Summary of the invention
For at least one of prior art the above defects or improvement requirements, the present invention provides a kind of based on certificateAuthentication system and method can detect the true and false of all kinds of certificates and differentiate the consistency of holder and the certificate owner.OnlyIt needs to be put into certificate in hardware device and accurate conclusion can be obtained, easy to operate efficient, accuracy rate is high.This authentication systemSystem can be applied to passport, Hongkong and Macro's pass, Taiwan compatriot certificate, Hongkong and Macro's home return permit, No.2 residence card, Hong Kong, Macao and Taiwan resident's residence permitAnd the certificates such as driver's license, and it is not limited to the above certificate.
To achieve the above object, it is another aspect of this invention to provide that additionally providing a kind of authentication system based on certificateSystem, comprising: image acquisition units, OCR unit, certificate false distinguishing unit, electronic chip recognition unit and recognition of face verifying are singleMember；
Described image acquisition unit is used to acquire including the certificate image and ultraviolet light conditions under the conditions of white light, infrared lightLower certificate fluorescence reaction image；
Certificate image under the conditions of the infrared light that the OCR unit is used to acquire described image acquisition unit is knownNot, the machine-readable code in figure, including uniline and the machine-readable code of multirow are identified；
Certificate image, purple under the conditions of the infrared light that the certificate false distinguishing unit is used to acquire described image acquisition unitCertificate fluorescence reaction image carries out false distinguishing under the conditions of outer light, judges the true and false of certificate；
The electronic chip recognition unit is for reading certificate chip interior information；
The recognition of face authentication unit carries out Face datection for capturing to holder, at least with described imageAcquisition unit acquisition white light conditions under certificate image in facial image carry out consistency discrimination, more preferably also with certificate coreThe facial image extracted in piece carries out three's consistency discrimination.
Preferably, described image acquisition unit includes camera and polychromatic source, the polychromatic source include white light source,Infrared light light source, ultraviolet source；
Preferably, the certificate includes various countries' passport, Hongkong and Macro's pass, Taiwan compatriot certificate, Hongkong and Macro's home return permit, middle state-two for identityCard, driver's license, Hong Kong, Macao and Taiwan resident certificate, foreigner's residence permit.
To achieve the above object, it is another aspect of this invention to provide that additionally providing a kind of authentication system based on certificateThe verification method of system, includes the following steps:
(1) certificate image acquisition step: certificate information homepage or certificate information are inserted into hardware device down, hardware is setStandby self-starting is acquired under the image W under white light conditions, image IR and ultraviolet source under infrared light supply by built-in cameraImage UV.
(2) the machine-readable code identification step of certificate: identifying the machine-readable code S1 in image IR using OCR technique, and verify S1, schoolIt tests and passes through, then carry out in next step；Otherwise, machine-readable code identification mistake is reported；
(3) reading machine reading code step: activating electronic chip reading unit using machine-readable code, obtains the machine-readable of the interior storage of chipCode S2 and facial image A (if chipless, which is omitted)；
(4) certificate information verification step: extracting feature to image W and facial image A and C by face recognition algorithms,Carry out consistency checking；Consistency checking is carried out to S1, S2 simultaneously (to be determined as genuine according to the character for being S1, S2 corresponding positionUnanimously).Consistency checking is true twice, then carries out in next step；Otherwise, report consistency checking failure；
(5) certificate image false distinguishing step: image IR and image UV is used, the anti-fake point of summary and corresponding false distinguishing are passed throughMethod obtains false distinguishing result.Certificate false distinguishing is by then carrying out in next step；Otherwise, certificate false distinguishing mistake is reported；
(6) the facial image C for obtaining holder recognition of face verification step: is shot by external camera；It is calculated with YOLO2Method carries out Face datection to image C and image W respectively and obtains facial image D and facial image E；If certificate has electronic chip,By facial image A and facial image D and facial image E input ResNet (residual error network) carry out feature extraction obtain feature toCosine similarity is measured and calculated, is then differentiated；Otherwise, facial image E and facial image D is inputted into ResNet (residual error netNetwork) it carries out feature extraction acquisition feature vector and calculates cosine similarity, then differentiated.If being verified, then it represents that bodyPart is verified；Otherwise, report authentication fails and calls Security Services；
Preferably, in step (2), the method for OCR technique HSV space binaryzation first carries out machine-readable code precise regionBinaryzation, is then partitioned into each character with sciagraphy, and the feature for finally needing to extract character picture (extracts character picture featureMethod be statistics character row pixel, column pixel, piecemeal pixel and combine), then matched with character feature table,The smallest remainder chordal distance is result；For the character easily distinguished, matching result is final recognition result, and for easily mixedThe character (such as ' A ' and ' 4 ') to confuse, by the methods of analysis symmetry (analyzing symmetry by statistics left and right pixel ratio) to matchingAs a result it carries out further division and obtains final recognition result.
Preferably, in step (4), image is pre-processed, pretreatment carries out gray processing to image first, thenEdge detection is carried out using Sobel operator, is finally smoothed using gaussian filtering；
In step (4), pretreated image is used into the spatial pyramid matching process based on SURF algorithm, is realizedThe extraction of the local and global feature of face；Wherein, the spatial pyramid matching process based on SURF algorithm is by original when matchingReason is changed to the characterization (SURF feature) of characteristic information, so that next step support vector machines and Adaboost carry out recognition of face；
In step (4), combination supporting vector machine and Adaboost are merged by the group of local feature and global feature, it canJudge whether image is consistent.
Preferably, it in step (5), for image IR and image UV, is primarily based on improved discrete cosine transform and carries outAnti-fake zone location and it is partitioned into anti-fake region using sciagraphy, then using decorative pattern detection, fluorescent fiber detection, data fieldThe false distinguishings method such as detection and personal photo area detection, anti-fake points a variety of to machine-readable code area, passport NO., anti-fake decorative pattern etc. carry outFalse distinguishing processing.
Preferably, in step (6), Face datection is carried out using YOLO2 algorithm, the purpose of Face datection is by face figureAs being elected from background center, comprises the concrete steps that adjustment image size first, then run neural network, finally carry out non-very bigValue inhibits.
Preferably, in step (6), facial image input ResNet (residual error network) that Face datection is obtained is extracted specialLevy vector.ResNet (residual error network) introduces " jump connection " on the basis of CNN (convolutional neural networks).Wherein, " jumpConnection " is defined by following formula:
Y=F (x, { Wi})+x
The x, y are respectively the input of building block, output vector, and F is mapping function, WiFor i-th layer of weight.
Preferably, in step (6), differentiation foundation be calculate image feature vector cosine similarity, if be higher than orEqual to the threshold k of setting, then it is judged as same people, is then determined as different people lower than threshold value.
In above-mentioned steps, the true and false of certificate is determined by step (4) and step (5), by step (6) determine holder withThe consistency of the certificate owner.
Above-mentioned preferred feature can be combined with each other as long as they do not conflict with each other.
In general, through the invention it is contemplated above technical scheme is compared with the prior art, have below beneficial to effectFruit:
1, the authentication system of the invention based on certificate, has that structure is simple, easy to use, to adopt plot quality height etc. excellentPoint, and it is able to ascend hardware system stability.
2, the authentication system and method for the invention based on certificate supports false distinguishing and the electronic chip of multiclass certificateRecognition, including and it is not limited to identity card, Taiwan compatriot certificate, Hongkong and Macro's pass, Hongkong and Macro's home return permit, various countries' passport, driver's license, foreigner residenceFirmly card etc., and the new part of issuing licence such as Hong Kong, Macao and Taiwan resident certificate is supported for the first time.
3, the authentication system and method for the invention based on certificate, OCR identification compatibility is strong, can distinguish uniline andThe machine-readable code of multirow and identity card, drivers license number, while can handle certificate inclination when the case where.
4, the authentication system and method for the invention based on certificate, has carried out in detail all kinds of certificate false proof featuresAnalysis and summary, using the relative positioning technology of improved discrete cosine transform, using decorative pattern detection, fluorescent fiber detection, numberAccording to a variety of false distinguishing means such as area's detection and personal photo area detection, the true and false of certificate itself is verified, false distinguishing accuracy rate is improved.
5, the authentication system and method for the invention based on certificate, the face recognition technology of use, Face datection portionDivide and use YOLO2 algorithm, recognition of face part uses ResNet, can get higher identification while faster processing speedAccuracy rate.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, rightThe present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, andIt is not used in the restriction present invention.As long as in addition, technical characteristic involved in the various embodiments of the present invention described belowNot constituting a conflict with each other can be combined with each other.The present invention is described in more detail With reference to embodiment.
As a kind of better embodiment of the invention, tested as shown in Figure 1, the present invention provides a kind of identity based on certificateCard system includes: image acquisition units, OCR (optical character identification) unit, travel document false distinguishing unit, electronic chip recognition listMember, recognition of face authentication unit.Image acquisition units capture mode using camera and carry out certificate information collection, and acquisition includes whiteLight, infrared, under the conditions of ultraviolet source certificate information image；OCR unit carries out the image under the conditions of the infrared light supply of acquisitionIdentification identifies the machine-readable code in figure, including uniline or multirow situation.Image false distinguishing unit, which is mainly responsible for, judges the true of certificatePuppet carries out the certificate true and false using face alignment and specific false distinguishing method and the means such as chip interior information and OCR informationDifferentiate.Recognition of face authentication unit captures holder, carries out Face datection, extracts feature vector and calculates images matchDegree, obtains the differentiation of holder Yu certificate owner consistency.Authentication system based on certificate of the invention, has structureSimply, easy to use, adopt the advantages that plot quality is high, and be able to ascend hardware system stability.
Described image acquisition unit includes camera and polychromatic source, and the polychromatic source includes white light source, infrared lightLight source, ultraviolet source.
The certificate includes various countries' passport, Hongkong and Macro's pass, Taiwan compatriot certificate, Hongkong and Macro's home return permit, Chinese China second-generation identity card, drivesCard, Hong Kong, Macao and Taiwan resident certificate, foreigner's residence permit, however it is not limited to this.
The present invention also provides a kind of verification methods of authentication system based on certificate, as shown in Fig. 2, application scenariosIt is illustrated for carrying out authentication based on international passport, the specific steps are as follows:
(1) image acquisition units obtain passport white light, infrared, under the conditions of ultraviolet source certificate information image；
(2) OCR unit identifies that the machine-readable code of certificate, activation electronic chip distinguish unit, obtain the card stored in electronic chipPart information and facial image.
(3) certificate information and OCR identification information consistency in electronic chip are verified；Facial image and white i.e. in proofing chipThe striation part following figure seems no consistent；Carry out the infrared false distinguishing with certificate information image under the conditions of ultraviolet source.
(4) external camera of recognition of face authentication unit captures the real-time face image of holder；
(5) the three width image of facial image under face in real-time face image, chip and white light is verified.
In this example, step (2) specific steps are as follows:
(2.1) to the machine-readable code precise region of certificate information image interception under the conditions of infrared light supply；
(2.2) binaryzation is carried out to machine-readable code precise region using the method for HSV space binaryzation first；Then using masterThe principal direction that constituent analysis (PCA) algorithm finds dumper reading code image is corrected tilted image using rotation transformation；ThenEach character is partitioned into sciagraphy；Finally need to extract character picture feature (extract character picture feature method be systemMeter character row pixel, column pixel, piecemeal pixel are simultaneously combined), it is then matched with character feature table, remainder chordal distanceThe smallest is result；For the character easily distinguished, matching result is final recognition result, and for confusing character(such as ' A ' and ' 4 ') carries out matching result by the methods of analysis symmetry (analyzing symmetry by statistics left and right pixel ratio)Further division obtains final recognition result.
(2.3) card reader in electronic chip recognition unit after the machine-readable code check of identification, will be passed to, carries out card reading behaviourMake.The information stored in passport chip includes machine-readable code and certificate owner's image.
In this example, step (3) specifically:
(3.1) facial image in image under white light conditions and chip is pre-processed (pretreatment operation include gray processing,Sobel operator edge detection, gaussian filtering smoothing)；
(3.2) local feature is obtained using the spatial pyramid matching process based on SURF algorithm to pretreated imageWith the combination of global characteristics；
(3.3) combination supporting vector machine and Adaboost are merged by the group of local feature and global feature, determines white lightUnder the conditions of image and facial image in chip and live face it is whether consistent；
(3.2) the machine-readable code identified according to OCR, distinguishes the information such as passport-type, belonging country, according to belonging country andPassport-type determines different anti-fake regions using the relative positioning technology of corresponding discrete cosine transform；
(3.3) anti-fake region segmentation is come out using sciagraphy；
(3.4) decorative pattern detection, fluorescent fiber detection, data field detection and individual are used to the anti-fake region splitThe false distinguishings methods such as photo area detection, reflect to 13, machine-readable code area, passport NO., anti-fake decorative pattern etc. stable general anti-fake pointsPuppet processing.
In this example, step (5) specifically:
(5.1) Face datection is carried out using YOLO2 algorithm to facial image in real-time face image, chip and calibrated, frameSelect facial image；
(5.2) facial image is inputted into ResNet, extracts image feature vector；
(5.4) COS distance for calculating feature vector, is determined according to the reasonable threshold value of setting, is then tested higher than threshold valueCard passes through, and does not otherwise pass through.
It is worth noting that of the invention it is critical that the OCR identification technology of use can be compatible with uniline and multirow is machine-readableCode situation, and chip recognition can be compatible with standards such as " ICA09303 standard Part1-4 ", greatly simplify system, improve and beSystem stability；In addition it is trueer than counterpart method verifying certificate itself that image false distinguishing method, image shallow-layer feature, machine-readable code are usedPuppet, while using YOLO2 algorithm+ResNet and holder and certificate owner's consistency are determined, it has been obviously improved thisThe accuracy and safety of authentication system；This system integrated level is high, and easy to operate, authentication is high-efficient, can be very bigSaving authentication time, improve verification efficiency.
As it will be easily appreciated by one skilled in the art that the foregoing is merely illustrative of the preferred embodiments of the present invention, not toThe limitation present invention, any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should all includeWithin protection scope of the present invention.