Full-automatic video identification authentication system and method
Technical field
The present invention relates to a kind of identity authorization system and method, a kind of specifically full-automatic video identification authentication system and method.
Background technology
Along with Internet development, there is increasing application scenario to need the identity of authenticated user.A common example is exactly all kinds of social network sites, need judge in these websites whether the photo that the user uploads is my photo.Relatively Chang Yong a kind of authentication mode is a video authentication, usually computer system can require the user at first to upload the live photo of oneself, and then take some video photo upload with camera, contrast living photo and video photograph on the backstage, website by the customer service personnel, thereby judge whether video takes is user's picture.This video identification authentication method has significant disadvantages, and promptly it depends on artificial judgement, and workload is bigger, thereby cost is than higher.Simultaneously, the cycle of artificial judgement is long, and authentication speed is slow, and generally the user can't obtain authentication result in real time.In addition, traditional mode of utilizing customer service personnel manual identified need be built corresponding customer service system, complex structure, and the system building cost is very high.
Summary of the invention
The objective of the invention is to propose a kind of the needs manually adjudicates, and can reduce labor workload greatly, in real time access authentication result's full-automatic video identification authentication system and method.
Purpose of the present invention can be achieved through the following technical solutions:
Full-automatic video identification authentication system comprises a line module, is connected with computer with alternant way; A computer that comprises camera is used to import the video photo that the user comprises people's face; A video identification authentication module is used to accept the photo that the user sends, and sends instruction to the user, finishes the authentication of end user's identity.
The full-automatic video identification authentication method may further comprise the steps:
(1) user uploads a video photo that comprises people's face of taking then and there;
(2) receive the initial video photo that step (1) uploads after, computer sends action command, require the user make some action (such as, cover some zone of face with hand, perhaps open one's mouth, perhaps close one's eyes, or the like);
(3) user uploads the new photo that satisfies command request of taking then and there according to action command;
(4) computer judges whether the new photo of receiving that at the appointed time the user uploads, if, then carry out next step, if not, then to judge overtimely, authentification failure finishes authentication;
(5) computer adopts the photo of uploading based on the photo control methods contrast step (1) and the step (3) of recognition of face, judges whether to be same people, if, then carry out next step, if not, then authentification failure finishes authentication;
(6) computer judges whether to have satisfied the requirement of contrast number of times, if then authentication success finishes authentication, if not, then returns to step (2).
Purpose of the present invention can also further realize by following technical measures:
Aforesaid full-automatic video identification authentication system also comprises a server, and the video identification authentication module is contained in the server, connects by network between computer and the server.
Aforesaid full-automatic video identification authentication system, wherein said video identification authentication module comprises interface module, in order to finishing the transmission of photo and computer instruction, and identification module, in order to finish based on the photo contrast of recognition of face and the judgement of user identity.
Aforesaid full-automatic video identification authentication method, wherein said photo control methods based on recognition of face may further comprise the steps:
1. define two class people faces: model people face, i.e. one or more people's face as standard of the initial input of user; Wait to adjudicate people's face, promptly the user is after computer sends action command, and the user is according to people's face of action command input;
2. to the model people face of user input with wait to adjudicate people's face and carry out people's face and detect, and the people's face picture after detecting is carried out feature location, normalization and preliminary treatment;
3. model people face with wait to adjudicate people's face and be divided into two zones, a zone is the zone that the user operates according to instruction, promptly instructed the zone that influences, another zone to be the zone that user's action should not cause the marked change of people's face picture, promptly do not instructed the zone that influences;
4. to being instructed the region decision that influences whether the variation of instruction regulation has taken place, to not instructed the region decision that influences whether to be same people;
5. comprehensive step judged result is 4. made the face authentication judgement.
Aforesaid full-automatic video identification authentication method, wherein said step 4. in, judge and not instructed the zone that influences whether to be that same people's method is for based on PCA (PCA), linear discriminant analysis method (LDA), elastic graph matching algorithm (EBGM) or local binary pattern method (LBP), judge whether instructed the zone that influences whether the method that the instruction regulation changes has taken place takes place for adopting grader to adjudicate an action, described grader is AdaBoost, utilize the nearest neighbor classifier or the Bayes classifier of principal component analysis (PCA).
Aforesaid full-automatic video identification authentication method, wherein said photo control methods based on recognition of face may further comprise the steps:
1. define two class people faces: model people face, i.e. one or more people's face as standard of the initial input of user; Wait to adjudicate people's face, promptly the user is after computer sends action command, and the user is according to people's face of action command input;
2. to the model people face of user input with wait to adjudicate people's face and carry out people's face and detect, and the people's face picture after detecting is carried out feature location, normalization and preliminary treatment;
3. model people face with wait to adjudicate the grid that people's face is divided into a plurality of fixed sizes zone;
4. each net region is calculated the cost function of face authentication judgement;
The cost function of the 5. comprehensive All Ranges face authentication decision of conducting oneself.
Aforesaid full-automatic video identification authentication method, wherein said step 4. in, the implementation method of cost function is to adopt local binary pattern (LBP) to extract the textural characteristics histogram in zone, E (m, n, i, c)=Dist (LBP (m
i), LBP (n
i)) * α (c
Mi, c
Ni), m representation model people face in the formula, n represents to wait to adjudicate people's face, mi represents i the zone of people's face m, c represents the influence degree that regional i is instructed, E (m, n, i, c) the regional i of expression confirms that m and n are same individual's cost functions, what α represented is the weight coefficient that regional i is influenced by computer instruction, the Dist function representation be two distances between the LBP histogram, the distance that can adopt comprises X
2, L1, L2 distance etc.
Aforesaid full-automatic video identification authentication method, the method that wherein said people's face detects is Adaboost or SVMs (SVM) or neural net or latent markov model, the preliminary treatment of described people's face picture is image alignment, geometric correction or brightness adjustment.
Advantage of the present invention is: the present invention is a kind of full automatic video identification authentication system and method, and it has adopted the computer face recognition technology, does not need manually to adjudicate, and can reduce labor workload greatly.Simultaneously its access authentication result has in real time improved recognition speed greatly.Native system is simple in structure clear, does not need complicated backstage customer service system, is easy to realize that the system building cost is very low.This method is widely used in the occasion of the various people's of needs face authentication, can effectively discern the true and false of user identity, promotes the development of various internet, applications.
Description of drawings
Fig. 1 is the system construction drawing of full-automatic video identification authentication system standalone version of the present invention.
Fig. 2 is the system construction drawing of the full-automatic video identification authentication system network edition of the present invention.
Fig. 3 is the flow chart of full-automatic video identification authentication method of the present invention.
Fig. 4 is the flow chart based on the photo comparison method one of recognition of face.
Fig. 5 is the flow chart based on the photo comparison method two of recognition of face.
Embodiment
The composition of full-automatic video identification authentication system of the present invention has multiple mode, comprises the standalone version and the network edition, the standalone version system configuration as shown in Figure 1, the system configuration of the network edition is as shown in Figure 2.In the standalone version identity authorization system, the module that relates to comprises user and a computer that comprises camera.Authentication module resides in the computer with software or hardware mode.By the mode of user and computer interactive, authentication module is finished the authentication of user identity.In network edition identity authorization system, the module that relates to comprises the user, a computer that comprises camera, and a station server.Connect (as the Internet, perhaps a wireless network) by network between local computer and the server.Local computer be used for finishing and user's mutual and local computer and server between communication.Authentication module resides in the server with software or hardware mode, finishes the authentication of user identity.The function that authentication module is finished comprises accepts the photo that the user sends, and sends instruction to the user, and the authentication of end user's identity.
The video identification authentication module comprises interface module, in order to finishing the transmission of photo and computer instruction, and identification module, in order to finish based on the photo contrast of recognition of face and the judgement of user identity.The function that identification module is carried out is according to the full-automatic video identification authentication method, produces the computer action instruction, and the recording instruction state is accepted the picture that the user uploads, and the authentication judgement is carried out in the photo control methods based on recognition of face according to the present invention.
Full-automatic video identification authentication method flow of the present invention may further comprise the steps as shown in Figure 3:
(1) user uploads a video photo that comprises people's face of taking then and there by camera;
(2) receive the initial video photo that step (1) uploads after, computer sends action command, require the user make some action (such as, cover some zone of face with hand, perhaps open one's mouth, perhaps close one's eyes, or the like);
(3) user uploads the new photo of the camera shooting of satisfying command request according to action command;
(4) computer judges whether the new photo of receiving that at the appointed time the user uploads, if, then carry out next step, if not, then to judge overtimely, authentification failure finishes authentication;
(5) computer adopts the photo of uploading based on the photo control methods contrast step (1) and the step (3) of recognition of face, judges whether to be same people, if, then carry out next step, if not, then authentification failure finishes authentication;
(6) computer judges whether to have satisfied the requirement of contrast number of times, if then authentication success finishes authentication, if not, then returns to step (2).
Photo comparison based on recognition of face can have multiple implementation method, the basic thought of a kind of method wherein is at first people's face to be divided into zones of different by the influence of computer action, the facial image of comprehensive then these zoness of different changes adjudicates, its flow process as shown in Figure 4, idiographic flow is as follows:
Define two class people faces: model people face (that is one or more people's face as standard of the initial input of user); And wait to adjudicate people's face (after computer sends instruction, people's face of user input).For people's face of user input, at first carry out people's face and detect and obtain human face region.People's face detects has multiple implementation method can select (as Adaboost, SVMs (SVM), neural net, latent markov model etc.).People's face can carry out image alignment after detecting, geometric correction, optionally picture preliminary treatment such as brightness adjustment.Then, according to the difference of computer instruction, model people face with wait to adjudicate people's face and be divided into two zones, a zone is the zone that the user operates according to instruction, as the zone of opening one's mouth, the zone of closing one's eyes, the zone that blindfolds with hand etc.; The zone that the action that second zone is the user should not cause the marked change of people's face picture, such as, the user carries out when closing one's eyes instruction, and mouth region should not change significantly.The various computing machine instructs corresponding area dividing template to establish according to different instructions in advance.Given model people face and wait to adjudicate people's face is divided these facial images respectively according to the area dividing template.Suppose face with m representation model people, n represents to wait to adjudicate people's face, C represents the zone of user according to instruction manipulation, B represents the zone of not operating, the zone according to instruction manipulation of Cm representation model people face then, Bm representation model people face not according to the zone of instruction manipulation, f represents the function that two zones of a judgement are whether similar, g represents to adjudicate the function whether a zone has carried out required movement, and then the criterion of face authentication is as follows:
If (Bm, (Cn Cm) satisfies the condition that zone C executes instruction to f, and then adjudicating m and n is same individual Bn) to satisfy similar condition in two zones of Bm and Bn and g.Otherwise judgement m is different people with n.Function f can be anyly can determine whether two people's face image-regions are functions of same individual, substantially can adopt any face recognition algorithms, as based on principal component analysis (PCA), linear discriminant analysis (LDA), elastic graph matching algorithm (EBGM), the method for local binary pattern (LBP) etc.Function g is used for adjudicating a zone whether required movement has taken place, and can adopt different decision devices according to different actions.Whether the grader that a kind of implementation method is different to different action trainings is in advance adjudicated an action with these graders and is taken place.Operable grader comprises AdaBoost, utilizes arest neighbors (NN) grader of PCA, Bayes (Bayes) grader etc.
The second method of comparing based on the photo of recognition of face is people's face of importing for the user, at first carrying out people's face detects and obtains human face region, then people's face is divided into the area grid of some fixed sizes, last comprehensive All Ranges grid is adjudicated, and its flow process as shown in Figure 5.People's face detects has several different methods to realize (as Adaboost, SVMs (SVM), neural net, latent markov model etc.).People's face can carry out image alignment after detecting, geometric correction, and optionally picture preliminary treatment such as brightness adjustment is carried out grid then and is divided.For example, can at first be normalized to 130 * 150 pixel sizes to everyone face, be divided into the fixed area grid of many 7 * 7 pixel sizes then.Suppose the face with m representation model people, n represents to wait to adjudicate people's face, and mi represents i of people's face m zone, and c represents that the influence degree that regional i instructed (represents that such as, c=1 the influence of being instructed certainly, c=0 represent the influence of not instructed certainly; The c value can be the two-value that disperses, and also can be continuous), E (m, n, i, c) the regional i of expression confirms that m and n are same individual's cost functions, then m and n be same individual cost E (m, n) be All Ranges sum total E (m, n)=∑ E (m, n, i, c).If (m, n) greater than some threshold values, then accept m and n is same individual to E.Otherwise, judge that m and n are not same individuals.(i c) can adopt multiple algorithm to cost function E for m, n.A kind of implementation method is to adopt local binary pattern (LBP) to extract the textural characteristics histogram of regional i, uses X
2Distance is represented two histogrammic distances of LBP, then E (m, n, i, c)=X
2(LBP (m
i), LBP (n
i)) * α (c
Mi, c
Ni), what α wherein represented is the weight coefficient that regional i is influenced by computer instruction.One of α simply follows the example of and is: if c=0, α=α 1; If c=1, α=-α 2.α 1 and α 2 are predetermined positive constants.Relation between α and c and the i also can be represented with other complicated any nonlinear functions.(i c) also can realize based on additive method cost function E for m, n, such as elastic graph matching algorithm (EBGM).