CN101233529A - Method and apparatus for enrollment and authentication of biometric images - Google Patents

Abstract

A method for enrolling biometric images including the steps of: a) capturing (310) a plurality of images for a user into a capture folder; b) selecting (318) one of the plurality of images in the capture folder and removing the selected image from the capture folder to an enroll folder; c) comparing (322) the selected image to each of the remaining images in the capture folder to generate a corresponding similarity score for each of the remaining images; d) determining (326) whether any of the corresponding similarity scores are at least equal to a predetermined score threshold, and removing each image having a corresponding similarity score at least equal to the predetermined score threshold from the capture folder to a delete folder (330); and e) determining (334) whether there is at least one image in the capture folder and if so repeating steps b) through d).

Description

The method and apparatus of biometric image registration and authentication

Technical field

The present invention relates generally to biometrical identification systems, and relate in particular to the user and register its biometric image and carry out the method and apparatus that the user checks based on registered image subsequently.

Technical background

The administration of justice and civil area in modern society are being played the part of important role based on the recognition system of biometric image.For example, public safety department is with criminal's the identification major part as any investigation now.Similarly, in civilian application, become the important component part of security procedure such as credit card or personal identity fraud, print identification.In all biometric characteristics (face, fingerprint, iris etc.), iris and retina are as the preferred biometric indicators that needs high security applications.However, very general because historical reasons reach the performance that has been confirmed in this field based on the check system of fingerprint, and facial images match is the second biometric characteristic index of using always in the identification.

Operation based on the automatic identification of biometric image for example, is carried out the identification of fingerprint, palmmprint or face-image, is made up of two stages usually.The firstth, registration or be called the typing stage, the secondth, identification, authentication or be called verification stage.At registration phase, the person's of being registered personal information and biometric image (as fingerprint, palmmprint, face-image etc.) are entered into system.Biometric image is caught by suitable sensor, generally can extract the feature of this biometric image then, for example the unique point of fingerprint.Usually with personal information and the feature extracted, may also comprise this image then, form a file record, be kept at the database that is used for discerning subsequently the person of being registered.

In identification/verification stage, need catch individual biometric image or obtain potential trace.Usually said searching record promptly is made up of feature and personal information that these images extract.Searching record will compare with registered (promptly filing) record in the database of recognition system subsequently.The result of comparison procedure will produce the tabulation of a coupling mark usually, and candidate's record is arranged according to the coupling mark.The coupling mark is that a kind of of characteristic similarity of identification search and file record measures.Usually, the coupling mark is high more, can think that the similarity of filing with searching record is big more.Like this, the candidate record on top is immediate coupling.

In recent years, the development of sensor technology make be used for registering and discern/that verification stage is used for the sensor that biometric image catches is more compact.Reducing on this size also makes the manufacturing cost of sensor reduce.For example, some manufacturer can place similar cellular hand-held wireless device with miniature non-optical formula fingerprint sensor (being solid state sensor).In above-mentioned example, the area of catching of sensor is generally less than the fingerprint total area that need catch, and this may cause the difficulty in the identification of the fingerprint that the sensor by these small sizes captures.The exemplary capture area of solid state fingerprint sensor only has 300 * 300 pixels.Yet the average area of finger to be caught can be its 3 times.

Same point in difference constantly (for example neutralizing in verification stage) impression of capturing in registration phase have only the possibility of very little overlapping region to cause when using these microsensors restriction about fingerprint recognition.More particularly, usually only can fingerprint image of typing file at registration phase a part of representative of the fingerprint of the actual finger of catching (only by), the feature of this image can be extracted and preserve be used for subsequently fingerprint search relatively.If the matching algorithm that is based on unique point that adopts has only under a small amount of overlapping situation in search and file fingerprint, the unique point quantity of pairing also is limited, so just causes the loss of matching accuracy.The loss of this accuracy may cause unwarranted personnel to be identified as authorized user by mistake, and perhaps authorized user is prevented from using this application program.No matter which kind of situation has all caused very big inconvenience to the user.Use the palmmprint identification of the area of the palm that the sensing area catches less than need also to be subjected to the similar restriction that is run into above-mentioned fingerprint recognition.

For the identification problem that above-mentioned microsensor causes, several known possible solutions are arranged.Yet each scheme all has its limitation.For example, can increase the size of sensor, but this needs more expensive sensor usually, therefore improved the cost of built-in sensors product.In addition, because the small size of product needed, this scheme may be used inapplicable to some.Another kind of solution is to use the image demonstration to come to provide guidance for the user who is registering.But be subjected to the restriction of size (such as the size of equipment), placing display on equipment may be unactual for some is used.Also have, other has solution to require the user that finger is placed on different positions when verification stage is caught fingerprint.For the user, consuming time in audit process of this scheme becomes longer, thereby makes its application in real society seem not too practical.

The solution of the identification problem that another causes for above-mentioned microsensor is referring to the process flow diagram among Fig. 2.In this case, produce the file record that mosaic fingerprint image is formed, replaced the person's of being registered single fingerprint image to be hunted down and be stored as the part of file record.For finishing this mosaic, sensor (210) is caught fingerprint image, until judge (214) its greater than a default quality threshold.If image has surpassed this quality threshold, it is an archival image (216) that this image is registered.Whether the quantity of judging (236) registered images then reaches the required quantity of default registered images just.If no, repeating step 210 to 216 is until the quantity that reaches required registered images.Then produce a mosaic image (240) by all enrolled images.Extract the feature (220) of this mosaic image again, mosaic image and corresponding matching characteristic are stored as file record (224).

The several difficult of this method makes it in the application of actual life and be not easy.For example, the process of combination mosaic image itself is exactly the process of a coupling, and the ridge of corresponding fingerprint line links with ridge in a plurality of images that need have no will capture mistakenly, the Gu Yugu link.Yet because the distortion of image, noise and other uncertain factor in catching, this does not normally accomplish.Correspondingly, the mosaic image of generation does not have level and smooth transition at the image boundary place of separate captured at large.Limitation in these mosaic image production processes can cause in verification stage, and unique point is detected by mistake, has also reduced the accuracy of coupling.

The front is mentioned, and facial images match is the second biometric characteristic index that is commonly used to discern.This has been implemented in the identity retrieval such as video monitoring identification, inlet control, crime survey database.A benefit of this identification is that the process right and wrong of obtaining are invasive, does not also need these personnel's cooperation.Yet its limitation is that the expression of common face-image or the viewing angle that captures may be different with registered image, caused the reduction of matching accuracy.The problem of this accuracy can by at registration phase from facial different angles and under difference expression, carry out catching of a plurality of different images and solve.But because the restriction of system storage capacity, the quantity of the face-image of catching has the limit of a reality, and the time of extra registration image also wishes to be controlled in the acceptable scope.

Therefore, the area of catching at the sensor that recognition system comprises is caught less than need in the biometrics authentication application of biometric characteristic area, and existence is for the demand of the method and apparatus of the biometric image (for example fingerprint, face-image and palmmprint) of differentiation and storage acceptable number.Can wish further that this method improves the possibility of correct identification in audit process, reduces the possibility of mistake identification.

Description of drawings

Now illustrate the preferred embodiments of the present invention, with reference to accompanying schematic figure, wherein:

Fig. 1 is the simple structure block diagram according to the biometrical identification systems of the embodiment of the invention;

Fig. 2 is a kind of process flow diagram of art methods of fingerprint register;

Fig. 3 is the process flow diagram according to a kind of biometric image register method of the embodiment of the invention;

Fig. 4 is the process flow diagram according to a kind of biometric image register method of the embodiment of the invention;

Fig. 5 is the process flow diagram according to a kind of biometric image register method of the embodiment of the invention;

Fig. 6 be according to the embodiment of the invention for being identified for controlling the distribution curve of fingerprint register and the coupling print scores and the print scores that do not match of the threshold value of checking;

Fig. 7 is the process flow diagram according to a kind of biometric image checking method of the embodiment of the invention; With

Fig. 8 determines shown in Fig. 7 the process flow diagram of the method for used threshold value in the checking method.

Embodiment

Though the present invention can be with shown in the synoptic diagram and be about in this implemented in many forms of describing in detail and clearly, needing the method for clearly existing announcement be an illustration of the principle of the invention, and is not intended to limit the invention to the particular implementation form of showing and describing.In addition, term and the wording that adopts here only should not be considered to restrictive for descriptive.Equally should be clear and definite, for the terseness and the definition of showing, the element in the synoptic diagram etc. there is no the necessity of drawing to scale.For example, with respect to other element, amplified the size of subelement.In addition, its mark numbering in order to show corresponding element, has been reused in suitable place in reference diagram.

Fig. 1 is the simple structure block diagram according to the biometrical identification systems 10 of the embodiment of the invention.Illustrate, system 10 can be included in the fingerprint recognition system that embeds in the foregoing mobile phone, or embeds such as in the biometric identification application programs such as palmmprint identification and facial image recognition system.Ideally, system 10 comprises input and enrollment station 140, data storage and retrieval facility 100, one or more matched-field processors 120, and verification station 150.

Input is used to catch such as the biometric image of fingerprint with enrollment station 140 and extracts the matching characteristic relevant with this image alternatively to be used for comparison subsequently.Also can in input and enrollment station 140, create the file record of image that captures and the feature that extracts.Input and enrollment station 140 can be as mentioned below, are configured to realize registering functional according to the embodiment of the invention.Thereby the foregoing area of catching can be connected with input with enrollment station 140 less than the sensor of the image area that is hunted down.Sensor for example is optical sensor or solid state sensor.Input further is connected or combination with processor device with enrollment station 140, to realize other function.

Data storage and retrieval facility 100 are used to store and retrieve the file record that comprises matching characteristic, also can store and retrieve other and help realizing data of the present invention.Matched-field processors 120 can utilize the matching characteristic of the biometric image that extracts to determine its similarity or be configured to compare in the image aspect.This type of matched-field processors can be the conventional Feature Points Matching device that is used for the unique point that comparison two width of cloth fingerprints or palmprint image extract.In facial images match was used, matched-field processors can be made up of principal component analysis coupling, facial characteristics coupling, local feature analysis coupling or other matching algorithms.

Finally, verification station 150 is used and is checked matching result according to the method for the embodiment of the invention.Therefore, verification station 150 is used for catching such as the biometric image of fingerprint and extracts the matching characteristic relevant with this image alternatively and compares with the matching characteristic of one or more file record being used for.Verification station 150 also can be created the searching record of image that captures and the feature that extracts.Thereby verification station 150 also can be connected with the sensor of acquisition search image, and is connected or combination with processor device, to realize other function.

Although will be appreciated by those skilled in the art that in system 10, the box indicating that input and enrollment station 140 and verification station 150 usefulness are separated, these two websites can be merged into a website in another alternate embodiments.In addition, when system 10 is used for searching record with a specific people when comparing with a plurality of file record of different personnel, comprise distributed matching controller (not shown) in the system 10 alternatively, it can comprise processor, and more complicated or time-consuming matching process is more effectively coordinated in configuration.

Fig. 3 is the process flow diagram according to a kind of biometric image register method of the embodiment of the invention.This method can realize in one or more processors of system 10, and it is more convenient to make the set of diagrams picture (and corresponding feature) of catching in later time the person of being registered be carried out effective accurate recognition.For the convenience that illustrates, the method is described according to fingerprint recognition.In any case the method is realization, for example palmmprint or face-image registration in the biometric image registration of other type similarly.

According to method shown in Figure 3, can place on the sensor by finger that will the person of being registered and (310) a plurality of fingerprint images are caught and registered to the diverse location that moves around to the sensor.When finger touched sensor, sensor was caught the snapshot image of fingerprint continuously.The image of typically, catching can characterize the many different laps of fingerprint.Suppose that N image is registered and stores into one and catch file (310), wherein N may be a preset value, as makes a function of memory requirement (the more little required storage of N just is more little) and the required degree of accuracy of system (just the big more degree of accuracy that can reach of N is high more) balance.Similarly, a plurality of palmprint images or face-image can be caught and stored into and catch file.Preamble had explanation, also can catch the image of facial different angles and the portrait of different expressions.This catch file can be stored in input and memory device that enrollment station 140 is connected on, for example on data storage and the retrieval facility 100.

The feature of these N that is used to mate images, the unique point of fingerprint for example will be by typically but not necessarily extract, and be stored in equally and catch file (314).When image aspect (with respect to the feature aspect) upward compares image, it is unnecessary that feature extraction just becomes.Thereafter, in catching whole fingerprint images of file, select a width of cloth as searching for fingerprint image and being stored in the register-file folder, for example in data storage and the retrieval unit 100, remaining fingerprint image is stayed as one group of background file fingerprint image and is caught (318) in the file.The feature of search fingerprint image will compare with the feature of remaining each width of cloth background file fingerprint image, and mates mark (being called similarity score again) (322) by matched-field processors 120 (for example, Feature Points Matching device) for relatively producing each time.

If the coupling mark of determining corresponding to the background file fingerprint image is more than or equal to a preset threshold value Te (326), then this background file fingerprint image and corresponding matching characteristic thereof are moved and store into an interim deleted file folder (330) from catch file, for example in data storage and the retrieval unit 100.When definite (334) all fingerprint images all shift out from catch file, just move to interim deleted file folder or register-file folder, method shown in Figure 3 finishes.This moment, register-file folder was finished, and was stored in the comparison that wherein image will be used for the fingerprint of verification stage search subsequently.Otherwise method is got back to step 318, selects from catch file and with another width of cloth image, puts into register-file with its matching characteristic and press from both sides.

As shown in Figure 3, threshold value Te is controlling the quantity of registering the fingerprint image of storing in the file.Therefore, step 326 (as the function of Te) is used for having the image of too much similarity to remove with selected searching image on the register-file folder.Do the redundant image that can reduce in the register-file folder like this and occur, thereby reduced the memory requirement that register-file presss from both sides.The value of Te should be the function of at least one feature of adaptation use at first, with reference to shown in Figure 6.But, size sensor with wait to catch the size of images relation and also can influence the value of Te because the grade of coupling mark is different.

In order to estimate the degree of accuracy of biometric image adaptation, the degree of accuracy of fingerprint matching device for example must be collected in several fingerprints that the same hand refers to the mark that the produced distribution curve 620 of fingerprint (just match) and the mark (the just distribution curve 610 of non-matching fingerprint) of several fingerprints to being produced on different fingers.In typical commercial the application, Te value is selected in the intersection point place of coupling mark and non-coupling score distribution curve, or rate (EER) point such as employing statistics, as shown in Figure 6.At this threshold value place, mistake matching rate (FMR) equates with the non-matching rate of mistake (FNMR).FMR be when input from different man-hours, system judges that this people is he that people's of declaring a probability.FNMR works as input from identical man-hour, and system judges that this people is not that people's that he declared a probability.

Threshold value Te also may get a value that is greater than or less than EER, and this depends on the design standards of memory requirement or the fingerprint quantity of final register list needs.If design standards has been specified smaller memory requirement, just the fingerprint in final registration is fewer, should select a less Te threshold value so.On the contrary, if design standards has been specified bigger memory requirement, just the fingerprint in final registration is many, should select a bigger Te threshold value so.In addition, as shown in Figure 6, the minimum value T1 of threshold value is taken at 1 FNMR point, and the maximum of T 2 of threshold value is taken at 1 FMR point.If therefore the Te value has exceeded the scope of T1 and T2, can increase certain mistake, do not reduce another kind of mistake simultaneously, vice versa.So ideally, Te should be between T1 and T2 value.Threshold value T1 shown in Fig. 6, T2 and Te illustrate with fingerprint recognition.But those of ordinary skills will appreciate that, also can determine these threshold values similarly on the adaptation that uses such as coupling palmmprint and face-image the time.

Fig. 4 and Fig. 5 are the process flow diagrams according to a kind of biometric image register method of the embodiment of the invention.Similar with the method shown in Fig. 3, for the convenience that illustrates, will the method be described according to fingerprint recognition.In any case the method is realization, for example palmmprint or face-image registration in the biometric image registration of other type similarly.In this embodiment, ideally, the image that only reaches certain mass just can be hunted down and be stored in and catch in the file at registration phase.

According to register method shown in Figure 4, use sensor to catch the fingerprint image in the certain zone of (410) finger, extract (414) its matching characteristic alternatively.Typically, do not consider the quality of image, the quantity of the person's of being registered who catches image also has a peaked restriction, in order to avoid bring too much inconvenience for the person of being registered.Yet this requirement is not necessary.Whether the quality of judging the image that catch (418) then is more than or equal to default quality threshold.If, with image and corresponding matching characteristic thereof be stored in the register-file folder (422) if not, image and corresponding matching characteristic thereof are stored in the temporary folder (434).

About catching of fingerprint image, being used for of using in the step 418 quality threshold of selecting to put into the image of catching file be rule of thumb to flow to distribution to draw based on the defective fingerprint of offline database (just ropy fingerprint) of recognition system 10 design processes and effective crestal line between the qualified fingerprint (the goodish fingerprint of quality just).In Palm Print Recognition System, quality threshold is used with the similar mode of fingerprint recognition system and is determined.In facial match, quality threshold can loosen for each image registration of allowing to catch in system, allow accreditation process select final registered images then.

Whenever piece image stores into when catching in the file, judge that (426) catch the image whether file has comprised requirement (for example, default quantity).If, catch file and finish, the step 442 in the execution graph 5 is set up the register-file folder to 458 with the image of catching in the file then.Step 442 among Fig. 5 to 458 with Fig. 3 in step 318 be identical to 334.Therefore, for simplicity, no longer repeat specification step 442 is to 458.If but catch the image that file does not comprise requirement, also need further judgement (436) whether to reach the maximum times that trial is caught.If reach maximum times, then select image and corresponding matching characteristic thereof in the temporary folder, store into and catch file, catch the required amount of images of file (438) until reaching.Execution in step 442 to 458 then, set up the register-file folder with the image of catching in the file.In addition, if do not reached the maximum times that trial is caught as yet, program is got back to step 410, catches another width of cloth image (being another zone of finger ideally) of finger.

When piece image stores in the temporary folder, judge whether (436) have reached the maximum times that trial is caught.If reach maximum times, then select image and corresponding matching characteristic thereof in the temporary folder, store into and catch file, catch the required amount of images of file (438) until reaching.Execution in step 442 to 458 then, set up the register-file folder with the image of catching in the file.In addition, if do not reached the maximum times that trial is caught as yet, program is got back to step 410, catches another width of cloth image (being another zone of finger ideally) of finger.

Fig. 7 is the process flow diagram according to a kind of biometric image checking method of the embodiment of the invention, can be realized by verification station 150 (Fig. 1), and the one or more processors in can using system 10 is realized.In addition, the method can be applicable to polytype biometrics authentication, comprises fingerprint, palmmprint and facial image authentication.In order to check a user, for example authorize its access system, must retrieve its register-file folder and corresponding default verification threshold (710).In the system of a register-file folder of having stored a plurality of users, the retrieval of this information can be triggered by the personal information system of being input to the user, for example imports user's name or certain type suitable identification code.But for the system that only need check a user, mobile phone is for example only caught a searching image (714) and just can have been triggered retrieval that suitable register-file is pressed from both sides on sensor.

In case caught searching image,, then in this width of cloth searching image, extracted feature (718) if on the feature aspect, carry out comparison.Then the feature of every width of cloth image in the feature of searching image and the register-file folder is mated, and produce corresponding coupling mark (722).If confirm (726) each coupling mark all more than or equal to verification threshold, then granted access (735).If confirm (726) all coupling marks all less than verification threshold, then denied access (730).Alternatively, when confirming that number of times that (734) attempt checking is less than the maximum times (default number of attempt just) of permission, by catching another width of cloth searching image (714) to repeat this process.Otherwise, if reached maximum attempts, finish this process, and refuse this user capture system.The catching of searching image that allows repeatedly to attempt making at least one width of cloth that enough quality are arranged can be used for checking of user.The control number of attempt reaches maximum quantity, and user's is inconvenient minimized helpful for making in verification stage for this.

An advantage of this invention is that in multi-user's system, a verification threshold is not at all users.In this invention, each user's verification threshold is individually to determine.Fig. 8 is the process flow diagram of judging in the method that determines whether to authorize user's verification threshold of using in specific user's access system.The deleted file folder that produces for this user in the register method shown in Fig. 3 (and Fig. 5) obtains using in this embodiment.

Select the piece image of deleted file folder and mate with each width of cloth in the final registered images of the M width of cloth of user's register-file folder.The matching characteristic of each width of cloth image compared during matching process typically pressed from both sides with register-file by the matching characteristic with selected image in the deleted file folder, for example, use matched-field processors 120 (for example Feature Points Matching device), produces M and mate that mark (810) finishes.Select the highest mark Si in (814) this M the coupling mark.The selection of this highest score Si helps the smallest match mark of corresponding deleted image, thereby makes non-this user's searching image can not surpass verification threshold, even this image and this user's biometric image has certain similarity.Repeating step 810 to 814 all compares with the feature of every width of cloth image of register-file folder until the feature of every width of cloth image of confirming (818) deleted file folder, thereby has produced N-M coupling mark Si the highest.In ading up to the Si mark of N-M, select minimum mark (822), help guaranteeing that the searching image with any deleted image coupling surpasses verification threshold.Verification threshold Th can be made as the minimum Si coupling mark (826) that this is chosen then.

In addition, can determine (826) verification threshold Th according to following algorithm.If minimum Si coupling mark then mates mark as verification threshold Th with minimum Si greater than first default minimum threshold T1 and less than second default max-thresholds T2.If minimum Si coupling mark then is made as T1 with Th less than T1.In all other circumstances, Th is made as T2.This algorithm helps guaranteeing that verification threshold Th does not exceed the scope based on the Relational database of adaptation and corresponding pairing and non-matching image (for example, respectively, the distribution curve 620 and 610 among Fig. 6).

Under the situation of coupling fingerprint, threshold value T1 and T2 are based on that the statistical distribution curve of coupling print scores that adaptation uses and non-coupling print scores calculates in advance.More particularly, select T1 and T2 as Fig. 6, wherein T1 is 1 FNMR point, and T2 is 1 FMR point.The verification threshold Th that calculates can be stored under for example corresponding individual ID, and be used for determining whether to find coupling in verification stage.In addition, as previously mentioned, threshold value Th, T1 and T2 determine with similar method, for example in palmmprint identification and facial image recognition system.

Refer again to the audit process shown in the process flow diagram of Fig. 7, can further handle the image that causes denied access.For example, can will cause the image and the corresponding matching characteristic thereof of denied access to store in the searching record, and compare with file record in criminal's database, to confirm whether for example identity theft has taken place or confirmed whether the owner of this image is relevant with the crime survey record.In addition, be a certain position user's coupling if known one or more cause the image of denied access, these images can be added to this user's register-file folder, and the verification threshold new according to the image calculation of these addings.

Biometric image of the present invention is registered and is checked compared with prior art, has multinomial advantage.Several advantages of hereinafter listing should not be considered to only advantage, also should be in office where face is considered to limitation of the present invention.For example, registered a plurality of images at registration phase of the present invention, rather than single image or mosaic image, the matching accuracy of verification stage subsequently improved.In addition, the present invention makes the biometrics right discriminating system reach best accuracy and speed for the method for determining to provide system of the quantity of the image sets that should register and feature group, has kept minimum memory requirement simultaneously.

Though the present invention is described in conjunction with specific embodiment, those skilled in the art can easily draw its added benefit or modify.Therefore the present invention aspect widely, also is not limited to show and the specific detail of describing, representational equipment and illustrative example.According to the description of front, multiple possible change, modification and variation are conspicuous to those skilled in the art.Therefore should clear and definite the present invention be not limited only to the description of preamble, and comprise according to all changes, modification and variation in the spirit and scope of claim of the present invention.

Claims (10)

1. method that is used to register biometric image comprises step:

A) user's multiple image is captured to catches in the file;

B) select the described width of cloth in the multiple image in the file of catching, the described image of choosing is moved into the register-file folder from catching file;

C) with the image chosen with catch that remaining every width of cloth image compares in the file, be that remaining every width of cloth image produces corresponding similarity score;

D) similarity score that judges whether any correspondence equals the score threshold preset at least, and the every width of cloth image that described corresponding similarity score is equaled at least the preset fraction threshold value moves on to the deleted file folder from catching file; And

E) judge whether catch file has piece image at least, if, repeating step b) to d).

2. the method for claim 1, wherein use matched-field processors to choose image and describedly catch that remaining every width of cloth image compares in the file with described.

3. method as claimed in claim 2, wherein, the function of at least one feature that described score threshold is described matched-field processors, and select described score threshold, the max-thresholds that makes it to equal the minimum threshold of described matched-field processors at least and be not more than described matched-field processors.

4. method as claimed in claim 3, wherein:

Described minimum threshold is 1 mistake non-matching rate (FNMR) point of pairing image distribution curve;

Described max-thresholds is 1 mistake matching rate (FMR) point of non-matching image distribution curve; With

Described score threshold be described zero FMR point and described zero FNMR order etc. the rate point.

5. the method for claim 1, wherein as the function of default quality threshold, catch described multiple image in the file described catching.

6. method as claimed in claim 5 wherein, describedly is captured to the step of catching file with described multiple image and may further comprise the steps:

I) catch image;

Judge ii) whether described quality of having caught image equals described default quality threshold at least; With

If iii) its quality equals described default quality threshold at least, will catch image registration to the described file of catching.

7. method as claimed in claim 6, wherein, with the image capturing of predetermined number to the described file of catching, and, describedly described multiple image be captured to the step of catching file further may further comprise the steps:

Iv) quality is put into temporary folder less than the image of catching of described default quality threshold; With

The maximum that v) judges whether to reach default attempts catching number of times, and judges the described image whether file comprises described predetermined number of catching, and

If catch the image that file does not comprise described predetermined number, and do not reach described default maximum as yet and attempt catching number of times, then return step I); With

Attempt catching number of times if reached described default maximum, in described temporary folder, select image and put into the described file of catching, catch the image that file has comprised described predetermined number until described.

8. comparison based on the multiple image in one or more image in the deleted file folder and the register-file folder is pressed from both sides described in claim 1 and is produced for the user determines the method for verification threshold, described deletion and register-file, and described method comprises the steps:

A) in described deleted file folder, select piece image;

B) each width of cloth image in described image of choosing and the described register-file folder is compared, and be each described similarity score that relatively produces correspondence;

C) in described corresponding similarity score, select the highest similarity mark;

D) repeating step a), b) and c), each width of cloth image in described deleted file folder all with described register-file folder in each width of cloth image compare;

E) in the highest similarity mark of in all step c), selecting, select a minimum mark;

F) described user's verification threshold is defined as a function of described lowest fractional.

9. a method that is used to register biometric image comprises the steps:

A) user's multiple image is captured to catches in the file;

B) extract described corresponding matching characteristic of catching each width of cloth image in the file, store described corresponding matching characteristic and described image into described catching in the file;

C) in described multiple image of catching in the file, select a width of cloth, described image and the corresponding matching characteristic thereof chosen moved on to the register-file folder from catching file;

D) will choose the matching characteristic of image and the matching characteristic of catching remaining every width of cloth image in the file to compare, for remaining every width of cloth image produces corresponding similarity score;

E) similarity score that judges whether any correspondence equals the score threshold preset at least, and the every width of cloth image and the corresponding matching characteristic thereof that described corresponding similarity score are equaled at least the preset fraction threshold value move on to the deleted file folder from catching file; With

F) judge whether catch file has piece image at least, if, repeating step b) to d).

10. one kind is used for biometric image registration and the system of checking, and comprising:

A) be used for user's multiple image is captured to the device of catching file;

B) be used for selecting a described width of cloth of catching in the file multiple image, and the described image of choosing is moved into the device that register-file presss from both sides from catching file;

C) being used for comparing choosing image and catching remaining every width of cloth image of file, is the device that remaining every width of cloth image produces corresponding similarity score;

D) similarity score that is used to judge whether any correspondence equals the score threshold preset at least, and the every width of cloth image that described corresponding similarity score is equaled at least the preset fraction threshold value moves on to the device of deleted file folder from catching file;

E) be used for judging whether catch file has piece image at least, and if, with regard to repeating step b) to d) device;

F) be used for pressing from both sides with described register-file based on each width of cloth image of described deleted file folder in the comparison of each width of cloth image, determine the device of verification threshold for described user;

G) be used to catch the device of described user's piece image at least as searching image;

H) be used for described at least one width of cloth searching image and each width of cloth image of described register-file folder compares and produce the device of corresponding similarity score for register-file each width of cloth image in pressing from both sides;

I) being used for judging whether to have one at least at step h) the corresponding similarity score that produces equals described user's verification threshold at least, and if, the device of just authorizing this user capture system.

Images