CN110070005A

CN110070005A - Images steganalysis method, apparatus, storage medium and electronic equipment

Info

Publication number: CN110070005A
Application number: CN201910262290.XA
Authority: CN
Inventors: 何长伟; 汪铖杰; 李季檩; 熊意超; 李绍欣; 陈超; 葛彦昊; 倪辉; 吴永坚; 黄飞跃
Original assignee: Tencent Technology Shenzhen Co Ltd
Current assignee: Tencent Technology Shenzhen Co Ltd
Priority date: 2019-04-02
Filing date: 2019-04-02
Publication date: 2019-07-30

Abstract

The present invention provides a kind of images steganalysis method comprising: multiple identification images with relevance are obtained, and detect the Face datection frame and human testing frame in each identification image；Each Face datection frame identified in image and human testing frame are subjected to matching operation, to obtain the identification target in each identification image；Feature extraction is carried out to the identification target in each identification image, to obtain the target signature of the identification target in identification image；In chronological order, the target signature of corresponding identification target in multiple identification images is merged, to obtain the motion profile feature of corresponding identification target in multiple identification images.The present invention also provides a kind of images steganalysis devices, and the present invention is based on multiple target signatures of identification target, generate motion profile feature of the identification target in multiple identification images；The identification accuracy for identifying target in identification image is improved, the policer operation cost to identification target is reduced.

Description

Images steganalysis method, apparatus, storage medium and electronic equipment

Technical field

The present invention relates to image real time transfer fields, are situated between more particularly to a kind of images steganalysis method, apparatus, storage Matter and electronic equipment.

Background technique

With the development of science and technology, the picture control equipment of social public domain is more and more, so that entire public domain Safety index steps up.But the monitoring objective in existing monitoring image causes entirely to supervise generally by manually being set The cost of labor for controlling operation is higher.

Subsequent monitoring device developer detects the position of human body frame in each frame figure using human body detecting method, then adopts The position of human body frame of each frame is associated with global object association, incidence matrix is solved by linear integer programming.But When two people's clothes colors are close, corresponding position of human body frame is easy to appear tracking error.

Therefore existing monitoring image target identification method error rate is higher or policer operation higher cost.

Summary of the invention

The embodiment of the present invention provides that a kind of target identification accuracy is higher and the corresponding lower-cost image mesh of policer operation Mark recognition methods and images steganalysis device；To solve existing images steganalysis method and images steganalysis device The technical issues of error rate is higher or policer operation higher cost.

The embodiment of the present invention provides a kind of images steganalysis method comprising:

Multiple identification images with relevance are obtained, and detect the Face datection frame in each identification image and people Body detection block；

Each Face datection frame identified in image and human testing frame are subjected to matching operation, to obtain each institute State the identification target in identification image；

Feature extraction is carried out to the identification target in each identification image, to obtain the identification in the identification image At least a kind of target signature of target；And

In chronological order, the target signature of corresponding identification target in multiple identification images is merged, with The motion profile feature of corresponding identification target into multiple identification images.

The embodiment of the present invention also provides a kind of images steganalysis device comprising:

It identifies module of target detection, for obtaining multiple identification images with relevance, and detects each identification Face datection frame and human testing frame in image；

Object matching module is identified, for carrying out each Face datection frame identified in image and human testing frame Matching operation, to obtain the identification target in each identification image；

Target's feature-extraction module, for carrying out feature extraction to the identification target in each identification image, with To at least a kind of target signature of the identification target in the identification image；And

Fusion Features module is used in chronological order, to the targets of corresponding identification target in multiple identification images Feature is merged, to obtain the motion profile feature of corresponding identification target in multiple identification images.

In images steganalysis device described in the embodiment of the present invention, the identification module of target detection is same for obtaining Multiple video frames of one video file, as multiple identification images with relevance.

The embodiment of the present invention also provides a kind of storage medium, is stored with processor-executable instruction, described instruction by One or more processors load, to execute above-mentioned images steganalysis method.

The embodiment of the present invention also provides a kind of electronic equipment, including processor and memory, and the memory storage has meter Calculation machine program, the processor is by calling the computer program, for executing above-mentioned images steganalysis method.

Compared to the prior art, images steganalysis method, apparatus, storage medium and electronic equipment of the invention are based on knowing Multiple target signatures of other target generate motion profile feature of the identification target in multiple identification images；Improve identification figure The identification accuracy that target is identified as in, reduces the policer operation cost to identification target；The existing figure of effective solution As the technical issues of the error rate of target identification method and images steganalysis device is higher or policer operation higher cost.

Detailed description of the invention

Fig. 1 is the flow chart of the first embodiment of images steganalysis method of the invention；

Fig. 2 is the flow chart of the step S102 of the first embodiment of images steganalysis method of the invention；

Fig. 3 is the flow chart of the second embodiment of images steganalysis method of the invention；

Fig. 4 is the flow chart of the step S304 of the second embodiment of images steganalysis method of the invention；

Fig. 5 is the flow chart of the step S402 of the second embodiment of images steganalysis method of the invention；

Fig. 6 is the structural schematic diagram of the first embodiment of images steganalysis device of the invention；

Fig. 7 is the structural representation of the identification object matching module of the first embodiment of images steganalysis device of the invention Figure；

Fig. 8 is the structural schematic diagram of the second embodiment of images steganalysis device of the invention；

Fig. 9 is the structural schematic diagram of the Fusion Features module of the second embodiment of images steganalysis device of the invention；

Figure 10 is Fusion Features of the Fusion Features module of the second embodiment of images steganalysis device of the invention The structural schematic diagram of module；

Figure 11 a to Figure 11 c is images steganalysis method of the invention and three knowledges that images steganalysis device obtains The schematic diagram of other image；

Figure 12 is the images steganalysis flow chart of images steganalysis method and images steganalysis device of the invention；

Figure 13 is the working environment structural schematic diagram of the electronic equipment where images steganalysis device of the invention.

Specific embodiment

Schema is please referred to, wherein identical component symbol represents identical component, the principle of the present invention is to implement one It is illustrated in computing environment appropriate.The following description be based on illustrated by the specific embodiment of the invention, should not be by It is considered as the limitation present invention other specific embodiments not detailed herein.

In the following description, specific embodiments of the present invention will refer to the operation as performed by one or multi-section computer The step of and symbol illustrate, unless otherwise stating clearly.Therefore, these steps and operation be will appreciate that, mentioned for several times wherein having It include by representing with the computer disposal list of the electronic signal of the data in a structuring pattern to be executed by computer Member is manipulated.At this manipulation transforms data or the position being maintained in the memory system of the computer, it can match again Set or in addition change in a manner familiar to those skilled in the art the running of the computer.The maintained data knot of the data Structure is the provider location of the memory, has the specific feature as defined in the data format.But the principle of the invention is with above-mentioned Text illustrates, is not represented as a kind of limitation, those skilled in the art will appreciate that plurality of step as described below and Operation also may be implemented in hardware.

Images steganalysis method and images steganalysis device of the invention may be provided in any electronic equipment, use It identifies that target carries out identification operation in the video frame to video file, and then obtains fortune of the identification target in multiple video frames Dynamic rail mark.The electronic equipment includes but is not limited to wearable device, helmet, medical treatment & health platform, personal computer, service Device computer, hand-held or laptop devices, mobile device (such as mobile phone, personal digital assistant (PDA), media play Device etc.), multicomputer system, consumer electronic devices, minicomputer, mainframe computer including above-mentioned arbitrary system or Distributed computing environment of equipment, etc..The images steganalysis device is preferably monitoring image identification server, for prison It controls the identification target such as personage in video and carries out target identification operation or identification target trajectory identification operation.Figure of the invention As target identification method and images steganalysis device can effectively improve the accuracy of target identification, the prison of reduction identification target Control the difficulty and cost of operation.

Fig. 1 is please referred to, Fig. 1 is the flow chart of the first embodiment of images steganalysis method of the invention, the present embodiment Images steganalysis method above-mentioned electronic equipment can be used to be implemented, the images steganalysis method packet of the present embodiment It includes:

Step S101 obtains multiple identification images with relevance, and detects the Face datection in each identification image Frame and human testing frame；

Each Face datection frame identified in image and human testing frame are carried out matching operation, to obtain by step S102 Identification target in each identification image；

Step S103 carries out feature extraction to the identification target in each identification image, to obtain the knowledge in identification image At least a kind of target signature of other target；

Step S104, in chronological order, at least a kind of target signature of corresponding identification target in multiple identification images It is merged, to obtain the motion profile feature of corresponding identification target in multiple identification images.

The following detailed description of the images steganalysis process of the images steganalysis method of the present embodiment.

In step s101, images steganalysis device (i.e. monitoring image identification server), which obtains, has the more of relevance A identification image.Here identification image refers to the image at least one identification target, and identification target can be identification image In human target, animal target or object target (such as vehicle target).Multiple identification images with relevance refer to more There is same identification target in a identification image, and there is regular hour continuity and geographical position between multiple identification images Set continuity.

The multiple successive video frames for such as obtaining same monitor video file can be used as multiple identification figures with relevance Picture.The monitor video file generally monitor for a long time same geographical location or adjacent multiple geographical locations video pictures (depending on Frequency frame), therefore the video pictures have geographical location continuity；And multiple successive video frames of monitor video file are usually to have Having time is successional.

Subsequent picture Target Identification Unit detects the Face datection frame and human testing frame in each identification image.Such as identification Target is human target, then can detect the Face datection in each identification image by detection algorithms such as multi-class detection algorithms Frame and human testing frame.Here Face datection frame is with the parameter of human face characteristic point (position of such as human face characteristic point or face The size etc. of characteristic point) it is the main coordinate frame for identifying object, human testing frame is that (such as human body is special with the parameter of human body feature point Levy position or the size of human body feature point etc. of point) it is the main coordinate frame for identifying object.

In step s 102, for images steganalysis device using human target as identification target, each identification target is right A face detection block and a human testing frame are answered, therefore is needed in this step to the Face datection in each identification image Frame and human testing frame carry out matching operation, to obtain the identification target in each identification image.

It specifically can refer to Fig. 2, Fig. 2 is the stream of the step S102 of the first embodiment of images steganalysis method of the invention Cheng Tu.Step S102 includes:

Step S201, images steganalysis device obtain the face center coordinate in identification image in Face datection frame with And the number of people center point coordinate in human testing frame.

Step S202, due in the face center coordinate and human testing frame in the Face datection frame of each identification target Number of people center point coordinate should be consistent, therefore images steganalysis device can be used bipartite graph matching algorithm, according to people The distance relation of face center point coordinate and number of people center point coordinate obtains face center coordinate and human body in Face datection frame The matching relationship of number of people center point coordinate in detection block.

The people in the face center coordinate Ci and multiple human testing frames in multiple Face datection frames can specifically be obtained Head center point coordinate Hj, wherein i is the label of face detection block, and j is the label of human testing frame.Then calculate each face inspection Survey probability Wij that frame and each human testing frame are same identification target (the face center coordinate Ci of Face datection frame with it is right Answer the distance of the number of people center point coordinate Hj of human testing frame closer, probability Wij is bigger).KM algorithm specifically can be used (Kuhn-Munkers), probability and W of the face detection block with corresponding human testing frame for same identification target after matching are obtained Maximum value, it may be assumed that

Wherein λ is preset constant；

The number of people central point in the face center coordinate Ci and human testing frame in Face datection frame can be obtained in this way The matching relationship of the matching relationship of coordinate Hj, i.e. Face datection frame and corresponding human testing frame.

Here KM algorithm is a kind of bipartite graph best match algorithm, and wherein bipartite graph is that " one group of point set can be divided into two Point, and a point is mutually not attached in every part, can have side between two-part point ".Bipartite graph best match algorithm is i.e. for two A subset is found on all sides of component, which meets following two condition:

1, any two sides are all independent of the same point；

2, allow in this subset meet it is more as far as possible in the case where condition 1.

Step S203, images steganalysis device by step S202 obtain with the Face datection frame of matching relationship and people Body detection block synthesizes the identification target in identification image, so that subsequent step carries out identification clarification of objective analysis.

In step s 103, images steganalysis device carries out the step S102 identification target for obtaining each identification image Feature extraction, to obtain at least a kind of target signature of the identification target in identification image；It is specific to obtain in identification image The corresponding face characteristic class target signature of all identification targets, characteristics of human body's class target signature and position class target signature institute group At at least a kind of target signature selected in group.

Due to identification target between may exist identification target overlapping, here in order to will identify target carry out accurately Differentiation, here images steganalysis device can obtain identification target a variety of different classes of target signatures.

Specifically, images steganalysis device can obtain the corresponding face matter of all identification targets in each identification image Amount point, clothes attribute value, human face characteristic point, human body feature point and identification target position information.Wherein face quality point and Human face characteristic point is face characteristic class target signature, and clothes attribute value and human body feature point are characteristics of human body's class target feature, Identification target position information is position class target signature.Images steganalysis device may be selected to use human face characteristic point, people in this way Body characteristics point or identification target position information carry out Fusion Features.

In step S104, images steganalysis device in chronological order, to step S103 extract multiple identification images In it is corresponding identification target target signature merged.

Specifically, images steganalysis device is special to the motion profile of corresponding identification target in multiple identification images first Sign initializes.

The identification image that subsequent picture Target Identification Unit acquisition time sequence is 1, and the identification figure for being 1 by time sequencing As the target signature of corresponding identification target, the motion profile feature of corresponding identification target in image is identified directly as 1. The target signature for the corresponding identification target of identification image that subsequent acquisition time sequence is 1, is fused in 1 identification image corresponding Identification target motion profile feature in, it is special with the motion profile for obtaining corresponding identification target in 2 identification images Sign.... the target signature for the corresponding identification target of identification image that last acquisition time sequence is m is fused to m-1 identification In image in the motion profile feature of corresponding identification target, to obtain the movement of corresponding identification target in m identification image Track characteristic, wherein m is the quantity for identifying image.

The images steganalysis process and target movement of the images steganalysis method of the present embodiment are completed in this way Track acquisition process.

Multiple target signatures of the images steganalysis method of the present embodiment based on identification target generate identification target more Motion profile feature in a identification image；The identification accuracy for identifying target in identification image is improved, is reduced to identification The policer operation cost of target.

Referring to figure 3., Fig. 3 is the flow chart of the second embodiment of images steganalysis method of the invention, the present embodiment Images steganalysis method above-mentioned electronic equipment can be used to be implemented, the images steganalysis method packet of the present embodiment It includes:

Step S301 obtains multiple identification images with relevance, and detects the Face datection in each identification image Frame and human testing frame；

Each Face datection frame identified in image and human testing frame are carried out matching operation, to obtain by step S302 Identification target in each identification image；

Step S303 carries out feature extraction to the identification target in each identification image, to obtain the knowledge in identification image At least a kind of target signature of other target；

Step S304 in chronological order merges the target signature of corresponding identification target in multiple identification images, To obtain the motion profile feature of corresponding identification target in multiple identification images.

Wherein the first of step S301, the specific embodiment of step 302 and above-mentioned images steganalysis method is implemented The specific embodiment of step S101 and step S12 in example are same or similar, specifically refer to above-mentioned images steganalysis Associated description in the first embodiment of method.

In step S303, images steganalysis device carries out the step S302 identification target for obtaining each identification image Feature extraction, to obtain at least a kind of target signature of the identification target in identification image.

Wherein human face characteristic point is the characteristic point identified in the corresponding Face datection frame of target, the characteristic area of human face characteristic point Indexing can be judged by face quality point.Wherein face quality is divided into the feature fine degree of human face characteristic point, face quality Point higher, the fine degree of human face characteristic point is higher, easier to be distinguished by human face characteristic point to identification target.

Human body feature point is the characteristic point identified in the corresponding human testing frame of target, the characteristic area indexing of human body feature point It can be judged by clothes attribute value.Clothes attribute value can be the primary color difference value etc. of human body feature point.Such as different identifications The different identification target of clothes attribute value is inevitable different on image.The identical identification mesh of clothes attribute value on difference identification image Mark, needs to distinguish by human body feature point.

Identification target position information is the location information for identifying target in identification image, it is however generally that, it is adjacent on the time Identification image in it is same identification target location information answer it is roughly the same or similar.

In step s 304, images steganalysis device in chronological order, to step S303 extract multiple identification images In it is corresponding identification target target signature carry out mixing operation.

Specifically referring to figure 4., Fig. 4 is the stream of the step S304 of the second embodiment of images steganalysis method of the invention Cheng Tu.Step S304 includes:

Step S401 initializes the motion profile feature of corresponding identification target in multiple identification images, And n=1 is set, wherein n is used for the counting operation of Fusion Features.

Here images steganalysis device can directly acquire corresponding identification target in the identification image that time sequencing is 1 Quantity q, the initialization motion profile feature of q identification target settable in this way, the corresponding 1 identification mesh of each motion profile feature Mark.

Step S402, the target signature for identifying corresponding identification target in image for being n by time sequencing, is fused to movement In track characteristic, to obtain the motion profile feature of corresponding identification target in n identification image.

As n=1, the corresponding identification target of identification image that time sequencing is directly 1 by images steganalysis device Target signature identifies the motion profile feature of corresponding identification target in image as 1.The motion profile feature includes but not It is limited to identify the human face characteristic point, human body feature point and identification target position information of target.

When n is greater than 1, the mesh for the corresponding identification target of identification image that time sequencing is n by images steganalysis device Feature is marked, is fused in n-1 identification image in the motion profile feature of corresponding identification target, to obtain n identification image In it is corresponding identification target motion profile feature.

Step S403 carries out counting operation, i.e. n=n+1, and return step S402 to n and carries out motion profile Fusion Features Operation, until obtaining the motion profile feature of corresponding identification target in m identification image, wherein m is the quantity for identifying image.

Specifically referring to figure 5., Fig. 5 is the stream of the step S402 of the second embodiment of images steganalysis method of the invention Cheng Tu.Step S402 includes:

Step S501, images steganalysis device judge whether to meet first condition, which is that time sequencing is n The corresponding identification target of identification image target signature in face quality point be greater than the first setting value, and n-1 identification is schemed Face quality point as in the motion profile feature of corresponding identification target is greater than the first setting value.Such as it is unsatisfactory for first Part goes to step S502, such as meets first condition, goes to step S503.

Step S502, images steganalysis device judge whether to meet second condition, which is that time sequencing is n Identification image it is corresponding identification target target signature in clothes attribute value, with corresponding identification in n-1 identification image Clothes attribute value in the motion profile feature of target is greater than the second setting value.Such as meet second condition, goes to step S504；Such as It is unsatisfactory for second condition, goes to step S505.

Step S503, if time sequencing is the face quality in the target signature for identifying the corresponding identification target of image of n Divide and be greater than the first setting value, and the face quality point in n-1 identification image in the motion profile feature of corresponding identification target Greater than the first setting value.Then explanation can calculate the phase that target is identified in different identification images by human face characteristic point in identification target Like degree.

Specifically, the target for the corresponding identification target of identification image that images steganalysis device is n according to time sequencing Face characteristic in human face characteristic point and n-1 identification image in feature in the motion profile feature of corresponding identification target Point determines that the corresponding identification target of identification image that time sequencing is n identifies the corresponding phase for identifying target in image with n-1 Like degree.

As time sequencing be n identification image it is corresponding identification target target signature in human face characteristic point be Fa, n-1 Human face characteristic point in a identification image in the motion profile feature of corresponding identification target is Fb, and it is n that wherein a, which is time sequencing, Identification image in identify target label, b be n-1 identification image in identify target label, then time sequencing for n knowledge The similarity S1 of identification target and the identification target in n-1 identification image in other image are as follows: S1=(Fa*Fb)/(| Fa | * | Fb|)；Then pass to step S506.

Step S504, is such as unsatisfactory for first condition, meets second condition, i.e. time sequencing is that the identification image of n is corresponding It identifies the clothes attribute value in the target signature of target, identifies that the motion profile of corresponding identification target in image is special with n-1 Clothes attribute value in sign is greater than the second setting value.Then explanation can calculate different identifications by human body feature point in identification target and scheme The similarity of target is identified as in.

Specifically, the target for the corresponding identification target of identification image that images steganalysis device is n according to time sequencing Characteristics of human body in human body feature point and n-1 identification image in feature in the motion profile feature of corresponding identification target Point determines that the corresponding identification target of identification image that time sequencing is n identifies the corresponding phase for identifying target in image with n-1 Like degree.

As time sequencing be n identification image it is corresponding identification target target signature in human body feature point be Ga, n-1 Human body feature point in a identification image in the motion profile feature of corresponding identification target is Gb, and it is n that wherein a, which is time sequencing, Identification image in identify target label, b be n-1 identification image in identify target label, then time sequencing for n knowledge The corresponding similarity S2 for identifying target in the other corresponding identification target of image and n-1 identification image are as follows: S2=(Ga*Gb)/ (|Ga|*|Gb|)；Then pass to step S506.

Step S505 is such as unsatisfactory for first condition and second condition simultaneously, then illustrate to identify the face quality point of target compared with Difference does not fit through the human face characteristic point in identification target and calculates the similarity for identifying target in different identification images；Know simultaneously The difference of the human body feature point of other target is smaller, is also not suitable for calculating different identification figures by the human body feature point in identification target The similarity of target is identified as in.At this moment different identification images can only be calculated by the identification target position information in identification target The similarity of middle identification target.

As time sequencing be n identification image it is corresponding identification target target signature in identification target position information be Identification target position information in Ha, n-1 identification images in the motion profile feature of corresponding identification target is Hb, wherein a For time sequencing be n identification image in identify target label, b be n-1 identification image in identification target label, then when Between sequentially for n identification image it is corresponding identification target with n-1 identification image in it is corresponding identify target similarity S3 are as follows: S3=iou (Ha, Hb)；Wherein iou function is to hand over and than function (Intersection-over-Union), for calculating Ha pairs The corresponding feature frame of the identification target position information answered (here can be corresponding human testing frame) identification target corresponding with Hb The degree of overlapping of the corresponding feature frame of location information.Then pass to step S506.

Step S506, the corresponding identification target of identification image that acquisition time sequence is n from step S503 to step S505 Identify that images steganalysis device uses bipartite graph matching algorithm in image after the similarity of corresponding identification target with n-1, The corresponding identification target of identification image for being n according to time sequencing is corresponding with n-1 identification image to identify the similar of target Degree, the corresponding identification target of identification image that acquisition time sequence is n identify of corresponding identification target in image with n-1 With relationship.

Specifically can acquisition time sequence be n the corresponding multiple identification targets of identification image and n-1 identification image in Corresponding multiple identification targets, and the identification image that above-mentioned time sequencing is n is obtained by step S503 to step S505 and is corresponded to Any identification target and n-1 identification image in it is corresponding it is any identification target between similarity；It then can be used two points Figure matching algorithm, calculate matching after time sequencing be n the corresponding identification target of identification image with it is right in n-1 identification image The identification target answered is the maximum value of the similarity sum of same identification target.In this way can acquisition time sequence be n identification figure As corresponding identification target identifies the corresponding matching relationship for identifying target in image with n-1.

Step S507, the target for the corresponding identification target of identification image that time sequencing is n by images steganalysis device Feature is fused in n-1 identification image in the motion profile feature of the identification target with matching relationship, to obtain n knowledge To the motion profile feature of the identification target of drink in other image.

On the basis of first embodiment, the images steganalysis method of the present embodiment is according to different target characteristic attribute It identifies target, generates the similarity for identifying target in different identification images in different ways, improve and obtain different identifications The accuracy of corresponding identification target, to improve the accuracy of Fusion Features operation, and then further improves knowledge in image The identification accuracy of target is identified in other image.

The present invention also provides a kind of images steganalysis devices, please refer to Fig. 6, and Fig. 6 is images steganalysis of the invention The structural schematic diagram of the first embodiment of device.Above-mentioned image object can be used to know for the images steganalysis device of the present embodiment The first embodiment of other method is implemented.The images steganalysis device 60 includes identification module of target detection 61, identification mesh Mark matching module 62, target's feature-extraction module 63 and Fusion Features module 64.

Identification module of target detection 61 is used to obtain multiple identification images with relevance, and detects each identification image In Face datection frame and human testing frame；Identify that object matching module 62 is used for the Face datection frame in each identification image Matching operation is carried out with human testing frame, to obtain the identification target in each identification image；Target's feature-extraction module 63 is used In carrying out feature extraction to the identification target in each identification image, to obtain at least a kind of of the identification target in identification image Target signature；Fusion Features module 64 is in chronological order, identifying at least the one of target to corresponding in multiple identification images Class target signature is merged, to obtain the motion profile feature of corresponding identification target in multiple identification images.

Fig. 7 is please referred to, Fig. 7 is the identification object matching module of the first embodiment of images steganalysis device of the invention Structural schematic diagram.The identification object matching module 62 includes coordinate acquisition submodule 71, coordinate matching submodule 72 and closes At submodule 73.

Coordinate acquisition submodule 71 be used for obtains identify image in Face datection frame in face center coordinate and Number of people center point coordinate in human testing frame；Coordinate matching submodule 72 is used to use bipartite graph matching algorithm, according to face The distance relation of center point coordinate and number of people center point coordinate, the face center coordinate and human body obtained in Face datection frame are examined Survey the matching relationship of the number of people center point coordinate in frame；Synthesis submodule 73 be used for will have matching relationship Face datection frame and Human testing frame synthesizes the identification target in identification image.

The images steganalysis device 60 of the present embodiment is in use, identification module of target detection 61 first is obtained with association Multiple identification images of property.Here identification image refers to the image at least one identification target.With the more of relevance A identification image, which refers to, has same identification target in multiple identification images, and has the regular hour between multiple identification images Continuity and geographical location continuity.

Then identification module of target detection 61 detects the Face datection frame and human testing frame in each identification image.As known Other target is human target, then can detect the face inspection in each identification image by detection algorithms such as multi-class detection algorithms Survey frame and human testing frame.

Then for identification object matching module 62 using human target as identification target, each identification target corresponds to a people Face detection block and a human testing frame, therefore identify that object matching module is needed to the Face datection frame in each identification image Matching operation is carried out with human testing frame, to obtain the identification target in each identification image.

Specifically can include:

Identify that the coordinate acquisition submodule 71 of object matching module 62 obtains the face in identification image in Face datection frame Number of people center point coordinate in center point coordinate and human testing frame.

Identify that the coordinate matching submodule 72 of object matching module 62 uses bipartite graph matching algorithm, according to face center The distance relation of coordinate and number of people center point coordinate obtains in the face center coordinate and human testing frame in Face datection frame Number of people center point coordinate matching relationship.

Identification object matching module 62 synthesis submodule 73 will acquire with the Face datection frame of matching relationship and people Body detection block synthesizes the identification target in identification image, carries out identification clarification of objective analysis so as to subsequent.

The identification target that 63 pairs of module of subsequent target's feature-extraction obtains each identification image carries out feature extraction, to obtain Identify at least a kind of target signature of the identification target in image.；The specific all identification targets pair obtained in identification image It is selected in face characteristic class target signature, characteristics of human body's class target signature and the formed group of position class target signature answered At least a kind of target signature.

Specifically, target's feature-extraction module 63 can obtain the corresponding face of all identification targets in each identification image Quality point, clothes attribute value, human face characteristic point, human body feature point and identification target position information.Wherein face quality point with And human face characteristic point is face characteristic class target signature, clothes attribute value and human body feature point are that characteristics of human body's class target is special Sign, identification target position information are position class target signature.In this way images steganalysis device may be selected using human face characteristic point, Human body feature point or identification target position information carry out Fusion Features.

Last Fusion Features module 64 in chronological order, to the mesh of identification target corresponding in multiple identification images of extraction Mark feature is merged.

Specifically, motion profile feature of the Fusion Features module 64 first to corresponding identification target in multiple identification images It initializes.

The identification image that subsequent 64 acquisition time sequence of Fusion Features module is 1, and the identification image for being 1 by time sequencing The target signature of corresponding identification target identifies the motion profile feature of corresponding identification target in image directly as 1.With The target signature for the corresponding identification target of identification image that acquisition time sequence is 1 afterwards, is fused to corresponding in 1 identification image In the motion profile feature for identifying target, to obtain the motion profile feature of corresponding identification target in 2 identification images.…… The target signature for the corresponding identification target of identification image that last acquisition time sequence is m, it is right in m-1 identification image to be fused to In the motion profile feature for the identification target answered, to obtain the motion profile feature of corresponding identification target in m identification image, Wherein m is the quantity for identifying image.

The images steganalysis process and target fortune of the images steganalysis device 60 of the present embodiment are completed in this way Dynamic rail mark acquisition process.

Multiple target signatures of the images steganalysis device of the present embodiment based on identification target generate identification target more Motion profile feature in a identification image；The identification accuracy for identifying target in identification image is improved, is reduced to identification The policer operation cost of target.

Fig. 8 is please referred to, Fig. 8 is the structural schematic diagram of the second embodiment of images steganalysis device of the invention.This reality The second embodiment of above-mentioned images steganalysis method can be used to be implemented for the images steganalysis device for applying example.The image Target Identification Unit 80 include identification module of target detection 81, identification object matching module 82, target's feature-extraction module 83 with And Fusion Features module 84.

Identification module of target detection 81 is used to obtain multiple identification images with relevance, and detects each identification image In Face datection frame and human testing frame；Identify that object matching module 82 is used for the Face datection frame in each identification image Matching operation is carried out with human testing frame, to obtain the identification target in each identification image；Target's feature-extraction module 83 is used In carrying out feature extraction to the identification target in each identification image, to obtain at least a kind of of the identification target in identification image Target signature；Fusion Features module 84 is in chronological order, identifying that the corresponding targets for identifying target are special in images to multiple Sign is merged, to obtain the motion profile feature of corresponding identification target in multiple identification images.

Fig. 9 is please referred to, Fig. 9 is the knot of the Fusion Features module of the second embodiment of images steganalysis device of the invention Structure schematic diagram.This feature Fusion Module 84 includes initialization submodule 91, Fusion Features submodule 92 and counting submodule 93.

Initialization submodule 91 is used to carry out just the motion profile feature of identification target corresponding in multiple identification images Beginningization operation, and n=1 is set；Fusion Features submodule 92 is used to time sequencing be corresponding identification mesh in the identification image of n Target target signature is fused in motion profile feature, to obtain the motion profile of corresponding identification target in n identification image Feature；Counting submodule 93 is used to carry out counting operation to n, until obtaining the fortune of corresponding identification target in m identification image Dynamic track characteristic, wherein m is the quantity for identifying image.

Figure 10 is please referred to, Figure 10 is the Fusion Features module of the second embodiment of images steganalysis device of the invention The structural schematic diagram of Fusion Features submodule.It includes fisrt feature integrated unit 101 and second that this feature, which merges submodule 92, Fusion Features unit 102.

Fisrt feature integrated unit 101 is used for as n=1, the corresponding identification target of identification image for being 1 by time sequencing Target signature, as 1 it is described identification image in it is corresponding identification target motion profile feature；Second feature integrated unit For by the target signature for the corresponding identification target of identification image that time sequencing is n, being fused to n-1 knowledge when n is greater than 1 In other image in the motion profile feature of corresponding identification target, to obtain the fortune of corresponding identification target in n identification image Dynamic track characteristic.

Wherein second feature integrated unit 102 includes the first judgment sub-unit 1021, the second judgment sub-unit 1022, first Similarity obtains subelement 1023, the second similarity obtains subelement 1024, third similarity obtains subelement 1025, matching Unit 1026 and fusion subelement 1027.

For first judgment sub-unit 1021 for judging whether to meet first condition, first condition is the knowledge that time sequencing is n Face quality point in the corresponding target signature for identifying target of other image is greater than the first setting value, and the n-1 identifications figures Face quality point as in the motion profile feature of corresponding identification target is greater than the first setting value.

Second judgment sub-unit 1022 then judges whether to meet second condition, Article 2 for being such as unsatisfactory for first condition Part is the clothes attribute value in the target signature for the corresponding identification target of identification image that time sequencing is n, is schemed with n-1 identification Clothes attribute value as in the motion profile feature of corresponding identification target is greater than the second setting value.

First similarity obtains subelement 1023 for such as meeting first condition, then the identification figure for being n according to time sequencing As the fortune of corresponding identification target in the human face characteristic point and n-1 identification image in the corresponding target signature for identifying target Human face characteristic point in dynamic track characteristic determines the corresponding identification target of identification image and n-1 identification figure that time sequencing is n The similarity of corresponding identification target as in.

Second similarity obtains subelement 1024 for such as meeting second condition, then the identification figure for being n according to time sequencing As the fortune of corresponding identification target in the human body feature point and n-1 identification image in the corresponding target signature for identifying target Human body feature point in dynamic track characteristic determines the corresponding identification target of identification image and n-1 identification figure that time sequencing is n The similarity of corresponding identification target as in.

Third similarity obtains subelement 1025 for being such as unsatisfactory for second condition, then the identification for being n according to time sequencing Corresponding identification in identification target position information and n-1 identification image in the target signature of the corresponding identification target of image Identification target position information in the motion profile feature of target determines that time sequencing is the corresponding identification mesh of identification image of n Mark the similarity that corresponding identification target in image is identified with n-1.

Coupling subelement 1026 is used to use bipartite graph matching algorithm, corresponding according to the identification image that time sequencing is n Identify that target identifies the similarity of corresponding identification target in image with n-1, the identification image that acquisition time sequence is n is corresponding Identification target with n-1 identify image in it is corresponding identification target matching relationship.

The target signature for merging subelement 1027 to be used for the corresponding identification target of identification image for being n by time sequencing, melts It is bonded in n-1 identification image in the motion profile feature of the identification target with matching relationship, to obtain in n identification image The motion profile feature of corresponding identification target.

The images steganalysis device 80 of the present embodiment is in use, identification module of target detection 81 first is obtained with association Property multiple identification images, and detect it is each identification image in Face datection frame and human testing frame.

Then identify object matching module 82 by each Face datection frame identified in image and the progress of human testing frame With operation, to obtain the identification target in each identification image.

Then 83 pairs of the target's feature-extraction module identification targets for obtaining each identification image carry out feature extraction, to obtain Identify at least a kind of target signature of the identification target in image.

Specifically, target's feature-extraction module 83 can obtain the corresponding face of all identification targets in each identification image Quality point, clothes attribute value, human face characteristic point, human body feature point and identification target position information.Wherein face quality point with And human face characteristic point is face characteristic class target signature, clothes attribute value and human body feature point are that characteristics of human body's class target is special Sign, identification target position information are position class target signature.In this way images steganalysis device may be selected using human face characteristic point, Human body feature point or identification target position information carry out Fusion Features.

Last Fusion Features module 84 in chronological order, to the mesh of identification target corresponding in multiple identification images of extraction It marks feature and carries out mixing operation.

It specifically includes:

Movement rail of the initialization submodule 91 of Fusion Features module 84 to corresponding identification target in multiple identification images Mark feature initializes, and n=1 is arranged, and wherein n is used for the counting operation of Fusion Features.

Time sequencing is corresponding identification mesh in the identification image of n by the Fusion Features submodule 92 of Fusion Features module 84 Target target signature is fused in motion profile feature, to obtain the motion profile of corresponding identification target in n identification image Feature.

As n=1, the fisrt feature integrated unit 101 of Fusion Features submodule 92 directly by time sequencing be 1 identification The target signature of the corresponding identification target of image identifies the motion profile feature of corresponding identification target in image as 1.It should Motion profile feature includes but is not limited to the human face characteristic point for identifying target, human body feature point and identification target position information.

When n is greater than 1, time sequencing is the identification figure of n by the second feature integrated unit 102 of Fusion Features submodule 92 As the target signature of corresponding identification target, it is fused to the motion profile feature of corresponding identification target in n-1 identification image In, to obtain the motion profile feature of corresponding identification target in n identification image.

The counting submodule 93 of Fusion Features module 84 carries out counting operation, i.e. n=n+1, and backout feature fusant to n Module carries out the operation of motion profile Fusion Features, until the motion profile for obtaining corresponding identification target in m identification image is special Sign, wherein m is the quantity for identifying image.

Wherein the process of the progress of second feature integrated unit 102 motion profile Fusion Features operation includes:

First judgment sub-unit 1021 of second feature integrated unit 102 judges whether to meet first condition, this first Part is that the face quality point in the target signature for the corresponding identification target of identification image that time sequencing is n is greater than the first setting Value, and the face quality point in n-1 identification image in the motion profile feature of corresponding identification target is greater than the first setting value.

It is such as unsatisfactory for first condition, then the second judgment sub-unit 1022 of second feature integrated unit 102 judges whether full Sufficient second condition, the second condition are the clothes in the target signature for the corresponding identification target of identification image that time sequencing is n Attribute value, the clothes attribute value identified in image in the motion profile feature of corresponding identification target with n-1 are greater than second and set Definite value.

Such as meet first condition, i.e. time sequencing is the people in the target signature for identifying the corresponding identification target of image of n Face quality point is greater than the first setting value, and the face in n-1 identification image in the motion profile feature of corresponding identification target Quality point is greater than the first setting value.Then explanation can be calculated in different identification images by human face characteristic point in identification target and identify mesh Target similarity.

Specifically, it is n's that the first similarity of second feature integrated unit 102, which obtains subelement 1023 according to time sequencing, Identify corresponding identification mesh in the human face characteristic point and n-1 identification image in the target signature of the corresponding identification target of image Human face characteristic point in target motion profile feature determines the corresponding identification target of identification image and n-1 that time sequencing is n Identify the similarity of corresponding identification target in image.

As time sequencing be n identification image it is corresponding identification target target signature in human face characteristic point be Fa, n-1 Human face characteristic point in a identification image in the motion profile feature of corresponding identification target is Fb, and it is n that wherein a, which is time sequencing, Identification image in identify target label, b be n-1 identification image in identify target label, then time sequencing for n knowledge The similarity S1 of identification target and the identification target in n-1 identification image in other image are as follows: S1=(Fa*Fb)/(| Fa | * | Fb|)。

Such as be unsatisfactory for first condition, meet second condition, i.e., time sequencing be n the corresponding identification target of identification image Clothes attribute value in target signature, the clothes identified in image in the motion profile feature of corresponding identification target with n-1 Attribute value is greater than the second setting value.Then explanation can be calculated in different identification images by human body feature point in identification target and identify mesh Target similarity.

Specifically, it is n's that the second similarity of second feature integrated unit 102, which obtains subelement 1024 according to time sequencing, Identify corresponding identification mesh in the human body feature point and n-1 identification image in the target signature of the corresponding identification target of image Human body feature point in target motion profile feature determines the corresponding identification target of identification image and n-1 that time sequencing is n Identify the similarity of corresponding identification target in image.

As time sequencing be n identification image it is corresponding identification target target signature in human body feature point be Ga, n-1 Human body feature point in a identification image in the motion profile feature of corresponding identification target is Gb, and it is n that wherein a, which is time sequencing, Identification image in identify target label, b be n-1 identification image in identify target label, then time sequencing for n knowledge The corresponding similarity S2 for identifying target in the other corresponding identification target of image and n-1 identification image are as follows: S2=(Ga*Gb)/ (|Ga|*|Gb|)。

If being unsatisfactory for first condition and second condition simultaneously, then illustrates to identify that the face quality point of target is poor, be not suitable for The similarity that target is identified in different identification images is calculated by the human face characteristic point in identification target；The people of target is identified simultaneously The difference of body characteristics point is smaller, is also not suitable for by identifying mesh in the different identification images of human body feature point calculating in identification target Target similarity.At this moment it can only be calculated in different identification images by the identification target position information in identification target and identify target Similarity.

Specifically, it is n's that the third similarity of second feature integrated unit 102, which obtains subelement 1025 according to time sequencing, Identify corresponding identification mesh in the human body feature point and n-1 identification image in the target signature of the corresponding identification target of image Human body feature point in target motion profile feature determines the corresponding identification target of identification image and n-1 that time sequencing is n Identify the similarity of corresponding identification target in image.

As time sequencing be n identification image it is corresponding identification target target signature in identification target position information be Identification target position information in Ha, n-1 identification images in the motion profile feature of corresponding identification target is Hb, wherein a For time sequencing be n identification image in identify target label, b be n-1 identification image in identification target label, then when Between sequentially for n identification image it is corresponding identification target with n-1 identification image in it is corresponding identify target similarity S3 are as follows: S3=iou (Ha, Hb)；Wherein iou function is to hand over and than function (Intersection-over-Union), for calculating Ha pairs The corresponding feature frame of the identification target position information answered (here can be corresponding human testing frame) identification target corresponding with Hb The degree of overlapping of the corresponding feature frame of location information.

The corresponding identification target of identification image and the corresponding identification mesh in n-1 identification image that acquisition time sequence is n After target similarity, the coupling subelement 1026 of second feature integrated unit 102 uses bipartite graph matching algorithm, suitable according to the time The corresponding identification target of identification image that sequence is n identifies the similarity of corresponding identification target in image, acquisition time with n-1 The corresponding identification target of identification image that sequence is n identifies the matching relationship of corresponding identification target in image with n-1.

Specifically can acquisition time sequence be n the corresponding multiple identification targets of identification image and n-1 identification image in Corresponding multiple identification targets, and obtain the corresponding any identification target of identification image and n-1 knowledge that above-mentioned time sequencing is n Similarity in other image between corresponding any identification target；Bipartite graph matching algorithm then can be used, after calculating matching The corresponding identification target of the identification image that time sequencing is n is same identification with corresponding identification target in n-1 identification image The maximum value of the similarity sum of target.In this way can acquisition time sequence be n the corresponding identification target of identification image and n-1 it is a Identify the matching relationship of corresponding identification target in image.

The corresponding identification of identification image that time sequencing is n by the fusion subelement 1027 of second feature integrated unit 102 The target signature of target is fused in n-1 identification image in the motion profile feature of the identification target with matching relationship, with It obtains in n identification image to the motion profile feature of the identification target of drink.

The images steganalysis process and target fortune of the images steganalysis device 80 of the present embodiment are completed in this way Dynamic rail mark acquisition process.

On the basis of first embodiment, the images steganalysis device of the present embodiment is according to different target characteristic attribute It identifies target, generates the similarity for identifying target in different identification images in different ways, improve and obtain different identifications The accuracy of corresponding identification target, to improve the accuracy of Fusion Features operation, and then further improves knowledge in image The identification accuracy of target is identified in other image.

Illustrate images steganalysis method and images steganalysis device of the invention below by a specific embodiment Working principle.Please refer to Figure 11 a to Figure 11 c and Figure 12, Figure 11 a to Figure 11 c be images steganalysis method of the invention and The schematic diagram for three identification images that images steganalysis device obtains.Figure 12 is images steganalysis method and figure of the invention As the images steganalysis flow chart of Target Identification Unit.

Monitoring image identification server of the invention carries out target identification operation and identification to above-mentioned identification image automatically Target trajectory identification operation.The target identification process includes:

Step S1201, monitoring image identify that server obtains Face datection frame and human testing frame in identification image, point Wei not obtain Face datection frame 11a1 in Figure 11 a, Face datection frame 11a2, Face datection frame 11a3, human testing frame 11a4, Human testing frame 11a5, human testing frame 11a6；Obtain Face datection frame 11b1, the Face datection frame 11b2, face in Figure 11 b Detection block 11b3, human testing frame 11b4, human testing frame 11b5, human testing frame 11b6；Obtain the Face datection in Figure 11 c Frame 11c1, Face datection frame 11c2, Face datection frame 11c3, human testing frame 11c4, human testing frame 11c5, human testing frame 11c6。

Step S1202, monitoring image identify server using bipartite graph matching algorithm to the Face datection in identification image Frame and human testing frame carry out matching operation, and the Face datection frame 11a1 and human testing frame 11a4 in final Figure 11 a are synthesized Identify the identification target 111 in image 11a, Face datection frame 11a2 and human testing frame 11a5 are synthesized in identification image 11a Identification target 112, Face datection frame 11a3 and human testing frame 11a4 synthesize the identification target 113 in identification image 11a； Similarly the identification target 121 in synthesis identification image 11b, identification target 122 and identification target 123；It identifies in image 11c Identify target 131, identification target 132 and identification target 133.

Step S1203, monitoring image identify that server carries out the identification target in identification image 11a, 11b and 11c Feature extraction, obtain respectively face quality point, clothes attribute value, human face characteristic point, the human body feature point of all identification targets with And the target signatures such as identification target position information.

Step S1204, since identification image 11a is first frame picture, monitoring image identifies server for identification image Identification target in 11a carries out motion profile feature initialization operation, that is, is initialized 3 tracks, and every track includes one Identify target.

It will then identify the target signature of the corresponding identification target of image 11a as corresponding identification in identification image 11a The motion profile feature of target.It will identify identification target 111 corresponding movement rail of the feature of target 111 as identification image Mark feature will identify that the feature in target 112 as the corresponding motion profile feature of identification target 112 of identification image, will be known Identification target 113 corresponding motion profile feature of the feature as identification image in other target 113.

Step S1205, monitoring image identify that server arrives the target feature fusion of the identification target identified in image 11b In the motion profile feature for identifying the identification target in image 11a.

Monitoring image identifies that server extracts the target signature that target is identified in identification image 11b, then passes through following public affairs Formula calculates the similarity of identification target and identification target corresponding in identification image 11a in identification image 11b:

Wherein F1 is to identify the human face characteristic point that target is identified in image 11a, and F2 is the identification target identified in image 11b Human face characteristic point, P1 is the face quality point for identifying the identification target in image 11a, and P2 is the identification in identification image 11b The face quality of target point, Pmin are the first setting value；G1 is the human body feature point for identifying the identification target in image 11a, G2 For the human body feature point of the identification target in identification image 11b, r1 is the clothes attribute for identifying the identification target in image 11a Value, r2 are the clothes attribute value for identifying the identification target in image 11b, and rset is the second setting value, and H1 is in identification image 11a Identification target identification target position information, H2 be identify image 11b in identification target identification target position information.

The meaning of above-mentioned formula is that the priority of first row calculating formula of similarity is greater than secondary series calculating formula of similarity Priority, while the priority of secondary series calculating formula of similarity be greater than third column calculating formula of similarity priority.

The phase of the corresponding identification target of identification image 11b identification target corresponding with image 11a is identified can be obtained in this way Like degree, subsequent monitoring image identification server uses bipartite graph matching algorithm, obtain the corresponding identification target of identification image 11b and Identify the matching relationship of the corresponding identification target of image 11a.Identify the identification target 111 and identification image 11b in image 11a In identification target 121 match, identify image 11a in identification target 112 with identify image 11b in identification target 122 Match, identifies that the identification target 113 in image 11a is matched with the identification target 123 in identification image 11b.

The target signature that will then identify the identification target in image 11b, being fused in identification image 11a, there is matching to close In the motion profile feature of the identification target of system, i.e., it will identify the face quality in the identification target 121 in image 11b point fusion Into the face quality point of the identification target 111 in identification image 11a, thus obtain identification target 111 identification image 11a with Identify the face quality point of the motion profile feature in image 11b, specific amalgamation mode can be to take in identification target 121 here Face quality point and identification target 111 face quality point in maximum value or take face quality point in identification target 121 With the average value of the face quality point of identification target 111.Identification target 111 can be successively obtained in this way in identification image 11a and knowledge Clothes attribute value, human face characteristic point and the human body feature point of the motion profile feature of other image 11b.

The identification target position information of the motion profile feature of fused identification target 111 is changed to identification target 111 It identifies target position information and identifies the set of the identification target position information of target 121, to be subsequently generated identification target 111 Motion profile.

This generates the motion profile features of identification image 11a and the identification target for identifying image 11b.

Step S1206, monitoring image identify that server arrives the target feature fusion of the identification target identified in image 11c In the motion profile feature for identifying the identification target of image 11a and identification image 11b.

Monitoring image identifies that server extracts the target signature that target is identified in identification image 11c, then passes through following public affairs Formula calculates the similarity of identification target and identification target corresponding in identification image 11a, 11b in identification image 11c:

Wherein F3 is to identify the human face characteristic point that target is identified in image 11c, and F12 is the knowledge identified in image 11a, 11b The human face characteristic point of other target, P3 are the face quality point for identifying the identification target in image 11c, P12 be identify image 11a, The face quality of identification target in 11b point, Pmin are the first setting value；G3 is the people for identifying the identification target in image 11c Body characteristics point, G12 are the human body feature point for identifying the identification target in image 11a, 11b, and r3 is the identification identified in image 11c The clothes attribute value of target, r12 are the clothes attribute value for identifying the identification target in image 11a, 11b, and rset is the second setting Value, H3 are the identification target position information for identifying the identification target in image 11c, and H2 is the identification target identified in image 11b Identification target position information.

The corresponding identification target of identification image 11c identification target corresponding with identification image 11a, 11b can be obtained in this way Similarity, subsequent monitoring image identification server uses bipartite graph matching algorithm, obtains the corresponding identification mesh of identification image 11c Mark the matching relationship of identification target corresponding with identification image 11a, 11b.It identifies the identification target 131 in image 11c and knows Identification target 111 in other image 11a matches, and identifies the identification in the identification target 132 and identification image 11a in image 11c Target 112 matches, and identifies that the identification target 133 in image 11c is matched with the identification target 113 in identification image 11a.

Then will identify image 11c in identification target target signature, be fused to identification image 11a, 11b in In the motion profile feature of identification target with relationship, i.e., it will identify the face quality in the identification target 131 in image 11c point In the face quality point of the identification target 111 and identification target 121 that are fused in identification image 11a, 11b, to be known Other target 111 is in identification image 11a, identification image 11b and the face quality for identifying the motion profile feature in image 11c Point, here specific amalgamation mode can be take face quality point in identification target 131, the face quality point of identification target 111, It identifies the maximum value in the face quality point of target 121 or the face quality in identification target 131 is taken to divide, identify target 111 The average value of face quality point, the face quality point of identification target 121.Identification target 111 can be successively obtained in this way in identification figure As 11a, identification image 11b and clothes attribute value, human face characteristic point and the human body of the motion profile feature for identifying image 11c Characteristic point.

The identification target position information of the motion profile feature of fused identification target 111 is changed to identification target 111 Identify the identification target position information of target position information, the identification target position information of identification target 121 and identification target 131 Set, so as to be subsequently generated identification target 111 motion profile.As identified in Figure 11 a, the motion profile of target 112 is A, figure The motion profile that target 113 is identified in 11a is B etc..

So i.e. complete this specific embodiment as target identification method and images steganalysis device target identification Operation and identification target trajectory identification operation.

The multiple target signatures of images steganalysis method, apparatus and storage medium of the invention based on identification target, it is raw At motion profile feature of the identification target in multiple identification images；It improves in identification image and identifies that the identification of target is accurate Property, reduce the policer operation cost to identification target；Effective solution existing images steganalysis method and image mesh Mark the technical issues of error rate is higher or policer operation higher cost of identification device.

" component ", " module ", " system ", " interface ", " process " etc. are generally intended to as used herein the term Refer to computer related entity: hardware, the combination of hardware and software, software or software in execution.For example, component can be but not It is limited to be the process on a processor of running, processor, object, executable application, thread, program and/or the computer executed. By diagram, both the application and the controller run on the controller can be component.One or more components can have It is in the process executed and/or thread, and component can be located on a computer and/or be distributed in two or more meters Between calculation machine.

Figure 13 and the discussion below, which provide, sets the electronics where realizing images steganalysis device of the present invention Brief, summary the description of standby working environment.The working environment of Figure 13 be only an example of working environment appropriate simultaneously And suggestion is not intended to about the purposes of working environment or any restrictions of the range of function.Example electronic equipment 1312 includes but not It is limited to wearable device, helmet, medical treatment & health platform, personal computer, server computer, hand-held or on knee sets Standby, mobile device (such as mobile phone, personal digital assistant (PDA), media player etc.), multicomputer system, consumption Type electronic equipment, minicomputer, mainframe computer, distributed computing environment including above-mentioned arbitrary system or equipment, etc..

Although not requiring, in the common background that " computer-readable instruction " is executed by one or more electronic equipments Lower description embodiment.Computer-readable instruction can be distributed via computer-readable medium and (be discussed below).It is computer-readable Instruction can be implemented as program module, for example executes particular task or realize the function of particular abstract data type, object, application Programming interface (API), data structure etc..Typically, the function of the computer-readable instruction can be in various environment arbitrarily Combination or distribution.

Figure 13 illustrates the electronic equipment including one or more embodiments in images steganalysis device of the invention 1312 example.In one configuration, electronic equipment 1312 includes at least one processing unit 1316 and memory 1318.According to The exact configuration and type of electronic equipment, memory 1318 can be volatibility (such as RAM), it is non-volatile (such as ROM, flash memory etc.) or both certain combination.The configuration is illustrated in Figure 13 by dotted line 1314.

In other embodiments, electronic equipment 1312 may include supplementary features and/or function.For example, equipment 1312 is also It may include additional storage device (such as removable and/or non-removable) comprising but it is not limited to magnetic memory apparatus, light Storage device etc..This additional memory devices are illustrated in Figure 13 by storage device 1320.In one embodiment, for real The computer-readable instruction of existing one or more embodiments provided in this article can be in storage device 1320.Storage device 1320 other computer-readable instructions that can also be stored for realizing operating system, application program etc..Computer-readable instruction It can be loaded into memory 1318 and be executed by such as processing unit 1316.

Term as used herein " computer-readable medium " includes computer storage medium.Computer storage medium includes The volatibility that any method or technique of the information of such as computer-readable instruction or other data etc is realized for storage With non-volatile, removable and nonremovable medium.Memory 1318 and storage device 1320 are the realities of computer storage medium Example.Computer storage medium includes but is not limited to RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, number Universal disc (DVD) or other light storage devices, cassette tape, tape, disk storage device or other magnetic storage apparatus can be with Any other medium for storing expectation information and can be accessed by electronic equipment 1312.Any such computer storage is situated between Matter can be a part of electronic equipment 1312.

Electronic equipment 1312 can also include the communication connection 1326 for allowing electronic equipment 1312 to communicate with other equipment.It is logical Letter connection 1326 can include but is not limited to modem, network interface card (NIC), integrated network interface, radiofrequency launcher/ Receiver, infrared port, USB connection or other interfaces for electronic equipment 1312 to be connected to other electronic equipments.Communication Connection 1326 may include wired connection or wireless connection.Communication connection 1326 can emit and/or receive communication medium.

Term " computer-readable medium " may include communication media.Communication media typically comprises computer-readable instruction Or other data in " the own modulated data signal " of such as carrier wave or other transmission mechanisms etc, and passed including any information Send medium.Term " own modulated data signal " may include such signal: one or more of the characteristics of signals is according to general Mode of the information coding into signal is set or changed.

Electronic equipment 1312 may include input equipment 1324, for example, keyboard, mouse, pen, voice-input device, touch it is defeated Enter equipment, infrared camera, video input apparatus and/or any other input equipment.It also may include that output is set in equipment 1312 Standby 1322, such as one or more displays, loudspeaker, printer and/or other any output equipments.1324 He of input equipment Output equipment 1322 can be connected to electronic equipment 1312 via wired connection, wireless connection or any combination thereof.In a reality It applies in example, input equipment or output equipment from another electronic equipment are used as the input equipment of electronic equipment 1312 1324 or output equipment 1322.

The component of electronic equipment 1312 can be connected by various interconnection (such as bus).Such interconnection may include outer Enclose component interconnection (PCI) (such as quick PCI), universal serial bus (USB), firewire (IEEE 1394), optical bus structure etc. Deng.In another embodiment, the component of electronic equipment 1312 can pass through network interconnection.For example, memory 1318 can be by Multiple physical memory cells arcs composition in different physical locations, by network interconnection.

It would be recognized by those skilled in the art that the storage equipment for storing computer-readable instruction can be across network point Cloth.For example, can store via the electronic equipment 1330 that network 1328 accesses for realizing one provided by the present invention or The computer-readable instruction of multiple embodiments.The accessible electronic equipment 1330 of electronic equipment 1312 and downloading computer is readable What is instructed is part or all of for execution.Alternatively, electronic equipment 1312 can be downloaded a plurality of computer-readable as needed It instructs or some instruction can execute at electronic equipment 1312 and some instructions can be held at electronic equipment 1330 Row.

There is provided herein the various operations of embodiment.In one embodiment, one or more operations can be with structure At the computer-readable instruction stored on one or more computer-readable mediums, will make to succeed in one's scheme when being executed by electronic equipment It calculates equipment and executes the operation.Describing the sequences of some or all of operations, to should not be construed as to imply that these operations necessarily suitable Sequence is relevant.It will be appreciated by those skilled in the art that the alternative sequence of the benefit with this specification.Furthermore, it is to be understood that Not all operation must exist in each embodiment provided in this article.

Moreover, although the disclosure, this field skill has shown and described relative to one or more implementations Art personnel will be appreciated that equivalent variations and modification based on the reading and understanding to the specification and drawings.The disclosure include it is all this The modifications and variations of sample, and be limited only by the scope of the following claims.In particular, to by said modules (such as element, Resource etc.) the various functions that execute, term for describing such components is intended to correspond to the specified function for executing the component The random component (unless otherwise instructed) of energy (such as it is functionally of equal value), even if illustrated herein with execution in structure The disclosure exemplary implementations in function open structure it is not equivalent.In addition, although the special characteristic of the disclosure Through being disclosed relative to the only one in several implementations, but this feature can with such as can be to given or specific application For be expectation and one or more other features combinations of other advantageous implementations.Moreover, with regard to term " includes ", " tool Have ", " containing " or its deformation be used in specific embodiments or claims for, such term be intended to with term The similar mode of "comprising" includes.

Each functional unit in the embodiment of the present invention can integrate in a processing module, be also possible to each unit list It is solely physically present, can also be integrated in two or more units in a module.Above-mentioned integrated module can both use Formal implementation of hardware can also be realized in the form of software function module.If the integrated module is with software function The form of module is realized and when sold or used as an independent product, also can store in computer-readable storage Jie In matter.Storage medium mentioned above can be read-only memory, disk or CD etc..Above-mentioned each device or system, can be with Execute the method in correlation method embodiment.

Although the serial number before embodiment only makes for convenience of description in conclusion the present invention is disclosed above with embodiment With not causing to limit to the sequence of various embodiments of the present invention.Also, above-described embodiment is not intended to limit the invention, this field Those of ordinary skill, without departing from the spirit and scope of the present invention, can make it is various change and retouch, therefore it is of the invention Protection scope subjects to the scope of the claims.

Claims

1. a kind of images steganalysis method characterized by comprising

Multiple identification images with relevance are obtained, and detect the Face datection frame in each identification image and human body inspection Survey frame；

Each Face datection frame identified in image and human testing frame are matched, to obtain each identification figure Identification target as in；

Feature extraction is carried out to the identification target in each identification image, to obtain the identification target in the identification image At least a kind of target signature；And

In chronological order, at least a kind of target signature of corresponding identification target in multiple identification images is merged, To obtain the motion profile feature of corresponding identification target in multiple identification images.

2. images steganalysis method according to claim 1, which is characterized in that described obtain has the multiple of relevance The step of identifying image are as follows:

The multiple video frames for obtaining same video file, as multiple identification images with relevance.

3. images steganalysis method according to claim 1, which is characterized in that it is described will be in each identification image Face datection frame and human testing frame carry out matching operation, with obtain it is each it is described identification image in identification target the step of Include:

It obtains in the number of people in the face center coordinate and the human testing frame in the Face datection frame in identification image Heart point coordinate；

Using bipartite graph matching algorithm, according to the distance relation of the face center coordinate and the number of people center point coordinate, Obtain the matching of the number of people center point coordinate in the face center coordinate and the human testing frame in the Face datection frame Relationship；

By with matching relationship Face datection frame and human testing frame synthesize it is described identification image in identification target.

4. images steganalysis method according to claim 1, which is characterized in that described in each identification image Identification target carry out feature extraction, with obtain it is described identification image in identification target at least one kind target signature the step of Include:

Obtain each corresponding face characteristic class target signature of all identification targets identified in image, characteristics of human body's classification At least a kind of target signature selected in mark feature and the formed group of position class target signature；Wherein face characteristic class target Feature includes face quality point and human face characteristic point, characteristics of human body's class target signature include clothes attribute value and characteristics of human body Point, position class target signature include identification target position information；

It is described that the target signature of corresponding identification target in multiple identification images is merged in chronological order, with Into multiple identification images, the step of motion profile feature of corresponding identification target, includes:

The motion profile feature of corresponding identification target in multiple identification images is initialized, and n=is set 1；

The target signature of corresponding identification target, is fused to the motion profile feature in the identification image for being n by time sequencing In, to obtain the motion profile feature of corresponding identification target in the n identification images；

N=n+1, return movement track characteristic fusion steps, until obtaining the movement of corresponding identification target in m identification image Track characteristic, wherein m is the quantity for identifying image.

5. images steganalysis method according to claim 4, which is characterized in that the identification for being n by time sequencing The target signature of corresponding identification target, is fused in the motion profile feature in image, to obtain the n identification images In it is corresponding identification target motion profile feature the step of include:

As n=1, by the target signature for the corresponding identification target of identification image that time sequencing is 1, as 1 identification The motion profile feature of corresponding identification target in image；

When n is greater than 1, by the target signature for the corresponding identification target of identification image that time sequencing is n, it is fused to n-1 institute It states in identification image in the motion profile feature of corresponding identification target, to obtain corresponding identification in the n identification images The motion profile feature of target.

6. images steganalysis method according to claim 5, which is characterized in that the identification for being n by time sequencing The target signature of the corresponding identification target of image is fused to the movement rail of corresponding identification target in the n-1 identification images In mark feature, to include: the step of obtaining the motion profile feature of corresponding identification target in the n identification images

Judge whether to meet first condition, the first condition is the corresponding identification mesh of identification image that the time sequencing is n Face quality point in target target signature is greater than the first setting value, and corresponding identification target in the n-1 identification image Motion profile feature in face quality point be greater than the first setting value；

Such as meet the first condition, then the target for the corresponding identification target of identification image for being n according to the time sequencing is special Face in human face characteristic point and the n-1 identification image in sign in the motion profile feature of corresponding identification target is special Point is levied, determines that the time sequencing is the corresponding identification target of identification image and the corresponding knowledge in the n-1 identification image of n The similarity of other target；

Using bipartite graph matching algorithm, according to the corresponding identification target of identification image that the time sequencing is n and the n-1 The similarity for identifying corresponding identification target in image, obtains the corresponding identification target of identification image that the time sequencing is n The matching relationship of corresponding identification target in image is identified with described n-1；

The target signature for the corresponding identification target of identification image that time sequencing is n is fused in the n-1 identification images In the motion profile feature of identification target with matching relationship, to obtain corresponding identification target in the n identification images Motion profile feature.

7. images steganalysis method according to claim 6, which is characterized in that the identification for being n by time sequencing The target signature of the corresponding identification target of image is fused to the movement rail of corresponding identification target in the n-1 identification images In mark feature, to obtain n described the step of identifying the motion profile feature of corresponding identification target in images further include:

It is such as unsatisfactory for the first condition, then judges whether to meet second condition, the second condition is that the time sequencing is n Identification image it is corresponding identification target target signature in clothes attribute value, it is corresponding with the n-1 identification image Identify that the clothes attribute value in the motion profile feature of target is greater than the second setting value；

Such as meet the second condition, then the target for the corresponding identification target of identification image for being n according to the time sequencing is special Human body in human body feature point and the n-1 identification image in sign in the motion profile feature of corresponding identification target is special Point is levied, determines that the time sequencing is the corresponding identification target of identification image and the corresponding knowledge in the n-1 identification image of n The similarity of other target；

8. images steganalysis method according to claim 7, which is characterized in that the identification for being n by time sequencing The target signature of the corresponding identification target of image is fused to the movement rail of corresponding identification target in the n-1 identification images In mark feature, to obtain n described the step of identifying the motion profile feature of corresponding identification target in images further include:

It is such as unsatisfactory for the second condition, then the target for the corresponding identification target of identification image for being n according to the time sequencing In identification target position information and the n-1 identification image in feature in the motion profile feature of corresponding identification target Identification target position information, determine the time sequencing be n the corresponding identification target of identification image with described n-1 identify The similarity of corresponding identification target in image；

9. a kind of images steganalysis device characterized by comprising

It identifies module of target detection, for obtaining multiple identification images with relevance, and detects each identification image In Face datection frame and human testing frame；

Object matching module is identified, for matching each Face datection frame identified in image and human testing frame Operation, to obtain the identification target in each identification image；

Target's feature-extraction module, for carrying out feature extraction to the identification target in each identification image, to obtain State at least a kind of target signature of the identification target in identification image；And

Fusion Features module, in chronological order, identifying at least a kind of of target to corresponding in multiple identification images Target signature is merged, to obtain the motion profile feature of corresponding identification target in multiple identification images.

10. images steganalysis device according to claim 9, which is characterized in that the identification object matching module packet It includes:

Coordinate acquisition submodule, for obtaining face center coordinate and the people in the Face datection frame in identification image Number of people center point coordinate in body detection block；

Coordinate matching submodule, for using bipartite graph matching algorithm, according in the face center coordinate and the number of people The distance relation of heart point coordinate obtains the people in the face center coordinate and the human testing frame in the Face datection frame The matching relationship of head center point coordinate；And

Submodule is synthesized, for synthesizing the Face datection frame with matching relationship and human testing frame in the identification image Identification target.

11. images steganalysis device according to claim 9, which is characterized in that

The target's feature-extraction module, it is special for obtaining the corresponding face of all identification targets in each identification image At least classification selected in sign class target signature, characteristics of human body's class target signature and the formed group of position class target signature Mark feature；Wherein face characteristic class target signature includes face quality point and human face characteristic point, characteristics of human body's class target signature Including clothes attribute value and human body feature point, position class target signature includes identification target position information；

The Fusion Features module, comprising:

Initialization submodule carries out initial for the motion profile feature to corresponding identification target in multiple identification images Change operation, and n=1 is set；

Fusion Features submodule, for the target signature for identifying corresponding identification target in image for being n by time sequencing, fusion Extremely in the motion profile feature, to obtain the motion profile features of corresponding identification target in the n identification images；And

Counting submodule, for carrying out counting operation to n, until obtaining the movement of corresponding identification target in m identification image Track characteristic, wherein m is the quantity for identifying image.

12. images steganalysis device according to claim 11, which is characterized in that

The Fusion Features submodule includes:

Fisrt feature integrated unit is used for as n=1, by the target for the corresponding identification target of identification image that time sequencing is 1 Feature, the motion profile feature as corresponding identification target in 1 identification image；

Second feature integrated unit, for being the mesh of the corresponding identification target of identification image of n by time sequencing when n is greater than 1 Feature is marked, is fused in the n-1 identification images in the motion profile feature of corresponding identification target, it is described to obtain n Identify the motion profile feature of corresponding identification target in image.

13. images steganalysis device according to claim 12, which is characterized in that the second feature integrated unit packet It includes:

First judgment sub-unit meets first condition for judging whether, the first condition is the knowledge that the time sequencing is n Face quality point in the corresponding target signature for identifying target of other image is greater than the first setting value, and the n-1 identification is schemed Face quality point as in the motion profile feature of corresponding identification target is greater than the first setting value；

Second judgment sub-unit then judges whether to meet second condition, the Article 2 for being such as unsatisfactory for the first condition Part is the clothes attribute value in the target signature for the corresponding identification target of identification image that the time sequencing is n, with the n-1 Clothes attribute value in a identification image in the motion profile feature of corresponding identification target is greater than the second setting value；

First similarity obtains subelement, for the first condition as described in meeting, then the identification figure for being n according to the time sequencing As corresponding identification target in the human face characteristic point and the n-1 identification image in the corresponding target signature for identifying target Motion profile feature in human face characteristic point, determine the time sequencing for the corresponding identification target of identification image of n and institute State the similarity of corresponding identification target in n-1 identification image；

Second similarity obtains subelement, for the second condition as described in meeting, then the identification figure for being n according to the time sequencing As corresponding identification target in the human body feature point and the n-1 identification image in the corresponding target signature for identifying target Motion profile feature in human body feature point, determine the time sequencing for the corresponding identification target of identification image of n and institute State the similarity of corresponding identification target in n-1 identification image；

Third similarity obtains subelement, for being such as unsatisfactory for the second condition, then the identification for being n according to the time sequencing It is corresponding in identification target position information and the n-1 identification image in the target signature of the corresponding identification target of image It identifies the identification target position information in the motion profile feature of target, determines that the identification image that the time sequencing is n is corresponding Identification target with described n-1 identify image in it is corresponding identification target similarity；

Coupling subelement, for using bipartite graph matching algorithm, the corresponding identification of identification image for being n according to the time sequencing Target identifies the similarity of corresponding identification target in image with described n-1, obtains the identification image that the time sequencing is n Corresponding identification target identifies the matching relationship of corresponding identification target in image with described n-1；

Subelement is merged, for the target signature for the corresponding identification target of identification image for being n by time sequencing, is fused to n-1 In a identification image in the motion profile feature of the identification target with matching relationship, to obtain the n identification images In it is corresponding identification target motion profile feature.

14. a kind of storage medium is stored with processor-executable instruction, described instruction is by one or more processors Load, to execute such as images steganalysis method any in claim 1 to 8.

15. a kind of electronic equipment, including processor and memory, the memory storage has computer program, the processor By calling the computer program, for executing such as images steganalysis method any in claim 1 to 8.