CN106203329B - A method of identity template is established based on eyebrow and carries out identification - Google Patents
A method of identity template is established based on eyebrow and carries out identification Download PDFInfo
- Publication number
- CN106203329B CN106203329B CN201610536639.0A CN201610536639A CN106203329B CN 106203329 B CN106203329 B CN 106203329B CN 201610536639 A CN201610536639 A CN 201610536639A CN 106203329 B CN106203329 B CN 106203329B
- Authority
- CN
- China
- Prior art keywords
- eyebrow
- image
- knowledge representation
- feature
- representation model
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- General Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of methods established identity template based on eyebrow and carry out identification, belong to identity identification technical field.The present invention includes: Zheng Lian topography of the acquisition containing eyebrow, is pre-processed after being partitioned into pure eyebrow image;The contour detecting that pure eyebrow image is carried out using the edge detection operator based on pseudo- ball, initial curve of the coarse eyebrow outline that will acquire as level set movements realize that accurate eyebrow outline is extracted using the Level Set Models dividing method corrected based on biased field;The knowledge representation model based on eyebrow feature set is constructed in a manner of feature vector, feature vector is made of shape feature, direction character and textural characteristics item, can carry out identification by eyebrow knowledge representation model.For the living things feature recognitions such as face and iris, the extraction of face outer profile and iris profile is carried out using the present invention, and calculates the characteristic value based on chamfered shape to construct knowledge representation model, therefore the present invention has preferable scalability.
Description
Technical field
The present invention relates to identity identification technical fields, establish identity mould based on eyebrow more specifically to one kind
Plate and the method for carrying out identification.
Background technique
Eyebrow recognition is the new biometrics identification technology of rising in recent years, and correlative study shows that eyebrow can be used as one
The independent biometric technology for identification of kind, such as discloses one kind application No. is 200610000048.8 patent and is based on
The authentication identifying method of eyebrow recognition.
But researcher is general to the research of the biological characteristics such as iris, fingerprint, eyes relatively broad at this stage, it is often right
Eyebrow concern is seldom.Eyebrow has distinct contour feature, and its shape has specificity and stability, efficiently uses eyebrow
Profile information or geometrical characteristic based on eyebrow information construct the knowledge representation model based on eyebrow, are applied to identification, mesh
Preceding correlative study and report is also seldom.
The feature extraction of eyebrow is the key technology of identification system, the eyebrow feature extracting method of mainstream include it is main at
Divide analysis, two-dimensional principal component analysis, hidden Markov model etc., but these methods are all based on the spy of the algebra in statistical significance
Sign is all to be projected image by a linear transformation to achieve the purpose that dimensionality reduction, and then the low-dimensional for obtaining eyebrow is special
Sign.As document " studies " (" computer application research " 2009 the 11st based on the eyebrow authentication system of wavelet transformation and SVM
Phase) it is the scheme that a kind of method by wavelet transformation extracts the textural characteristics of eyebrow.
Number of patent application 201310149305.4, the applying date are on April 26th, 2013, invention and created name are as follows: based on dilute
The eyebrow recognition method indicated is dredged, this application specifically comprises the following steps: to acquire original eyebrow image, be partitioned into after gray processing
Pure eyebrow image, and be normalized;Every pure eyebrow image array is connected by row and transposition is at column vector;Calculate to
Measure mean value;Calculate overall Scatter Matrix;Find out the preceding m characteristic value and feature vector of overall Scatter Matrix C;Calculate eyebrow sample
With the inner product of feature vector;Construct the vector approximation of eyebrow sample;Input test eyebrow image, and it is handled, it is tested
The vector approximation of eyebrow image;Calculate optimal sparse coefficient vector;Calculate the difference of test eyebrow and image;Determine test eyebrow institute
The classification of category.This application enhances the robustness of image recognition, it can also be used to the other biologicals feature such as face, fingerprint, iris
Identification, but the essence of this application eyebrow feature extraction is also the method using principal component analysis.
Based on analysis above it is found that at this stage eyebrow feature extracting method mainly study be all eyebrow algebraic characteristic or
Textural characteristics, these features are easy to be influenced by the quality and illumination variation of eyebrow image itself, so as to cause to know
Not Shi Xiao the problem of, application on there are certain deficiencies, it is still desirable to further improve.
Summary of the invention
1. technical problems to be solved by the inivention
The present invention is based on biometric technology applications, provide one kind and establish identity template based on eyebrow and carry out body
Part knows method for distinguishing;Present invention is generally directed to the pure eyebrow images in front face, on the basis of extracting accurate eyebrow outline, with
The mode of feature vector constructs effective eyebrow knowledge representation model, is used for identification;How to obtain accurate eyebrow outline with
And on the basis of profile, effective feature set how about is designed to construct eyebrow knowledge representation model, is difficult point institute of the invention
?;The present invention is introduced pseudo- ball edge detection operator in the level set algorithm corrected based on biased field, and eyebrow is effectively partitioned into
Hair wheel is wide, and on the basis of gained eyebrow outline, calculates shape, the direction character of eyebrow, in conjunction with textural characteristics, jointly
Eyebrow knowledge representation model is constructed, it also can be beauty culture and case which, which can not only realize identification,
Eyebrow feature involved in investigative technique provides key technique reference.
2. technical solution
In order to achieve the above objectives, technical solution provided by the invention are as follows:
A kind of method for establishing identity template based on eyebrow of the invention, the steps include:
Step 1: Zheng Lian topography of the acquisition containing eyebrow;
Step 2: intercept pure brow region, generate the pure brow region image of rectangle and simultaneously pre-process, respectively obtain right eyebrow with
Left eyebrow image;
Step 3: obtaining eyebrow edge line using edge detection operator, eyebrow edge line inner area is filled using morphology
Domain obtains initial profile line of the rough closed curve as level set movements, realizes eyebrow coarse segmentation;
Step 4: being developed using partial differential equation guidance level set function, the accurate outer contour of eyebrow is obtained, realizes eyebrow
It is accurate to extract;
Step 5: calculating separately the shape feature and direction character of eyebrow on the basis of step 4;
Step 6: the shape feature that step 5 is obtained, direction character construct eyebrow in conjunction with eyebrow textural characteristics jointly
Knowledge representation model;
Step 7: repeat step 1 to step 63 times, i.e. the eyebrow knowledge representation model of 3 same persons of random acquisition,
The identity template for forming the people, is added to the eyebrow validation database of identification.
Further, position of the camera apart from face face 1m in step 1, face or so and downward shift angle
No more than 10 °, and the top and the bottom of camera rectangle view-finder, respectively close among forehead and in the middle part of the bridge of the nose, left-right parts are close
Face right boundary acquires the Zheng Lian topography containing eyebrow.
Further, Zheng Lian topography obtained by step 1 is divided into left eyebrow and right eyebrows area first by step 2
Domain arbitrarily chooses one group of point to left and right eyebrow image respectively and is linked to be the polygon containing brow region, which requires to maximize
Brow region is included, the minimum circumscribed rectangle comprising the polygon is then calculated, generates left and right pure eyebrow image;Institute
Having the left and right pure eyebrow image of acquisition need to be pre-processed, including gray processing and size normalization, respectively indicate normalizing with W and H
The width and height of eyebrow image after change, it is desirable that 32≤H≤128 (unit: pixel), and W is 2 to 5 times of H.
Further, step 3 utilizes the edge detection operator based on pseudo- ball to pure eyebrow image obtained by step 2 first
Contour detecting is carried out, the coarse contour line of detection is not closure, is filled up in contour line using the closed operation in morphology
Slight crack and small hole, then using area filling will fill inside eyebrow edge line region, obtain a region interior pixels
Value is 1, and the bianry image that outside is 0 is denoted as P, P is brought into as in undefined level set initialization function:
φ0(x, y)=2c0(0.5-P)
Wherein c0For non-negative constant value;The symbolic measurement that level set initialization function takes first closure curve to generate is first
Beginning condition.
Further, shape feature and direction character are expressed as follows respectively in step 5:
1) shape feature: the accurate eyebrow outline figure that step 4 is obtained, calculate separately left and right brow region perimeter,
5 area, width, height, eccentricity shape features, are denoted as L respectivelyi, Si, Wi, Hi, Ei, as i=l, indicate left eyebrow;I=
When r, right eyebrow is indicated;It is indicated with feature vector are as follows:
2) direction character: representing the bending direction and degree of eyebrow using the matched curve of eyebrow outline curve, calculates
Eyebrow direction character is portrayed at the inclination angle of 5 tangent lines in matched curve, and 5 inclinations angle are denoted as respectively: θi1、θi2、θi3、
θi4And θi5, as i=l, indicate left eyebrow;When i=r, indicates right eyebrow, is indicated with feature vector are as follows:
Further, the textural characteristics in step 6 are respectively angle second order away from Mi1, contrast ratio Mi2, correlation Mi3And entropy
Mi4, as i=l, indicate left eyebrow;When i=r, indicates right eyebrow, is indicated with feature vector are as follows:
Shape feature, direction character combination eyebrow textural characteristics, the eyebrow knowledge representation model constructed jointly are denoted as:
R=(R1,R2,R3)。
A kind of method that identification is carried out based on eyebrow of the invention, when identification, by tested people according to step
One obtains its eyebrow knowledge representation model to be identified to step 6, then by eyebrow knowledge representation model to be identified and database
In 3 eyebrows verifying knowledge representation model of same people carry out similarity detection respectively, and by the affiliated people of the identity of best match
It exports as a result.
Further, the knowledge representation model of eyebrow is indicated in a manner of eyebrow feature vector, and characteristic
There is three classes difference scale rank, therefore calculates separately out the opposite Euclidean distance D between shape featureshape, between direction character
Relative distance DdirectionThe distance between textural characteristics Dtexture, wherein DshapeIs defined as:
Wherein, p is the Shape feature set in the eyebrow knowledge representation model of people to be identified, and q is in eyebrow validation database
Shape feature set;DdirectionAnd DtextureIt is defined above.
Further, the distance between comprehensive three category features, portray similarity S between eyebrow imagesim, it is defined as follows:
Further, if eyebrow knowledge representation model to be identified and 3 eyebrows verifying of people same in database are known
Know at least two similarity S of expression modelsim75% or more, then it is assumed that identification success, by the identity of the best match
Affiliated people exports as a result.
3. beneficial effect
Using technical solution provided by the invention, compared with existing well-known technique, there is following remarkable result:
(1) a kind of method that identity template is established based on eyebrow of the invention, in view of the water corrected based on biased field
Flat set algorithm, which only considered, is added to inhibition image overall gray scale inhomogeneities in conventional flat set algorithm for biased field, but does not examine
The local edge information for considering target itself in image, often will appear segmentation effect difference and the number of iterations more than problem, selection
The coarse contour line of eyebrow is detected based on the edge detection operator of pseudo- ball, pseudo- ball edge detection operator is in the item for keeping flatness
Under part, compared with common edge detection operator, higher edge precision can be obtained, and can be while detecting edge
Weaken noise, realizes very fast and complete Ground Split eyebrow outer profile;
(2) a kind of method that identity template is established based on eyebrow of the invention, it is initial bent by original manual regulation
Line is changed to the profile point using the automatic detection image of edge detection operator, obtains the rough wheel that can substantially reflect eyebrow outline shape
Profile, and using this as the initial curve of level set, one kind is realized based on area-of-interest and sets initial profile, is finally utilized
Eyebrow segmentation is carried out based on the iterative process of the level set algorithm of biased field, segmentation effect substantially improves;
(3) a kind of method for carrying out identification based on eyebrow of the invention, on the basis of being based on eyebrow, then calculates eyebrow
The shape feature and geometrical characteristic of hair, effectively prevent the disturbing factors such as eyebrow gray scale, and calculation amount is smaller and easy calculating;By eyebrow
Feature and textural characteristics on hair geometric meaning consider description eyebrow collectively as a kind of knowledge representation of eyebrow more fully hereinafter
Hair picture material each characteristic parameter, be able to solve because it is single consider eyebrow feature in a certain respect caused by identity know
Not Shi Xiao the problem of;
(4) a kind of method that identification is carried out based on eyebrow of the invention, to living things feature recognitions such as face and irises
For, it can use scheme provided by the invention and carry out the extraction of face outer profile and iris profile, and calculate a series of
Characteristic value based on chamfered shape is to construct their knowledge representation model, therefore the present invention has preferable scalability.
Detailed description of the invention
Fig. 1 is the flow chart of the personal identification method based on eyebrow in the present invention;
(a)~(d) in Fig. 2 is original pure eyebrow image;(e)~(h) in Fig. 2 is rectangle initial profile, in Fig. 2
(i)~(l) is that (m)~(p) based on (e) in Fig. 2~(h) initial profile segmentation result figure, in Fig. 2 is base of the invention
Initial profile figure after edge detection operator, (q)~(t) in Fig. 2 are points based on (m) in Fig. 2~(p) initial profile
Cut result figure;
Fig. 3 is the schematic diagram that eyebrow direction character is sought in the present invention;
Fig. 4 is the work flow diagram for carrying out authentication in the present invention based on eyebrow.
Specific embodiment
To further appreciate that the contents of the present invention, the present invention is described in detail in conjunction with the accompanying drawings and embodiments.
Embodiment 1
A kind of method for establishing identity template based on eyebrow of the present embodiment, the hardware device being related to include that image is adopted
Collect equipment and computer, wherein image capture device is by image pick-up card CG300, CP240 Panasonic video camera and 75mm high-precision
Japanese import camera lens composition, computer are selected as the Dell type of Intel (R) Pentium (R) CPU, dominant frequency 2.13GHz, RAM 2GB
Number.
The implementing procedure of the present embodiment is referring to Fig. 1, and its step are as follows:
Step 1: Zheng Lian topography of the acquisition containing eyebrow, specifically:
Under general illumination condition, by image capture device, to N (≤500) individual, everyone acquires original graph of 3 width containing eyebrow
Picture generates (3*N) width original image containing eyebrow library, and to the good affiliated personal identification of eyebrow image tagged.In acquisition original graph
As during, it is desirable that camera lens are installed on the position of face face distance 1m, and face or so and downward shift angle do not surpass
10 ° are crossed, in order to simplify subsequent calculating, directly by the top and the bottom of camera rectangle view-finder respectively close to forehead centre and the bridge of the nose
Middle part, left-right parts acquire everyone approximate region containing eyebrow close to face right boundary.
Step 2: intercept pure brow region, generate the pure brow region image of rectangle and simultaneously pre-process, respectively obtain right eyebrow with
Left eyebrow image library;Specifically:
By every width original image containing eyebrow in image library obtained by step 1, by computer substantially from nose middle position
Left eyebrow and right eyebrows administrative division map are vertically cut out, computer is passed through successively to every obtained left and right eyebrow image respectively
It arbitrarily chooses one group of point and is linked to be the polygon (polygon maximizes be included brow region as far as possible) comprising brow region,
The minimum circumscribed rectangle comprising the polygon is calculated, the pure brow region of rectangle is intercepted, generates left and right pure eyebrow image library.Pre- place
Reason includes that gray processing and size normalize, the width and height of all eyebrow images after respectively indicating normalization with W and H, so that 32≤H
≤ 128 (units: pixel), and pretreating effect is best when W is 2 to 5 times of H, takes H=50, W=150 specific to the present embodiment;
Obtain the left eyebrow image library and corresponding right eyebrow image library of marked good personal identification.
Step 3: obtaining eyebrow edge line using edge detection operator, eyebrow edge line inner area is filled using morphology
Domain obtains initial profile line of the rough closed curve as level set movements;Specifically:
The present embodiment randomly chooses 4 pure eyebrow images of different people from pure right eyebrow image library, with (a) in Fig. 2
4 pure eyebrow images shown in~(d) are illustrated.
Biased field is added to conventional flat in view of only considered based on the level set algorithm that biased field is corrected by the present embodiment
Inhibit image overall gray scale inhomogeneities in set algorithm, but not in view of the local edge information of target in image itself, often
Will appear segmentation effect difference and the number of iterations more than problem, selection the coarse of eyebrow is detected based on the edge detection operator of pseudo- ball
Contour line, pseudo- ball edge detection operator under conditions of keeping flatness, compared with common edge detection operator (such as Roberts,
Laplace, Sobel and Canny) for, higher edge precision can be obtained, and can subtract while detecting edge
Small noise realizes very fast and complete Ground Split eyebrow outer profile.The detective operators based on pseudo- ball are utilized to pure eyebrow image first
Contour detecting is carried out, a rough pure eyebrow outline binary map is obtained, the coarse contour line of detection is not generally closure,
Slight crack and small hole in contour line are filled up using the closed operation in morphology, then using area is filled eyebrow edge line
Filling inside region, obtaining a region internal pixel values is 1, and the bianry image that outside is 0 is denoted as P, then by this binary map
P is brought into the initialization function of level set, this function is defined as:
φ0(x, y)=2c0(0.5-P)
Wherein c0For non-negative constant value, general value is 1.
The symbolic measurement that level set initialization function usually takes first closure curve to generate is primary condition, and symbol
Distance function meet partial differential equation beI.e. gradient-norm is 1.For above formula, P is a bianry image, when P is
When 1, i.e., point is inside enclosed region, φ0Take -1;When P is 0, i.e., point is outside enclosed region, φ01 is taken, so, φ0Take
Value is only 1 or -1, thereforeMeet the definition of symbolic measurement.
This closed curve is extracted using the contour function contour carried in Matlab, as the first of level set movements
Beginning contour line, formula are as follows:
[c, h]=contour (φ, [0,0], ' k')
Wherein, c is equivalent wire matrix, stores the coordinate data of contour;H is the handle of contour figure, and φ is described
Level set initialization function, i.e. the condition that is met of z-axis data, [0,0] indicates to draw the contour that a height is 0, i.e.,
Initial profile line, ' k' expression indicates with black lines, shown in (m)~(p) in effect such as Fig. 2.
Initial profile line in traditional level set algorithm is (e) in manual cutting one closed curve, such as Fig. 2
Shown in~(h), setting the rectangle frame among eyebrow image and be used as initial profile, (i)~(l) in Fig. 2 is segmentation result figure, by
Segmentation result can be seen that the method is more time-consuming and fails to make full use of the information of image itself, lead to final segmentation effect simultaneously
Poorly, the present embodiment is to overcome this defect, improves the initial curve segment of level set, initial bent by original manual regulation
Line is changed to the profile point using the automatic detection image of edge detection operator, obtains the rough wheel that can substantially reflect eyebrow outline shape
Profile, and using this as the initial curve of level set, one kind is realized based on area-of-interest and sets initial profile, is finally utilized
Eyebrow segmentation is carried out based on the iterative process of the level set algorithm of biased field.
Step 4: developing using partial differential equation guidance level set function, the accurate outer contour of eyebrow is obtained;Specifically:
It after step 3 obtains initial profile line, is split, guides using based on the modified Level Set Method of biased field
The partial differential equation of iterative segmentation process are as follows:
Wherein, the initial value of level set function φ is φ0,It is gradient operator, div () is divergence operator, and v is long
Energy term coefficient is spent, μ is punishment energy term coefficient, and δ () is Dirac function, dpIs defined as:
Wherein, p is potential function, the deviation being able to guide between level set function φ and symbolic measurement.
eiIt is defined as follows:
ei(x)=∫ K (y-x) | I (x)-b (y) ci|2dy;I=1,2
Wherein, K is non-negative window function, and also referred to as kernel function, b are biased field, ciFor gray scale constant value, I is pure eyebrow
Image, K are defined as follows:
Wherein, α is normaliztion constant, and σ is Gaussian kernel scale parameter.
The present embodiment is in eyebrow segmentation, and experiment parameter is in addition to the number of iterations is different, and miscellaneous stipulations are as follows: kernel function K
In σ=4.0, v=0.001 × 2552, μ=1.0, in time step t=0.1, Fig. 2 (i)~(l) and (q)~(t) from a left side to
Right iteration number is successively set as 5,10,10,5, shown in (q)~(t) in final segmentation effect such as Fig. 2, it can be seen that in phase
Under same the number of iterations, the present embodiment segmentation is more accurate, can faster develop onto the profile of eyebrow, comes to (d) image in Fig. 2
It says, it is little with original difference after improvement when intensity profile is more uniform.In order to objectively evaluate the superiority of the present embodiment, use
Jaccard similarity (JS) index judges segmentation precision, i.e.,
Wherein S1, S2The segmentation result for being accurate segmentation result (general by taking off by hand) respectively and needing to judge, this refers to
It marks higher, it is meant that the performance of algorithm is better, and the accuracy of contours segmentation is higher.By the present embodiment and traditional based on biased field
Level set algorithm (Li model) [Li C M, Huang R, Ding Z H, the et al.A Level Set Method for of correction
Image Segmentation in the Presence of Intensity Inhomogeneities with
Application to MRI [C] .IEEE Transactions on image processing, 2011,20 (7): 2007-
2016.] it is compared, as shown in table 1:
1 segmentation precision contrast table (%) of table
Table 1 is the segmentation precision contrast table that the 8 width eyebrow images extracted from right eyebrow image library at random are done, by table 1
It is found that being compared to traditional algorithm, the present embodiment segmentation is more effective.
For other eyebrow images in right eyebrow image library, the accurate outer contour segmentation of eyebrow similarly can be obtained
Figure.
Step 5: calculating separately the shape feature and direction character of eyebrow on the basis of step 4;Specifically:
1) shape feature: carrying out binary conversion treatment to the eyebrow segmentation figure that step 4 obtains, and pixel is 1 in contour area,
Region exterior pixel be 0, then to this bianry image carry out thresholding, delete bianry image in be possible to occur small area and
Isolated area, obtaining pixel in a brow region is 1, and the eyebrow bianry image that background pixel is 0 is denoted as BW, utilizes Matlab
Regionprops function in software carries out the calculating of shape feature value, regionprops function grammer are as follows:
STATS=regionprops (L, properties)
Wherein, L is the matrix with BW same size, the class label comprising each connected region in BW is marked, STATS
It is the structural array that a length is max (L (:)), the corresponding field of structural array defines under each region respective attributes
Measurement.
The present embodiment has acquired 5 characteristic attributes: the perimeter L of brow regioni, area Si, width Wi, height Hi, eccentricity
Ei(the elliptical eccentricity with brow region with identical standard second-order moment around mean), i.e. properties takes respectively '
Perimeter', ' Area', ' BoundingBox' and ' Eccentricity', what wherein BoundingBox was asked is brow region
Minimum circumscribed rectangle, the width of upper right corner origin coordinates and matrix comprising boundary rectangle and the width of height and eyebrow
And height value.As i=l, left eyebrow is indicated;When i=r, indicates right eyebrow, is indicated with feature vector are as follows:
2) direction character: the present embodiment represents the bending direction and journey of eyebrow using the matched curve of eyebrow outline curve
Degree, the inclination angle (value range be [0,180) of 5 tangent lines on digital simulation curve) portray eyebrow direction character, 5 points
Inclination angle is denoted as respectively: θi1、θi2、θi3、θi4And θi5(as i=l, indicate left eyebrow;When i=r, right eyebrow is indicated), eyebrow
Direction character is indicated with feature vector are as follows:
It is worth noting that direction character directly reflects the bending direction and degree of eyebrow, direction character and eyebrow wheel
Profile is related, and about eyebrow segmentation figure, eyebrow outline line is a closed curve, but due to being influenced by various noises,
It inevitably will appear not smooth enough the phenomenon of contour line, therefore eyebrow outline line sought using least-square fitting approach
The curved course of eyebrow is portrayed in matched curve, first coordinate at calculating contour line and the matched curve left side and the right endpoint, note
Two endpoints are respectively h1And h5, then calculate h1And h5Midpoint coordinates, midpoint is denoted as h3, similarly, then calculate h1And h3Midpoint h2
Coordinate, h3And h5Midpoint h4Coordinate, after the coordinate for acquiring five points, calculate separately and matched curve cutting in these points
Line finally acquires the inclination angle (value range be [0,180) of tangent line), with (d) in Fig. 2 for original image, seek direction character
As shown in figure 3, being denoted as θ respectivelyr1、θr2、θr3、θr4And θr5。
Step 6: by shape feature, direction character, in conjunction with eyebrow textural characteristics, a kind of common knowledge for constructing eyebrow
Expression model;Specifically:
After the shape feature and direction character that acquire eyebrow by step 5, consider that not only outline shape has spy to eyebrow image
Anisotropic and stability, textural characteristics also have the characteristics that this.Gray level co-occurrence matrixes (GLCM) are the texture for gray level image
A kind of method that information is analyzed, embodies the texture information of gray level image from direction.It is relatively accurate in view of having obtained
Eyebrow outline will (back outside the profile of eyebrow segmentation figure in order to avoid requirement of the background gray scale unevenness to textural characteristics outside profile
Scape part) gray value is uniformly set as 0, and texture feature extraction is carried out on this image, realize step are as follows: calculate pure eyebrow image
Co-occurrence matrix (direction takes 0 °, 45 °, 90 °, 135 °) on 4 directions → 4 co-occurrence matrixs normalization → to being total to after normalization
Raw matrix calculates angle second order away from Mi1, contrast ratio Mi2, correlation Mi3With entropy Mi4[four parametric textures are specifically defined referring to document " base
In the texture feature extraction of gray level co-occurrence matrixes " (" computer system application " the 19th phase in 2010)], as i=l, indicate left
Eyebrow;When i=r, indicate that right eyebrow, texture feature vector may be expressed as:
Therefore, after the geometrical characteristic and the textural characteristics that merge eyebrow, eyebrow image knowledge expression model is with feature vector
Form indicates are as follows:
R=(R1,R2,R3)
That is:
For the ease of subsequent calculating, eyebrow knowledge representation model R is switched into one-dimensional row vector by bivector, is denoted as:
R=(Rl,Rr)
Wherein, RlRepresent the first row (left eyebrow feature set) in R, RrRepresent the second row (right eyebrow feature set) in R.
The present embodiment is on the basis of being based on eyebrow, then calculates the shape feature and geometrical characteristic of eyebrow, effectively prevents
The disturbing factors such as eyebrow gray scale, calculation amount is smaller and easy calculating;By on eyebrow geometric meaning feature and textural characteristics it is common
As a kind of knowledge representation of eyebrow, each characteristic parameter of description eyebrow picture material is considered more fully hereinafter, can be solved
Certainly because it is single consider the problems of eyebrow feature in a certain respect caused by identification fail.
Step 7: repeat step 1 to step 63 times, i.e. the eyebrow knowledge representation model of 3 same persons of random acquisition,
The knowledge representation model for calculating every width eyebrow image in the left eyebrow of (3*N) width and right eyebrow image library constructs (3*N) row
The eyebrow characteristic sequence database of 28 column (N number of people altogether, everyone acquires three times, each 14 characteristic values of left and right eyebrow) size, shape
At the identity template of the people, it is added to the eyebrow validation database of identification.
Embodiment 2
In conjunction with Fig. 4, a kind of method carrying out identification based on eyebrow of the present embodiment is different substantially with embodiment 1
Place is that tested people is obtained its eyebrow knowledge representation model to be identified according to step 1 to step 6 by the present embodiment,
Then 3 eyebrows of same people in eyebrow knowledge representation model to be identified and the built database of embodiment 1 are verified into knowledge representation
Model carries out similarity detection respectively, the feature of three kinds of different scales as involved in knowledge representation model, so remembering shape respectively
Opposite Euclidean distance between shape feature is Dshape, the relative distance between direction character be DdirectionBetween textural characteristics
Distance be Dtexture, the distance between three category features are finally integrated, similarity S is calculatedsim, define SsimAre as follows:
Wherein DshapeIs defined as:
P is the Shape feature set in the eyebrow knowledge representation model of people to be identified, and q is the shape in eyebrow validation database
Feature set;DdirectionAnd DtextureIt is defined above.
If at least 2 similarity Ssim75% or more, then it is assumed that identification success, and by the body of best match
People belonging to part exports as optimum.
For the living things feature recognitions such as face and iris, it can use scheme provided in this embodiment and carry out outside face
The extraction of profile and iris profile, and a series of characteristic values based on chamfered shape are calculated to construct their knowledge representation
Model, therefore the present embodiment has preferable replicability.
Schematically the present invention and embodiments thereof are described above, description is not limiting, institute in attached drawing
What is shown is also one of embodiments of the present invention, and actual structure is not limited to this.So if the common skill of this field
Art personnel are enlightened by it, without departing from the spirit of the invention, are not inventively designed and the technical solution
Similar frame mode and embodiment, are within the scope of protection of the invention.
Claims (10)
1. a kind of method for establishing identity template based on eyebrow, the steps include:
Step 1: Zheng Lian topography of the acquisition containing eyebrow;
Step 2: intercepting pure brow region, generating the pure brow region image of rectangle and pre-processing, right eyebrow and left eyebrow are respectively obtained
Hair image;
Step 3: obtaining eyebrow edge line using edge detection operator, eyebrow edge line interior zone is filled using morphology, is obtained
Initial profile line of the closed curve as level set movements roughly is obtained, realizes eyebrow coarse segmentation;
Step 4: developing using partial differential equation guidance level set function, the accurate outer contour of eyebrow is obtained, realizes that eyebrow is accurate
It extracts;
Step 5: calculating separately the shape feature and direction character of eyebrow on the basis of step 4;
Step 6: the shape feature that step 5 is obtained, direction character, common to construct knowing for eyebrow in conjunction with eyebrow textural characteristics
Know expression model;
Step 7: repeat step 1 to step 63 times, i.e. the eyebrow knowledge representation model of 3 same persons of random acquisition is formed
The identity template of the people is added to the eyebrow validation database of identification.
2. a kind of method for establishing identity template based on eyebrow according to claim 1, it is characterised in that: step 1
Position of the middle camera apart from face face 1m, face or so and downward shift angle are no more than 10 °, and camera rectangle is found a view
Respectively close among forehead and in the middle part of the bridge of the nose, left-right parts are acquired close to face right boundary containing eyebrow for the top and the bottom of frame
Zheng Lian topography.
3. a kind of method for establishing identity template based on eyebrow according to claim 1 or 2, it is characterised in that: step
Zheng Lian topography obtained by step 1 is divided into left eyebrow and right eyebrows region first by rapid two, respectively to left and right eyebrow image
It arbitrarily chooses one group of point and is linked to be the polygon containing brow region, which requires to maximize and be included brow region, so
The minimum circumscribed rectangle comprising the polygon is calculated afterwards, generates left and right pure eyebrow image;The left and right pure eyebrow figure of all acquisitions
Picture need to be pre-processed, including gray processing and size normalization, the width and height of eyebrow image after respectively indicating normalization with W and H,
It is required that 32≤H≤128 and W are 2 to 5 times of H.
4. a kind of method for establishing identity template based on eyebrow according to claim 3, it is characterised in that: step 3
Contour detecting, the rough wheel of detection are carried out using the edge detection operator based on pseudo- ball to pure eyebrow image obtained by step 2 first
Profile is not closure, and the slight crack and small hole in contour line are filled up using the closed operation in morphology, then uses area
Domain filling will fill inside eyebrow edge line region, and obtaining a region internal pixel values is 1, the bianry image that outside is 0, note
For P, P is brought into as in undefined level set initialization function:
φ0(x, y)=2c0(0.5-P)
Wherein c0For non-negative constant value;The symbolic measurement that level set initialization function takes first closure curve to generate is initial strip
Part.
5. a kind of method for establishing identity template based on eyebrow according to claim 4, it is characterised in that: step 5
Middle shape feature and direction character are expressed as follows respectively:
1) shape feature: the accurate eyebrow outline figure that step 4 is obtained, calculate separately the perimeter of left and right brow region, area,
5 width, height, eccentricity shape features, are denoted as L respectivelyi, Si, Wi, Hi, Ei, as i=l, indicate left eyebrow;When i=r,
Indicate right eyebrow;It is indicated with feature vector are as follows:
2) bending direction and degree of eyebrow, digital simulation direction character: are represented using the matched curve of eyebrow outline curve
Eyebrow direction character is portrayed at the inclination angle of 5 tangent lines on curve, and 5 inclinations angle are denoted as respectively: θi1、θi2、θi3、θi4With
θi5, as i=l, indicate left eyebrow;When i=r, indicates right eyebrow, is indicated with feature vector are as follows:
6. a kind of method for establishing identity template based on eyebrow according to claim 5, it is characterised in that: step 6
In textural characteristics be respectively angle second order away from Mi1, contrast ratio Mi2, correlation Mi3With entropy Mi4, as i=l, indicate left eyebrow;I=
When r, indicates right eyebrow, is indicated with feature vector are as follows:
Shape feature, direction character combination eyebrow textural characteristics, the eyebrow knowledge representation model constructed jointly are denoted as:
R=(R1,R2,R3)。
7. the side that a kind of method for establishing identity template based on eyebrow according to claim 6 carries out identification
Method, it is characterised in that: when identification, tested people is obtained into its eyebrow knowledge table to be identified according to step 1 to step 6
Up to model, 3 eyebrows of same people in eyebrow knowledge representation model to be identified and database are then verified into knowledge representation model
Similarity detection is carried out respectively, and the affiliated people of the identity of best match is exported as a result.
8. the side that a kind of method for establishing identity template based on eyebrow according to claim 7 carries out identification
Method, it is characterised in that: the knowledge representation model of eyebrow is to be indicated in a manner of eyebrow feature vector, and characteristic has three classes
Different scale ranks, therefore calculate separately out the opposite Euclidean distance D between shape featureshape, it is opposite between direction character
Distance DdirectionThe distance between textural characteristics Dtexture, wherein DshapeIs defined as:
Wherein, p is the Shape feature set in the eyebrow knowledge representation model of people to be identified, and q is the shape in eyebrow validation database
Shape feature set;DdirectionAnd DtextureIt is defined above.
9. the side that a kind of method for establishing identity template based on eyebrow according to claim 8 carries out identification
Method, it is characterised in that: the distance between comprehensive three category features portray similarity S between eyebrow imagesim, it is defined as follows:
10. the side that a kind of method for establishing identity template based on eyebrow according to claim 9 carries out identification
Method, it is characterised in that: if 3 eyebrows of same people verify knowledge representation in eyebrow knowledge representation model to be identified and database
At least two similarity S of modelsim75% or more, then it is assumed that identification success, by the affiliated people of the identity of the best match
It exports as a result.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610536639.0A CN106203329B (en) | 2016-07-08 | 2016-07-08 | A method of identity template is established based on eyebrow and carries out identification |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610536639.0A CN106203329B (en) | 2016-07-08 | 2016-07-08 | A method of identity template is established based on eyebrow and carries out identification |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106203329A CN106203329A (en) | 2016-12-07 |
CN106203329B true CN106203329B (en) | 2019-02-22 |
Family
ID=57473834
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610536639.0A Active CN106203329B (en) | 2016-07-08 | 2016-07-08 | A method of identity template is established based on eyebrow and carries out identification |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106203329B (en) |
Families Citing this family (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107015745B (en) * | 2017-05-19 | 2020-03-24 | 广东小天才科技有限公司 | Screen operation method and device, terminal equipment and computer readable storage medium |
CN107464253B (en) * | 2017-07-10 | 2020-11-20 | 北京小米移动软件有限公司 | Eyebrow positioning method and device |
CN108256440A (en) * | 2017-12-27 | 2018-07-06 | 长沙学院 | A kind of eyebrow image segmentation method and system |
CN108388842B (en) * | 2018-01-31 | 2019-07-23 | Oppo广东移动通信有限公司 | Intelligent prompt method and Related product |
CN109271922B (en) * | 2018-09-13 | 2022-01-04 | 深圳市梦网视讯有限公司 | Nasal part positioning method and system based on contrast |
CN110533001B (en) * | 2019-09-06 | 2022-02-08 | 厦门久凌创新科技有限公司 | Big data face recognition method based on face recognition |
CN113689325A (en) * | 2021-07-12 | 2021-11-23 | 深圳数联天下智能科技有限公司 | Method for digitizing beautiful eyebrows, electronic device and storage medium |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1645406A (en) * | 2005-02-24 | 2005-07-27 | 北京工业大学 | Identity discriminating method based on eyebrow identification |
CN1801180A (en) * | 2005-02-24 | 2006-07-12 | 北京工业大学 | Identity recognition method based on eyebrow recognition |
CN101246543A (en) * | 2008-03-18 | 2008-08-20 | 苏州纳米技术与纳米仿生研究所 | Examiner identity appraising system based on bionic and biological characteristic recognition |
CN102982320A (en) * | 2012-12-05 | 2013-03-20 | 山东神思电子技术股份有限公司 | Method for extracting eyebrow outline |
CN104951743A (en) * | 2015-03-04 | 2015-09-30 | 苏州大学 | Active-shape-model-algorithm-based method for analyzing face expression |
-
2016
- 2016-07-08 CN CN201610536639.0A patent/CN106203329B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1645406A (en) * | 2005-02-24 | 2005-07-27 | 北京工业大学 | Identity discriminating method based on eyebrow identification |
CN1801180A (en) * | 2005-02-24 | 2006-07-12 | 北京工业大学 | Identity recognition method based on eyebrow recognition |
CN101246543A (en) * | 2008-03-18 | 2008-08-20 | 苏州纳米技术与纳米仿生研究所 | Examiner identity appraising system based on bionic and biological characteristic recognition |
CN102982320A (en) * | 2012-12-05 | 2013-03-20 | 山东神思电子技术股份有限公司 | Method for extracting eyebrow outline |
CN104951743A (en) * | 2015-03-04 | 2015-09-30 | 苏州大学 | Active-shape-model-algorithm-based method for analyzing face expression |
Non-Patent Citations (1)
Title |
---|
Eyebrow Recognition Based on Sparsity Preserving Projections;Xinwu Yang 等;《IEEE Conference Anthology》;20131231;第1-4页 |
Also Published As
Publication number | Publication date |
---|---|
CN106203329A (en) | 2016-12-07 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106203329B (en) | A method of identity template is established based on eyebrow and carries out identification | |
US7873189B2 (en) | Face recognition by dividing an image and evaluating a similarity vector with a support vector machine | |
WO2016150240A1 (en) | Identity authentication method and apparatus | |
Dibeklioglu et al. | 3D facial landmarking under expression, pose, and occlusion variations | |
Ibrahim et al. | Leaf recognition using texture features for herbal plant identification | |
CN109376717A (en) | Personal identification method, device, electronic equipment and the storage medium of face comparison | |
CN109086809A (en) | Based on the improved early stage Alzheimer disease classification method supervised and be locally linear embedding into | |
CN106980825A (en) | A kind of face posture sorting technique based on normalization pixel difference feature | |
CN109145716B (en) | Boarding gate verifying bench based on face recognition | |
CN111274883A (en) | Synthetic sketch face recognition method based on multi-scale HOG (histogram of oriented gradient) features and deep features | |
Ebrahim et al. | Features selection for offline handwritten signature verification: state of the art | |
Althabhawee et al. | Fingerprint recognition based on collected images using deep learning technology | |
Alsawwaf et al. | In your face: person identification through ratios and distances between facial features | |
Liao et al. | Nonrigid brain MR image registration using uniform spherical region descriptor | |
CN108010015A (en) | One kind refers to vein video quality evaluation method and its system | |
Albadarneh et al. | Iris recognition system for secure authentication based on texture and shape features | |
CN106778491A (en) | The acquisition methods and equipment of face 3D characteristic informations | |
CN105844605A (en) | Face image synthesis method based on adaptive expression | |
CN116612538A (en) | Online confirmation method of electronic contract content | |
Ahdid et al. | Three dimensional face surfaces analysis using geodesic distance | |
Sun et al. | Virtual reality of recognition technologies of the improved contour coding image based on level set and neural network models | |
Malik et al. | An efficient retinal vessels biometric recognition system by using multi-scale local binary pattern descriptor | |
Almutiry | Efficient iris segmentation algorithm using deep learning techniques | |
Tian et al. | The research on rejoining of the oracle bone rubbings based on curve matching | |
Alasadi et al. | Dorsal hand-vein images recognition system based on grey level co-occurrence matrix and Tamura features |
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 |