A kind of face identification method of DNR dynamic noise reductionTechnical field
The present invention relates to three-dimensional face identification technology field, more particularly, to a kind of recognition of face side of DNR dynamic noise reductionMethod.
Background technology
Face identification system, using face recognition technology as core, is an emerging biological identification technology, is the current worldThe high-quality precision and sophisticated technology of sciemtifec and technical sphere tackling key problem.Not reproducible, collection is convenient, do not need the cooperation of one be shooted because having for face so thatFace identification system has a wide range of applications.Nowadays, face recognition technology has been widely used in the safety-security areas such as gate inhibition.
Face recognition technology is mainly or by two dimensional image identification method, and its method is according to two dimensional surface face silhouetteOr certain visual angle is extracted and recognizes face characteristic.The poor reliability of this method, the shadow of facial pose, illumination by identified personSound is larger.Accordingly, three-dimensional face identification technology degree of accuracy height, strong adaptability, attack tolerant be strong, anti-fraudulent is strong, than twoThe face recognition technology for tieing up image is relatively reliable.
However, existing three-dimensional face identification technology concern is primarily with how to facial crucial recognition site modeling andOvercome the influence that light is brought.But in fact, the particularity of the environment due to gate inhibition's application, identified person is likely to be at severe dayUnder gas system, such as drenching with rain, snow, haze, cause the unclear of identified person's local feature, and then had influence on three-dimensionalDegree of accuracy during face recognition.
The content of the invention
In order to overcome adverse effect of the meteorological condition to three-dimensional face identification technology in recognition accuracy, the present invention is providedA kind of face identification method of DNR dynamic noise reduction, including：
(1) voice signal of people to be identified is gathered, and the identity information that people to be identified claims is obtained based on this voice signal；
(2) whereabouts information that the natural environment residing for people to be identified rains or snow is gathered, to obtain people front to be identifiedRain or snow density information；
(3) the actual three dimensional face image information of people to be identified is gathered；
(4) identity information and the actual three-dimensional surface of people to be identified claimed according to the whereabouts information of rain or snow, people to be identifiedPortion's image information, it is determined that the initial position message of three-dimensional identification；
(5) according to initial position message, the three dimensional face information of people to be identified is recognized；
(6) according to the facial information of people to be identified, the actual identity information of people to be identified is determined.
Further, the step (1) includes：
(11) voice messaging of prompting problem is provided to people to be identified, the sound letter of people to be identified in the given time is obtainedBreath；
(12) vocal print of the acoustic information of people to be identified is obtained；
(13) vocal print of the acoustic information of people to be identified is compared with default voiceprint set, according to comparing knotFruit determines the identity information that people to be identified claims.
Further, the prompting problem is the prompting problem provided at random.
Further, step (2) includes：
(21) locus in the acoustic information source of people to be identified is determined；
(22) rain estimation of rain or snow is gathered in people underfooting to be identified；
(23) when rain estimation exceedes default rain estimation threshold value, grid type screening is carried out in people's overhead to be identifiedGear, in order to which the rain estimation for dropping to the rain on the number of people to be identified or snow is controlled below default rain estimation threshold value；
(24) density information of the positive rain of people to be identified or snow is gathered.
Further, the step (24) includes：
(241) direct picture of multiple people to be identified is gathered between people to be identified and front scan camera；
(242) identity information claimed according to people to be identified, determines the frontal face image that people to be identified claims；
(243) in presupposition analysis region, it is determined that the direct picture of multiple people to be identified and the voice to be identifiedSharpness information c between the frontal face image of titlej, the presupposition analysis region is centered on default collection point, verticallyDirection, the region that radius is preset length R, the space Z-direction coordinate of the default collection point are predetermined for the people crown to be identifiedAt distance, and when progress grid type is blocked, the pre-determined distance, which is less than, to be carried out when grid type is blocked apart from the people crown to be identifiedDistance, the space X direction of the collection point and Y-direction coordinate are that X-coordinate and Y at the locus that the acoustic information is originated are satMark, the sharpness information cjBased on the physiological characteristic at facial each position of people to be identified, j is each position of faceQuantity and j >=5.
Further, the step (3) is at the locus that the acoustic information is originated including gathering people to be identifiedThe actual three dimensional face image information of picture centre.
Further, the step (4) includes：
(411) identity information claimed according to people to be identified, determines the frontal face image that people to be identified claims；
(412) sharpness information is added on the frontal face image, obtained with reference to identification image；
(413) it is compared described with reference to identification image and the actual three dimensional face image information of people to be identified, it is determined thatFacial zone in actual three dimensional face image information；
(414) in the facial zone in the actual three dimensional face image information, the actual three dimensional face figure is determinedMatch information between the frontal face image claimed as information and the people to be identified, the match information includes i with faceEach position physiological characteristic based on deformed region, wherein i be more than 10；
(415) the deformation coefficient respectively α of the i deformed regions is seti, according to using each deformed region inThe heart, deformation analysis radius r calculate for the following formula in multiple neighborhoods of radius, it is determined that calculating, obtained minimum value is corresponding, makeCentered on deformed region Amin, and the maximum value that calculating is obtained is corresponding, deformed region A as centermax：
Distance between deformed regions of the wherein r for some deformed region and around it, the deformation coefficient αiFor withIn each physiological characteristic region based on the physiological characteristic at each position for the frontal face image that people to be identified claims, tableShow the number that the pixel of the profile of the physiological characteristic occurs in the relevant position with reference to identification image；
(416) using T as the cycle, to the deformed region AminIt is the actual three dimensional face figure of picture centre at where centerAs information carries out p secondary acquisition, two-dimentional face-image, and the two-dimensional surface being extracted within each cycle described in determination are therefrom extractedSharpness information between the frontal face image that portion's image and the people to be identified claimThe sharpness informationTo treatBased on the physiological characteristic at facial each position for recognizing people, quantity and j >=5 of the j for each position of face；
(417) with each secondary acquisition during obtainBuild definition discrimination matrix, each behavior of the matrixThe facial corresponding definition in each position of the people to be identified obtained during one secondary acquisition, each be classified as of the matrix is treatedSome facial position of identification people carries out the definition obtained during each secondary acquisition, i.e.,：First during secondary acquisitionThe facial corresponding definition in each position of the secondary people to be identified collected is the first row, and what is collected for the second time is to be identifiedThe facial corresponding definition in each position of people is the second row, by that analogy；
(418) the variance D of each row of the definition discrimination matrix is calculatedq, wherein q is the definition discrimination matrixColumns；
(419) variance D in the definition discrimination matrix is removedqMaximum sharpness informationThe row at place, is passed throughCross the definition discrimination matrix of processing；
(420) for each row in the treated definition discrimination matrix, following after-treatment is carried out successively：RootAccording to centered on each deformed region, the deformation analysis radius r be that following formula in multiple neighborhoods of radius is calculated, it is determined that meterObtained minimum value：
(421) it is the minimum value obtained in the after-treatment is corresponding as in the geometry of the deformed region at centerThe heart is used as initial position message as initial position, the positional information of the geometric center.
Further, the rain estimation is the criteria for classifying according to precipitation in meteorology the preceding paragraph time.
Further, methods described also includes the reality of the identity information and people to be identified claimed according to the people to be identifiedIdentity information, determines the safety precaution grade of gate control system.
The beneficial effects of the invention are as follows：
(1) can improve in the rain or snow in recognition of face the degree of accuracy；
(2) can based on the multiple analysis to rain or snow, dynamically determine rather than still specified three-dimensional recognize when justBeginning position, compared with prior art in the commonly used position such as " nose ", " place between the eyebrows " that is directly defaulted as there is more extensive be applicableProperty and reliability, can avoid because the people to be identified physiological defect of itself causes recognition failures.
Brief description of the drawings
Fig. 1 shows the flow chart of the face identification method of the DNR dynamic noise reduction according to the present invention.
Embodiment
As shown in figure 1, the invention provides a kind of face identification method of DNR dynamic noise reduction, including：
(1) voice signal of people to be identified is gathered, and the identity information that people to be identified claims is obtained based on this voice signal；
(2) whereabouts information that the natural environment residing for people to be identified rains or snow is gathered, to obtain people front to be identifiedRain or snow density information；
(3) the actual three dimensional face image information of people to be identified is gathered；
(4) identity information and the actual three-dimensional surface of people to be identified claimed according to the whereabouts information of rain or snow, people to be identifiedPortion's image information, it is determined that the initial position message of three-dimensional identification；
(5) according to initial position message, the three dimensional face information of people to be identified is recognized；
(6) according to the facial information of people to be identified, the actual identity information of people to be identified is determined.
The step (1) includes：
(11) voice messaging of prompting problem is provided to people to be identified, the sound letter of people to be identified in the given time is obtainedBreath；
(12) vocal print of the acoustic information of people to be identified is obtained；
(13) vocal print of the acoustic information of people to be identified is compared with default voiceprint set, according to comparing knotFruit determines the identity information that people to be identified claims.
The prompting problem is the prompting problem provided at random.
Step (2) includes：
(21) locus in the acoustic information source of people to be identified is determined；
(22) rain estimation of rain or snow is gathered in people underfooting to be identified；
(23) when rain estimation exceedes default rain estimation threshold value, grid type screening is carried out in people's overhead to be identifiedGear, in order to which the rain estimation for dropping to the rain on the number of people to be identified or snow is controlled below default rain estimation threshold value；
(24) density information of the positive rain of people to be identified or snow is gathered.
The step (24) includes：
(241) direct picture of multiple people to be identified is gathered between people to be identified and front scan camera；
(242) identity information claimed according to people to be identified, determines the frontal face image that people to be identified claims；
(243) in presupposition analysis region, it is determined that the direct picture of multiple people to be identified and the voice to be identifiedSharpness information c between the frontal face image of titlej, the presupposition analysis region is centered on default collection point, verticallyDirection, the region that radius is preset length R, the space Z-direction coordinate of the default collection point are predetermined for the people crown to be identifiedAt distance, and when progress grid type is blocked, the pre-determined distance, which is less than, to be carried out when grid type is blocked apart from the people crown to be identifiedDistance, the space X direction of the collection point and Y-direction coordinate are that X-coordinate and Y at the locus that the acoustic information is originated are satMark, the sharpness information cjBased on the physiological characteristic at facial each position of people to be identified, j is each position of faceQuantity and j >=5.
It is picture centre that the step (3), which includes gathering people to be identified at the locus that the acoustic information is originated,Actual three dimensional face image information.
The step (4) includes：
(411) identity information claimed according to people to be identified, determines the frontal face image that people to be identified claims；
(412) sharpness information is added on the frontal face image, obtained with reference to identification image；
(413) it is compared described with reference to identification image and the actual three dimensional face image information of people to be identified, it is determined thatFacial zone in actual three dimensional face image information；
(414) in the facial zone in the actual three dimensional face image information, the actual three dimensional face figure is determinedMatch information between the frontal face image claimed as information and the people to be identified, the match information includes i with faceEach position physiological characteristic based on deformed region, wherein i be more than 10；
(415) the deformation coefficient respectively α of the i deformed regions is seti, according to using each deformed region inThe heart, deformation analysis radius r calculate for the following formula in multiple neighborhoods of radius, it is determined that calculating, obtained minimum value is corresponding, makeCentered on deformed region Amin, and the maximum value that calculating is obtained is corresponding, deformed region A as centermax：
Distance between deformed regions of the wherein r for some deformed region and around it, the deformation coefficient αiFor withIn each physiological characteristic region based on the physiological characteristic at each position for the frontal face image that people to be identified claims, tableShow the number that the pixel of the profile of the physiological characteristic occurs in the relevant position with reference to identification image；
(416) using T as the cycle, to the deformed region AminIt is the actual three dimensional face figure of picture centre at where centerAs information carries out p secondary acquisition, two-dimentional face-image, and the two-dimensional surface being extracted within each cycle described in determination are therefrom extractedSharpness information between the frontal face image that portion's image and the people to be identified claimThe sharpness informationTo treatBased on the physiological characteristic at facial each position for recognizing people, quantity and j >=5 of the j for each position of face；
(417) with each secondary acquisition during obtainBuild definition discrimination matrix, each behavior of the matrixThe facial corresponding definition in each position of the people to be identified obtained during one secondary acquisition, each be classified as of the matrix is treatedSome facial position of identification people carries out the definition obtained during each secondary acquisition, i.e.,：First during secondary acquisitionThe facial corresponding definition in each position of the secondary people to be identified collected is the first row, and what is collected for the second time is to be identifiedThe facial corresponding definition in each position of people is the second row, by that analogy；
(418) the variance D of each row of the definition discrimination matrix is calculatedq, wherein q is the definition discrimination matrixColumns；
(419) variance D in the definition discrimination matrix is removedqMaximum sharpness informationThe row at place, is passed throughCross the definition discrimination matrix of processing；
(420) for each row in the treated definition discrimination matrix, following after-treatment is carried out successively：RootAccording to centered on each deformed region, the deformation analysis radius r be that following formula in multiple neighborhoods of radius is calculated, it is determined that meterObtained minimum value：
(421) it is the minimum value obtained in the after-treatment is corresponding as in the geometry of the deformed region at centerThe heart is used as initial position message as initial position, the positional information of the geometric center.
The rain estimation is the criteria for classifying according to precipitation in meteorology the preceding paragraph time.
Methods described also includes the actual identity information of the identity information and people to be identified claimed according to the people to be identified,Determine the safety precaution grade of gate control system.
NM three-dimensional face identification is comprised the concrete steps that after above-mentioned initial position message is determined in the application, canWith what is carried out using various models and algorithm of the prior art, and not this Applicant's Abstract graph main technical schemes, and this Shen is not influenceedImplementation please, will not be repeated here.
It is the purpose to illustrate for the narration that presently preferred embodiments of the present invention is made above, and is not intended to limit essence of the inventionIt is really disclosed form, it is possible based on teaching above or to make an amendment or change from embodiments of the invention study, embodiment is for explanation principle of the invention and allows those skilled in the art to be existed with various embodiments using the present inventionSelect and describe in practical application, technological thought of the invention attempts to be determined by claim and its equalization.