CN105956515B - Solid-EO-1 hyperion face identification method based on pole imaging - Google Patents
Solid-EO-1 hyperion face identification method based on pole imaging Download PDFInfo
- Publication number
- CN105956515B CN105956515B CN201610247977.2A CN201610247977A CN105956515B CN 105956515 B CN105956515 B CN 105956515B CN 201610247977 A CN201610247977 A CN 201610247977A CN 105956515 B CN105956515 B CN 105956515B
- Authority
- CN
- China
- Prior art keywords
- face
- pole
- picture
- obtains
- image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/168—Feature extraction; Face representation
- G06V40/171—Local features and components; Facial parts ; Occluding parts, e.g. glasses; Geometrical relationships
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/172—Classification, e.g. identification
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- General Health & Medical Sciences (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Processing (AREA)
Abstract
The invention proposes a kind of solid-EO-1 hyperion recognition of face methods based on pole imaging, for solving recognition result stability difference present in existing multispectral face identification method and the low technical problem of recognition efficiency, include the following steps, 1, utilize auroral spectrum system acquisition face information；2, according to face information in two-dimentional pole figure picture to face carry out stereo reconstruction；3, region unit division is carried out to two-dimension human face, and region unit division is carried out to stereoscopic face and face spectral information according to the division result；4, the multi-feature vector of region unit is extracted；5, sample database data and human face data to be identified are obtained to step 4 using step 1 to multiple faces and face to be identified respectively；6, training sample database data obtain classifier；7, human face data to be identified is input in classifier, and identifies face.The present invention has the characteristics that stability is good and recognition efficiency is high, can be used for the identity authentication of financial transaction, safe defence, clearance safety check etc..
Description
Technical field
The invention belongs to biometrics identification technology fields, are related to a kind of face identification method, and in particular to one kind is based on
Solid-EO-1 hyperion face identification method of pole imaging, can be used for the identity authentication of financial transaction, safe defence, clearance safety check etc..
Background technique
Live the mankind that 6,700,000,000 appearances are different, the colour of skin is different in the world, the face biological characteristic important as one,
There is critical role in interpersonal exchange.In today's society, face has become a kind of important for identity authentication
Feature.Face identification system can be divided into three kinds by function: right towards security type, management supervision class and multimedia recreation class
The requirement of recognition accuracy also successively reduces.It is reported that thering is government's welfare fund of billions of dollars illegally to be emitted every year in the U.S.
Neck, in addition, also occurring asking including a variety of social safeties that the attack of terrorism threatens because lacking perfect identification system
Topic.The tragedy of the continuous attack of terrorism occurred in recent years examines the border control of various countries, immigration while reinforcing security implementations
It examines ability and has also beaten alarm bell.Therefore, more accurate, efficient face recognition technology of the research towards security type has clear
, urgent research significance.
Face recognition process is generally described as, and is given a static state or dynamic facial image, is utilized existing people
Face database determines the identity of one or more people in image.It mainly solves three tasks: authentication, identity are known
Not, monitoring.Authentication is to judge whether face in image is identity that he is stated, it is only necessary to by the facial image of input
It is one-to-one comparison compared with the template image of the identity stored in database pair；Identification is to utilize face number
The identity that face in image is determined according to library needs to compare identity template image all in the facial image of input and library
To and provide similarity, be one-to-many comparison to differentiate that input face is which identity (similarity highest) in library.Monitoring
Refer in monitoring range find face and track and split, and judge the face whether on watch-list, if it is
If, further confirm their identity.
The basic procedure of recognition of face mainly include data acquisition, Face datection, face registration, face characteristic extract and
The processes such as recognition of face.Facial image or video are obtained using imaging sensor such as CCD or CMOS etc., figure is improved by pretreatment
Then image quality amount carries out Face datection, Attitude estimation and positioning feature point, i.e., is partitioned into interested face area from image
Domain is registrated human face region and is normalized, then carries out feature extraction to the facial image after normalization, is input to classifier
In in face database template image feature vector carry out matching primitives, finally export recognition result.
Recognition of face can be divided into two major classes from the method for description face: based on geometrical characteristic face representation method and base
In the face representation method of texture.Recognition methods based on face geometry is believed using feature of the face in structure composition
Breath, seeks edge and profile as feature from face two dimensional image, and then in the way of aspect ratio pair between judging characteristic
Similarity.Substantially, from the subset that the information extracted in two-dimension picture is after being lost true three-dimension face depth dimension information, and
And the resemblance of face and unstable, it is easy to be illuminated by the light the influence with expression shape change, therefore based on the several of two-dimension human face image
What structural recognition method stability is not high, can even fail under the shadow effect of large area.To over the past two years based on three-dimensional
Or the face identification method of depth geometry feature is suggested, and is obtained by technologies such as laser scanning, three-dimensional imagings and is believed with depth
The human face three-dimensional model of breath, and effective feature is extracted from model, such as form using information such as forehead curvature, nose height
The feature vector of three-dimensional face is described, and then is identified by the method for template matching.The introducing of Spatial Multi-Dimensional degree information at
The influence for solving face characteristic and being illuminated by the light variation of function, makes recognition accuracy reach 90% or more.
On the other hand, researchers both domestic and external have also carried out the largely face identification method based on texture information and have ground
Study carefully.Such method from the spectrum of face tie up information, using spectrum face surface be distributed regularity and stationarity as
The feature of face describes, and then completes recognition of face using the method for pattern-recognition.Act on face light be divided into actively and by
Dynamic light source, the wavelength band of light source light spectrum also expand to near-infrared from visible light, and the resolving range of spectrum is from polychrome, multispectral
EO-1 hyperion is arrived again.The development for relying on the theory and technology of statistical-simulation spectrometry emerges people largely based on texture information
Face recognizer." eigenface (Eignface) " that nineteen nineties are suggested is representative recognition methods.
So far " eigenface " method is still to be used to evaluate and test one of benchmark algorithm of new algorithm in field of face identification, is applied to each
The recognition of face of spectral band.Rise and fast-developing, the methods of neural network, deep learning quilt with machine learning techniques
It is applied in Face datection and is successful, especially most pushing away with high-performance computer hardware over the past two years
Extensively, artificial intelligence technology has great breakthrough, using machine learning to obtaining good classifier after extensive sample training,
And then facial image or video are detected and identified.Although the application development of " machine learning " is like a raging fire,
In processing human face detection and tracing problem or exist obvious insufficient: a large amount of Training data being needed to be used as support,
To improve the generalization ability of system；It is seldom with complete tag information in current existing face database, limit big portion
Divide supervised learning algorithm；Model training time and resource consumption are very big, such as want adjusting training strategy and parameter, the week for needing to grow very much
Phase.
Currently, face identification method used in researcher from spatially via low-dimensional to higher dimensional space develop, from light
Develop via monochrome greyscale image to broadband or even height (more) spectrum in spectrum.On the one hand, pass through laser scanning, three-dimensional imaging
Etc. technologies obtain the human face three-dimensional model for having depth information, and effective geometrical characteristic is extracted from model, Spatial Dimension letter
The introducing of breath solves face characteristic and is illuminated by the light influence with expression shape change.On the other hand, glistening light of waves piece or Amici prism are utilized
Optical instrument obtains the spectral information of face, and the spectral signature of face part, light are extracted from the face spectral information obtained
The extension of spectrum dimensional information solves recognition of face to be influenced by shade and partial occlusion factor.For example, Chinese patent application, is awarded
Power notification number is CN102831400B, the patent of invention of entitled " a kind of multispectral face identification method and its system ", the hair
It is bright include spectrum imaging system, color camera, face recognition module, data memory module, Central Control Module and spectrometer, it is more
Spectrum imaging system exports the face image data of shooting to face recognition module, and face recognition module stores mould according to data
The information of standard faces database in block is identified, then exports the result of identification, Central Control Module control
The image capture of multi-optical spectrum imaging system and the identification of face recognition module.Although the invention can provide bright in identification process
Aobvious between class distance, but there are still insufficient at two first is that due to using multi-optical spectrum imaging system to acquire face image data, nothing
Method obtains the stereochemical structure information of face, is easy recognition result by illumination variation, the influence that coats and block, causes to identify
As a result stability is poor, and in addition the invention reduces recognition of face due to needing to be manually operated when intercepting human face region sample
Efficiency.
Summary of the invention
It is an object of the invention to overcome the problems of the above-mentioned prior art, propose a kind of based on the vertical of pole imaging
Body-EO-1 hyperion face identification method, it is poor for solving recognition result stability present in existing multispectral face identification method
The technical problem low with recognition efficiency.
To achieve the above object, the technical solution that the present invention takes, includes the following steps:
Step 1, human face data is acquired using Hyperspectral imager, obtains face in the two-dimentional pole figure picture of each spectral band
And the corresponding spectral information of pixel on the two dimension pole figure picture；
Step 2, it is corresponding that three high spectral bands of quality score are filtered out from the two-dimentional pole figure picture of obtained face
Pole figure picture merges the three pole figure pictures filtered out, and carries out affine transformation to fused pole figure picture, obtains face three
The image of a different points of view；
Step 3, using multi-viewpoint three-dimensional matching process, three-dimensional weight is carried out to the image for obtaining three different points of view of face
Structure obtains the three-dimensional structure of face；
Step 4, the full two-dimensional face in the fused pole figure picture is divided, obtains face with different points
The two-dimentional regional area block of class label；
Step 5, the two-dimentional regional area block that different classifications label is had according to obtained face, to the three-dimensional of the face
The corresponding spectral information of pixel carries out the division of regional area block in structure and the two-dimentional pole figure picture, obtains comprising three-dimensional structure
With multiple regional area blocks with tag along sort of spectral information；
Step 6, to each regional area block in obtained multiple regional area blocks, calculate the feature of three-dimensional structure to
Amount, while the feature vector of spectral information is extracted, and merge to the two feature vectors, multiple have for obtaining face is divided
The Local synthesis feature vector of class label；
Step 7, step 6 is arrived using step 1 to multiple sample faces, obtains multiple have comprising multiple sample faces and divides
The Face Sample Storehouse of the Local synthesis feature vector of class label；
Step 8, obtained Face Sample Storehouse data are input in linear SVM model and are trained, obtain needle
To the classifier of different target type；
Step 9, step 6 is arrived using step 1 to face to be identified, obtains the multiple with tag along sort of face to be identified
Local synthesis feature vector；
Step 10, it by multiple Local synthesis feature vectors with tag along sort of obtained face to be identified, is input to
Classify in the classifier, obtains target type belonging to the multiple regional area blocks of face to be identified, and according to wait know
Whether others belongs to same target type by multiple regional area blocks of face, judges whether the face is consistent with sample database.
Compared with the prior art, the invention has the following advantages:
1, the present invention is due to being utilized Hyperspectral imager when acquiring human face data, and according to obtaining two-dimentional pole figure picture
Three-dimensionalreconstruction is carried out to face, acquisition while can be realized three-dimensional structure and spectral information in face characteristic, with the prior art
It is middle to acquire human face data using multi-optical spectrum imaging system, and the two-dimensional geometry feature for obtaining face is compared with spectral information, identification
As a result the influence for not being illuminated by the light variation, coating and blocking has higher stability.
2, the present invention is used due to when dividing full two-dimensional face, not needing manually to participate in the prior art
The method of Manual interception human face region sample compare, improve the efficiency of recognition of face.
Detailed description of the invention
Fig. 1 is flow diagram of the invention；
Fig. 2 is that the Hyperspectral imager in the present invention acquires schematic diagram to human face data；
Fig. 3 is that Multi-visual point image schematic diagram is imaged in pole；
Fig. 4 is the multiple regional area block schematic diagrames of face in the present invention.
Specific embodiment
Below in conjunction with the drawings and specific embodiments, the invention will be further described.
Referring to Fig.1, solid-EO-1 hyperion face identification method based on pole imaging, includes the following steps:
Step 1, human face data is acquired using Hyperspectral imager, as shown in Fig. 2, obtaining face in each spectral band
The corresponding spectral information of pixel on two-dimentional pole figure picture and the two dimension pole figure picture, the human face data for acquiring each spectral band can be with
More fully reflect face characteristic, keeps recognition result more accurate；
Step 2, it is corresponding that three high spectral bands of quality score are filtered out from the two-dimentional pole figure picture of the face of acquisition
Pole figure picture merges the three pole figure pictures filtered out, and carries out affine transformation to fused pole figure picture, in affine transformation
When extract radial line (polar curve) after fusion in pole figure picture, and it is horizontal stack into a new image, pole according to Fig.3, at
As new images are divided into three parts by Multi-visual point image principle in proportion in the horizontal direction, three views of hyperspectral imager are respectively corresponded
Point, to obtain the image of three different points of view of face；
The high corresponding pole figure picture of three spectral bands of quality score is filtered out to realize in accordance with the following steps:
Step 2a using Y-PSNR function (Peak Signal-to-Noise Ratio, PSNR) and is based on structure
The evaluation function (Structural Similarity Image Measurement, SSIM) of similarity, to each spectrum wave of face
The pole figure picture of section carries out quality evaluation, obtains the Y-PSNR of each spectral band pole figure picture of face and based on structural similarity
Scoring, Y-PSNR function can efficiently obtain the less wave band of noise in high spectrum image, and be based on structural similarity
Evaluation function from the brightness of image, the content quality of three factors evaluation high spectrum images of contrast and structure, fusion mass
The purpose of higher band is to obtain the apparent single width two dimensional image of characteristic point, reconstructs for depth below and provide sufficient characteristic point
For Stereo matching；
Step 2b, Y-PSNR to each spectral band pole figure picture of obtained face and based on the scoring of structural similarity
Quantify simultaneously linear superposition, obtains the quality score of each spectral band pole figure picture, and screen from each spectral band pole figure picture
The corresponding pole figure picture of the high spectral band of at least two quality scores out, the present invention filters out three pole figure pictures, if filtering out
The effect of high quality can be not achieved in more images, and blending image too unification can be made by screening very little image then, cannot be complete
Description human face data；
Step 3, using multi-viewpoint three-dimensional matching process, using GPU parallel algorithm to three different points of view of obtained face
Image carry out stereo reconstruction, obtain the corresponding depth information of pixel in two-dimension human face to get arrive face three-dimensional structure；
Step 4, since the circular central region of face two dimension pole figure picture is the true view that face passes through hyperspectral imager
The picture of point, which can accurately show two-dimension human face, can be with by dividing the full two-dimensional facial image in the region
More accurate partitioning standards are provided to divide face three-dimensional structure and spectral information below, so to the fused two dimension
Full two-dimensional face in pole figure picture is divided, and the two-dimentional regional area block that face has different classifications label is obtained；
Divide the realization in accordance with the following steps of full two-dimensional face:
The circular image comprising complete face in pole figure picture after obtained face fusion is split, is obtained by step 4a
To complete two-dimension human face image；
Step 4b is sharpened processing to obtained complete two-dimension human face image, and detects two-dimentional people after Edge contrast
The marginal point of face image obtains two-dimension human face image marginal point；
Step 4c clusters obtained two-dimension human face image marginal point, obtain multiple in two-dimension human face image include
The classification of marginal point, cluster here are to divide classification by the difference of Different Organs on face, due to eyes on face,
The organs such as nose, lip are to the notable difference of spectral reflectivity, so being easy to divide region unit, this hair by facial image
It is bright middle face to be divided into 9 classifications；
Step 4d, the center position of multiple classifications comprising marginal point in the two-dimension human face image calculated separately,
And its tag along sort and area size are determined according to the position of central point of all categories, for example, for a class center point most
The classification on the left side, the present invention give this classification eyes label, and set its category regions shape as rectangle, and size is positive
The average length and width of normal human eye obtains the two-dimentional regional area block that face has different classifications label；
Step 5, the two-dimentional regional area block that different classifications label is had according to obtained face, to the three-dimensional of the face
The corresponding spectrum of pixel in structure (two dimensional image combination depth information forms face steric information) and the two-dimentional pole figure picture
Information carries out the division of regional area block, and the spectral information is when dividing only to the complete face of face pole figure inconocenter border circular areas
It is divided, obtains multiple regional area blocks with tag along sort comprising three-dimensional structure and spectral information, as shown in Figure 4；
Step 6, it in multiple regional area blocks of obtained face, is formed using the average height and gradient of regional area
The feature vector of three-dimensional structure, while son sky is tieed up using the spectrum of each region unit of HySime algorithm rapidly extracting based on GPU
Between, the feature vector of spectral information is obtained, and merge to two feature vectors of each regional area block, obtains face
Multiple Local synthesis feature vectors with tag along sort；
Step 7, step 6 is arrived using step 1 to multiple sample faces, obtains multiple have comprising multiple sample faces and divides
The Face Sample Storehouse of the Local synthesis feature vector of class label；
Step 8, the data volume of face three-dimensional structure and the Local synthesis feature vector of spectral information is big, manual sort region
Block is time-consuming, laborious, unrealistic, and the present invention uses linear SVM model, will there is the office of tag along sort in Face Sample Storehouse
Portion's comprehensive characteristics data, are input in linear SVM model and are trained, and support vector machines is substantially two points a kind of
Class model finds optimal segmentation hyperplane in two categories feature space, and target scene is divided in general application
Local feature be often more than two kinds, it is therefore desirable to multiple classification processing is carried out using " one-to-one " mode of support vector machines,
The classifier for different target types is obtained, which determines the mesh of the Local synthesis feature vector with tag along sort
Mark type, it can determine that the region units such as eyes or the nose of input belong to which of sample database face；
Step 9, step 6 is arrived using step 1 to face to be identified, obtains the multiple with tag along sort of face to be identified
Local synthesis feature vector；
Step 10, it by multiple Local synthesis feature vectors with tag along sort of obtained face to be identified, is input to
Classify in the classifier, target type belonging to the multiple regional area blocks of face to be identified is obtained, if multiple offices
Portion's region unit is all classified as the same target type, then recognition of face to be identified be the target type, otherwise face to be identified with
Sample database is not inconsistent, that is, is not belonging to sample database.
Referring to Fig. 2, the system that the present invention uses is by hyperspectral imager, rotary table internal reflector, scanning turntable, synchronous bracket
It is formed with fixture.Hyperspectral imager light spectrum image-forming range covers visible light-near infrared band from 400nm to 1000nm, by
Quite high reflectivity (r > 0.96) is remain in visible light-near-infrared spectrum range in metallic silver, is conducive to imaging spectral
Instrument is reflected by internally reflective the radial spectra collection of mirror, therefore selects the Coating Materials of silver-colored (Ag) as rotary table reflecting mirror inner wall.Due to
The spatial resolution of hyperspectral imager scan line is 1632 pixels, and spectrum is averaged minimum resolution as 0.64nm, that is, has 936
A band, so high spatially and spectrally resolution requirement scanning process bracket have good stability, it is contemplated that the present invention wants
Imager and internal reflector are combined into composition " high light spectrum image-forming " acquisition system, transformation is needed to have scanning turntable, is led to
Processing synchronous bracket and fixture are crossed to control the position of internal reflector, and the control mode that stepper motor is transformed makes high light spectrum image-forming
Instrument is synchronous with the holding of the scanning process of internal reflector.After hyperspectral imager and internal reflector cascade, need to utilize laser pair
Imaging optical path is calibrated and is tested.And the image-forming principle being imaged according to pole, determine the visual field of entire " Hyperspectral imager "
Angle and EFFECTIVE RANGE.
There are certain positional relationship between rotary table internal reflector and hyperspectral imager in Hyperspectral imager, in figure
R represents port radius on the left of internal reflector, and θ represents imager lens centre to the half of the subtended angle of internal reflector left-hand end mouth, β
Represent internal reflector lens barrel tangent plane angle with horizontal plane, due to face to be guaranteed only carried out in interior reflecting mirror it is Polaroid,
So lens barrel horizontal length l meets some requirements, l must meet l=2rcos (β) cos (θ -2 β) csc (θ -3 β).
Referring to Fig. 3, light forms multi-view image after truncated cones internal reflector in hyperspectral imager, according to
Length, subtended angle and the front and back radius of pre-designed reflecting mirror limit a certain range of light and only pass through primary event
Reach imaging focal plane, to complete the imaging of " an object point multiple views ", the space in operating distance be divided into three mesh areas,
Binocular area and other blind areas.The position of face is within three mesh areas, which refers to that any point can exist simultaneously on face
Radially three viewpoints are imaged, and 3 points of A, B, C in these three viewpoint, that is, figures, point B is the light of hyperspectral imager imaging len
The heart, point A and point C be respectively point B in radial directions about internal reflector at picture.Figure midpoint P is target scene space Zhong Chu
In three mesh areas of pole imaging system a bit, after internal reflector, be respectively formed on the focal plane of imaging spectrometer A,
B, the picture of tri- viewpoints of C, B point are projected in the central circular of pole figure picture, A and C point is projected on the peripheral annulus of pole figure picture
In region.
Referring to Fig. 4, face is divided into multiple regional area blocks such as eyebrow, eyes, nose, lip, each regional area
It all include three-dimensional structure (two dimensional image combination depth information) and spectral information (Spectral dimension evidence) in block, wherein spectral information
Comprising each point in complete face to the reflectivity of each spectral band.
Claims (3)
1. a kind of solid-EO-1 hyperion face identification method based on pole imaging, which comprises the steps of:
(1) human face data is acquired using Hyperspectral imager, obtain face in the two-dimentional pole figure picture of each spectral band and be somebody's turn to do
The corresponding spectral information of pixel on two-dimentional pole figure picture；
(2) the high corresponding pole figure picture of three spectral bands of quality score is filtered out from the two-dimentional pole figure picture of obtained face,
The three pole figure pictures filtered out are merged, and affine transformation is carried out to fused pole figure picture, obtain three differences of face
The image of viewpoint；
(3) multi-viewpoint three-dimensional matching process is utilized, stereo reconstruction is carried out to the image for obtaining three different points of view of face, obtains people
The three-dimensional structure of face；
(4) the full two-dimensional face in the fused pole figure picture is divided, obtains face with different classifications label
Two-dimentional regional area block；
(5) the two-dimentional regional area block that different classifications label is had according to obtained face, three-dimensional structure to the face and
The corresponding spectral information of pixel carries out the division of regional area block on the two dimension pole figure picture, obtains comprising three-dimensional structure and spectrum
Multiple regional area blocks with tag along sort of information；
(6) to each regional area block in obtained multiple regional area blocks, the feature vector of three-dimensional structure is calculated, is mentioned simultaneously
The feature vector of spectral information is taken, and the two feature vectors are merged, obtains the multiple with tag along sort of face
Local synthesis feature vector；
(7) step (6) are arrived using step (1) to multiple sample faces, obtains multiple with classification comprising multiple sample faces
The Face Sample Storehouse of the Local synthesis feature vector of label；
(8) obtained Face Sample Storehouse data are input in linear SVM model and are trained, obtained for difference
The classifier of target type；
(9) step (6) are arrived using step (1) to face to be identified, obtains multiple offices with tag along sort of face to be identified
Portion's multi-feature vector；
(10) by multiple Local synthesis feature vectors with tag along sort of obtained face to be identified, it is input to point
Classify in class device, obtains target type belonging to the multiple regional area blocks of face to be identified, and according to face to be identified
Whether multiple regional area blocks belong to same target type, judge whether the face is consistent with sample database.
2. solid-EO-1 hyperion face identification method according to claim 1 based on pole imaging, it is characterised in that: step
(2) the high corresponding pole figure picture of three spectral bands of quality score is filtered out described in, is realized in accordance with the following steps:
(2a) utilizes Y-PSNR function and the evaluation function based on structural similarity, to the pole figure picture of each spectral band of face
Quality evaluation is carried out, the Y-PSNR of each spectral band pole figure picture of face and the scoring based on structural similarity are obtained；
The scoring amount of progress of (2b) to the Y-PSNR of each spectral band pole figure picture of obtained face and based on structural similarity
Change simultaneously linear superposition, obtains the quality score of each spectral band pole figure picture, and be screened out from it three high spectrum of quality score
The corresponding pole figure picture of wave band.
3. solid-EO-1 hyperion face identification method according to claim 1 based on pole imaging, it is characterised in that: step
(4) division described in is realized in accordance with the following steps:
(4a) splits the circular image comprising complete face in pole figure picture after obtained face fusion, obtains complete
Two-dimension human face image；
(4b) is sharpened processing to obtained complete two-dimension human face image, and detects two-dimension human face image after Edge contrast
Marginal point obtains two-dimension human face image marginal point；
(4c) clusters obtained two-dimension human face image marginal point, obtains multiple in two-dimension human face image including marginal point
Classification；
The center position of multiple classifications comprising marginal point in the two-dimension human face image that (4d) is calculated separately, and according to each
The position of class center point determines its tag along sort and area size, obtains the two-dimentional partial zones that face has different classifications label
Domain block.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610247977.2A CN105956515B (en) | 2016-04-20 | 2016-04-20 | Solid-EO-1 hyperion face identification method based on pole imaging |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610247977.2A CN105956515B (en) | 2016-04-20 | 2016-04-20 | Solid-EO-1 hyperion face identification method based on pole imaging |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105956515A CN105956515A (en) | 2016-09-21 |
| CN105956515B true CN105956515B (en) | 2019-04-23 |
Family
ID=56917870
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610247977.2A Active CN105956515B (en) | 2016-04-20 | 2016-04-20 | Solid-EO-1 hyperion face identification method based on pole imaging |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105956515B (en) |
Families Citing this family (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106778681B (en) * | 2017-01-11 | 2019-12-17 | 厦门中控智慧信息技术有限公司 | face recognition method and face recognition equipment |
| CN106909944B (en) * | 2017-03-01 | 2020-05-26 | 西北大学 | Face picture clustering method |
| CN106991378B (en) * | 2017-03-09 | 2020-02-04 | Oppo广东移动通信有限公司 | Depth-based face orientation detection method and device and electronic device |
| CN107239833B (en) * | 2017-05-12 | 2018-04-10 | 吉林师范大学 | A kind of information processing system and method based on deep learning |
| CN107194369A (en) * | 2017-06-09 | 2017-09-22 | 西安电子科技大学 | A kind of face identification method and face recognition device based on pole imaging volume camera |
| CN107748859B (en) * | 2017-08-10 | 2021-04-27 | 东南大学 | Three-dimensional face recognition method under local shielding based on radial lines |
| CN107832793B (en) * | 2017-11-08 | 2021-08-06 | 深圳大学 | Hyperspectral image classification method and system |
| CN107991283B (en) * | 2017-12-26 | 2023-09-22 | 同方威视技术股份有限公司 | Raman spectrum detection device and Raman spectrum detection method |
| CN110309692B (en) * | 2018-03-27 | 2023-06-02 | 杭州海康威视数字技术股份有限公司 | Face recognition method, device and system, and model training method and device |
| CN108805862B (en) * | 2018-05-02 | 2021-08-13 | 南京大学 | Label identification method based on improved structure similarity |
| CN108764328A (en) * | 2018-05-24 | 2018-11-06 | 广东工业大学 | The recognition methods of Terahertz image dangerous material, device, equipment and readable storage medium storing program for executing |
| CN110022462A (en) * | 2019-03-29 | 2019-07-16 | 江西理工大学 | A kind of safety defense monitoring system based on multispectral camera |
| CN111368644B (en) * | 2020-02-14 | 2024-01-05 | 深圳市商汤科技有限公司 | Image processing method, device, electronic equipment and storage medium |
| CN111368681B (en) * | 2020-02-27 | 2023-09-01 | 深圳数联天下智能科技有限公司 | Living body screening method, device, equipment and storage medium based on multi-point positioning |
| CN115641649A (en) * | 2021-07-19 | 2023-01-24 | 清华大学 | Face recognition method and system |
| CN116168346B (en) * | 2023-04-21 | 2023-08-18 | 独数一智(北京)信息科技有限公司 | Remote accompanying-reading monitoring system based on student behavior capturing |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7469060B2 (en) * | 2004-11-12 | 2008-12-23 | Honeywell International Inc. | Infrared face detection and recognition system |
| CN101398886A (en) * | 2008-03-17 | 2009-04-01 | 杭州大清智能技术开发有限公司 | Rapid three-dimensional face identification method based on bi-eye passiveness stereo vision |
| CN102324022A (en) * | 2011-09-05 | 2012-01-18 | 辽宁工程技术大学 | Composite gradient vector-based face recognition method |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2014176485A1 (en) * | 2013-04-26 | 2014-10-30 | West Virginia High Technology Consortium Foundation, Inc. | Facial recognition method and apparatus |
-
2016
- 2016-04-20 CN CN201610247977.2A patent/CN105956515B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7469060B2 (en) * | 2004-11-12 | 2008-12-23 | Honeywell International Inc. | Infrared face detection and recognition system |
| CN101398886A (en) * | 2008-03-17 | 2009-04-01 | 杭州大清智能技术开发有限公司 | Rapid three-dimensional face identification method based on bi-eye passiveness stereo vision |
| CN102324022A (en) * | 2011-09-05 | 2012-01-18 | 辽宁工程技术大学 | Composite gradient vector-based face recognition method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105956515A (en) | 2016-09-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105956515B (en) | Solid-EO-1 hyperion face identification method based on pole imaging | |
| US10204262B2 (en) | Infrared imaging recognition enhanced by 3D verification | |
| CN105046219B (en) | A kind of face identification system | |
| KR101030613B1 (en) | The Region of Interest and Cognitive Information Acquisition Method at the Eye Image | |
| CN101999900B (en) | Living body detecting method and system applied to human face recognition | |
| CN103632132B (en) | Face detection and recognition method based on skin color segmentation and template matching | |
| Raghavendra et al. | Exploring the usefulness of light field cameras for biometrics: An empirical study on face and iris recognition | |
| CN110210276A (en) | A kind of motion track acquisition methods and its equipment, storage medium, terminal | |
| CN106845328B (en) | A kind of Intelligent human-face recognition methods and system based on dual camera | |
| Friedrich et al. | Seeing people in the dark: Face recognition in infrared images | |
| Medioni et al. | Identifying noncooperative subjects at a distance using face images and inferred three-dimensional face models | |
| CN105426843B (en) | Single-lens lower palm vein and palm print image acquisition device and image enhancement and segmentation method | |
| CN106295600A (en) | Driver status real-time detection method and device | |
| CN104597057B (en) | A kind of column Diode facets defect detecting device based on machine vision | |
| CN103116749A (en) | Near-infrared face identification method based on self-built image library | |
| CN101739546A (en) | Image cross reconstruction-based single-sample registered image face recognition method | |
| CN109684965A (en) | A kind of face identification system based near infrared imaging and deep learning | |
| CN108416291A (en) | Face datection recognition methods, device and system | |
| CN110472495B (en) | Deep learning face recognition method based on graphic reasoning global features | |
| Sun et al. | Context based face spoofing detection using active near-infrared images | |
| Bogo et al. | Automated detection of new or evolving melanocytic lesions using a 3D body model | |
| RU2488882C1 (en) | Method of personal identification | |
| Campadelli et al. | Localization of facial features and fiducial points | |
| Pandeeswari et al. | Fast evolutionary algorithm based identifying surgically distorted face for surveillance application | |
| KR102395866B1 (en) | Method and apparatus for object recognition and detection of camera images using machine learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |