 2015-05-27 | 中国科学院深圳先进技术研究院 | Tampering detection method and system for compressed image quantization table |
| CN105874472A (en) * | 2014-02-21 | 2016-08-17 | 三星电子株式会社 | Multi-band biometric camera system having iris color recognition |
| US20180054630A1 (en) * | 2016-08-19 | 2018-02-22 | Apple Inc. | Compression of image assets |
| CN107256385A (en) * | 2017-05-22 | 2017-10-17 | 西安交通大学 | Infrared iris Verification System and method based on 2D Log Gabor Yu composite coding method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Liu et al. | Point cloud quality assessment: Dataset construction and learning-based no-reference metric | |
| US20230215197A1 (en) | Systems and Methods for Detection and Localization of Image and Document Forgery | |
| Luo et al. | A survey of passive technology for digital image forensics | |
| CN111445454B (en) | Image authenticity identification method and application thereof in license identification | |
| JP4732660B2 (en) | Visual attention system | |
| CN103366178B (en) | A kind of method and apparatus for being used to carry out target image color classification | |
| CA2851498C (en) | System and method for digital image signal compression using intrinsic images | |
| CN104661037B (en) | The detection method and system that compression image quantization table is distorted | |
| JP2006505075A (en) | Nonlinear quantization and similarity matching method for video sequence retrieval with multiple image frames | |
| Akbari et al. | A new forensic video database for source smartphone identification: Description and analysis | |
| Lo et al. | Exploring semantic segmentation on the dct representation | |
| Abdulrahman et al. | Color image stegananalysis using correlations between RGB channels | |
| CN102903100B (en) | Detection method for digital photo image compression history | |
| Arbeláez et al. | A metric approach to vector-valued image segmentation | |
| Li et al. | Distinguishing computer graphics from photographic images using a multiresolution approach based on local binary patterns | |
| US8437545B1 (en) | System and method for digital image signal compression using intrinsic images | |
| Nirmalkar et al. | A review of image forgery techniques and their detection | |
| CN103796017B (en) | Image discriminating device and method | |
| CN110363207A (en) | Image detection and the generation method of sorter model, device and equipment | |
| Taya et al. | Detecting tampered regions in JPEG images via CNN | |
| CN106407323A (en) | Picture retrieving method based on diversity spatial distance | |
| Azarian-Pour et al. | An automatic jpeg ghost detection approach for digital image forensics | |
| CN111931850A (en) | Cross-modal fusion significance detection method based on feature refinement | |
| Chen et al. | Image filter identification using demosaicing residual features | |
| Muselet et al. | Rank correlation as illumination invariant descriptor for color object recognition |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| REG | Reference to a national code |
Ref country code: HK Ref legal event code: DE Ref document number: 40016160 Country of ref document: HK |