CN101923645B - Iris splitting method suitable for low-quality iris image in complex application context - Google Patents
Iris splitting method suitable for low-quality iris image in complex application context Download PDFInfo
- Publication number
- CN101923645B CN101923645B CN2009100722308A CN200910072230A CN101923645B CN 101923645 B CN101923645 B CN 101923645B CN 2009100722308 A CN2009100722308 A CN 2009100722308A CN 200910072230 A CN200910072230 A CN 200910072230A CN 101923645 B CN101923645 B CN 101923645B
- Authority
- CN
- China
- Prior art keywords
- iris
- image
- outline
- eyes
- edge
- 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.)
- Expired - Fee Related
Links
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses an iris splitting method suitable for low-quality iris images. The prior art can not carry out robust splitting to the low-quality iris image with mass interference and noise. The invention uses a human eye detection algorithm to preliminarily determine a subimage of the human eye and is applied to image splitting based on interdependent histogram and cluster and ellipse Hough conversion to accurately determine the subimage of the human eye; improved Hough conversion is adopted to position the outer contour of the iris and output the integrodifferential operator value CID of a detection result; if the outer contour of the iris is not accurately positioned, parabola approximation is carried out on the image to judge whether the image is an eye-closed image; for the non eye-closed image, complexion information is utilized to re-determine the outer contour of the iris; the palpebra superior is detected by one-dimensional signal detection and parabola integrodifferential operator; the palpebra inferior is detected by one-dimensional signal detection and an RANSAC algorithm; the histogram in the iris is calculated; and a threshold value is found to remove a highly bright spot. The invention is used for iris splitting of low-quality iris image in a complex application context.
Description
Technical field
The technical field that the present invention relates to comprises Flame Image Process, pattern-recognition and machine learning, specifically, has proposed a kind of iris splitting method to inferior quality iris image in the complex application context.
Background technology
Biometrics identification technology based on iris has the accuracy of identification height, can't forge and advantage such as non-infringement property; Gate control system at dwelling house and intelligent building; Customs's entry and exit and airport are with a wide range of applications in many fields such as identity validation of finance, security, insurance, social agency.Yet present iris authentication system requires highly cooperation of user, can only closely gather and require the user not move, so the application of iris authentication system received very big restriction, can't be applied in the application scenarios of various complicacies.
Modern iris authentication system is in order to adapt to the application scenarios of various complicacies, and the reply user has as far as possible little constraint, makes the existence of the imperceptible system of user, and iris recognition becomes inevitable development trend in therefore remote, the motion.Yet the iris image of in these complicated application scenarioss, gathering comprises various noises or interference, like head deflection, and eyes semi-closure or full cut-off, hair, eyelashes and eyelid block, and defocus and motion blur optical glasses, contact lenses interference etc.Therefore iris is partitioned into challenging problem in the complex scene, and the accuracy that iris is cut apart has directly determined the quality of the performance of an iris authentication system.
At present classical iris splitting method is based on the algorithm of integro differential operator and confirms the iris inside and outside contour.Horse strives the iris locating method of having invented based on mathematical morphology and probability statistics, and this method is at first utilized the round heart in the projection coarse positioning on the binaryzation iris image, according to the round heart in confirming, and radius of circle in search minimum mean sample is confirmed.Behind the circle, obtain the edge of cylindrical with binaryzation and Mathematical Morphology Method in confirming.Yuan Weiqi has invented a kind of human eye iris identification method, and this method is at first estimated pupil center, is set out by pupil center and detects four iris outer boundary points, confirms the iris center of circle at last.The iris splitting method of more than mentioning is very ripe, yet when iris capturing is a non-cooperation formula when gathering, the iris image of collection has a large amount of noises or disturbs, and these iris splitting methods just can not reach good effect.Therefore, we have invented a kind of iris splitting method that adapts to inferior quality iris image in the complex application context.
Summary of the invention
The purpose of this invention is to provide a kind of iris splitting method to inferior quality iris image in the complex application context; In the application scenarios of complicacy; The user is in the middle of iris capturing device distance far and is generally being moved; At this moment the iris image of gathering is because the less cooperation of user comprises much noise and interference, and current iris splitting method can't reach good segmentation effect, has a strong impact on the accuracy of iris authentication system.The present invention has solved this problem effectively.
The objective of the invention is to realize like this:
A kind of iris splitting method that adapts to inferior quality iris image in the complex application context; Use is tentatively confirmed the human eye subimage based on the human eye detection device of AdaBoost algorithm; Further use then based on the image segmentation and the oval Hough conversion of symbiosis histogram and K-Means cluster and confirm the human eye subimage more accurately; Adopt improved Hough conversion location iris outline and export the integro differential operator value CID of testing result; If the not accurate enough of iris outline location judges whether it is the image of closing one's eyes through iris image being carried out binaryzation, rim detection and fitting of parabola; Again confirm the iris outline for the non-imagery exploitation colour of skin information of closing one's eyes; Adopt the para-curve integro differential operator of one-dimensional signal detection algorithm and belt restraining to detect the upper eyelid at last, adopt one-dimensional signal detection algorithm and RANSAC algorithm to detect palpebra inferior, seek threshold value through the histogram that calculates iris inside and remove high bright spot.
The iris splitting method of inferior quality iris image in the above-mentioned adaptation complex application context, described based on the tentatively definite human eye subimage of human eye detection algorithm, be based on AdaBoost algorithm study human eye detection device, in iris image, eye areas is extracted.
The iris splitting method of inferior quality iris image in the above-mentioned adaptation complex application context; Described image segmentation and oval Hough conversion based on symbiosis histogram feature and K-Means cluster confirms that accurately the human eye subimage is to adopt symbiosis histogram between the self-adaptation dividing regions as proper vector; Based on the K-Means clustering algorithm subimage that step 1 obtains is cut apart; Generate edge image with the Canny edge detection method in the image after cutting apart, thereby use Hough conversion match eye contour to produce the elliptic region that comprises eyes more accurately then based on ellipse.
The iris splitting method of inferior quality iris image in the above-mentioned adaptation complex application context; Described improved Hough conversion based on circle confirms that the iris outline is meant in the elliptic region that obtains with Canny edge detection algorithm generation edge image; Use then based on the Hough conversion of circle and confirm the iris outline; When using Hough conversion statistics marginal point, need the gradient direction of CONSIDERING EDGE point simultaneously, to improve the accuracy of confirming the iris outline; Because candidate's circle of pixels statistics most number may not be the outline of real corresponding iris in the Hough conversion; Therefore in the Top 10 candidates circle that the Hough conversion obtains; Candidate's circle that the integro differential operator value is maximum is as best iris outline, and output integro differential operator value CID.
The iris splitting method of inferior quality iris image in the above-mentioned adaptation complex application context; The judgement of the described image of closing one's eyes is to use the integro differential operator value CID of outline to judge that whether this image is the candidate image of eyeball image of closing one's eyes; If CID less than certain threshold value, then is candidate image.
The iris splitting method of inferior quality iris image in the above-mentioned adaptation complex application context if think that the iris outline that obtains through claim 3 and 4 is enough accurate, need not to detect again; For candidate image, need further carry out rim detection, image binaryzation and fitting of parabola, the para-curve through match determines whether finally to be the image of closing one's eyes that if be judged as the image of closing one's eyes, then the iris cutting procedure finishes.
The iris splitting method of inferior quality iris image in the above-mentioned adaptation complex application context; Describedly confirm again that based on the colour of skin information iris outline is meant: when CID less than certain threshold value non-the closing one's eyes during image of image this moment, need utilize colour of skin information to confirm the iris outline once more; Detect skin pixel and non-skin pixel through skin color detector; Obtain binary image; The edge is detected in the bianry image back of expanding; Go out to comprise the elliptic region of eyes with the Hough change detection, the iris outline is confirmed in the improved Hough conversion of describing with claim 4 afterwards based on circle again.
The iris splitting method of inferior quality iris image in the above-mentioned adaptation complex application context; Described integro differential operator based on circle confirms that the pupil profile is meant: confirm the span in the pupil center of circle according to the center of circle of iris outline, the ratio of utilizing the interior profile of human eye iris outline and iris retrains the radius of pupil.Confirm the pupil profile in iris inside based on the integro differential operator of circle.
The iris splitting method of inferior quality iris image in the above-mentioned adaptation complex application context; The method that described detection iris upper eyelid is to use one-dimensional signal to detect is confirmed the marginal point in iris upper eyelid; Confirm the iris upper eyelid with the para-curve integro differential operator; In search procedure, remove obvious irrational candidate's para-curve as constraint with detected eyelid point; Described detection iris palpebra inferior is to use the one-dimensional signal detection method to confirm the part edge point of iris palpebra inferior equally, uses the para-curve of RANSAC algorithm match iris palpebra inferior.
The iris splitting method of inferior quality iris image in the above-mentioned adaptation complex application context; The inner high bright spot of described removal iris is meant: calculate the inner histogram of iris and this histogram is carried out Gauss level and smooth; The inner high bright spot of the corresponding iris in the peak on histogram right side, the paddy of getting this left side, peak is as the threshold value of removing high bright spot.
Advantage of the present invention:
1. progressively iris is cut apart through progressive method, and the accuracy of feedback mechanism to judge that iris is cut apart is provided;
2. merged multiple image processing techniques and utilized multiple image information, can be in the image that has much noise and interference robust, cut apart iris exactly.
3. should invention in the complex scene with the less mated condition of user under iris authentication system significant, make the iris recognition in remote, the motion become possibility, can expand the range of application of iris recognition greatly.
4. the present invention at first uses the human eye detection device that obtains based on machine learning method to detect eyes, obtains initial human eye subimage.Then based on gray scale symbiosis histogram or colour of skin information to image cut apart, rim detection and ellipse fitting, thereby obtain human eye subimage more accurately.The method of this progressively refinement has been removed non-iris region in the image, and follow-up operation is limited in the human eye subimage, so not only make associative operation more accurately and also efficient higher.
5. the present invention uses improved Hough conversion to confirm the iris outline, adopted a kind of feedback mechanism simultaneously so that the iris excircle configuration accurately and reliably.
6. the present invention proposes the para-curve integro differential operator of belt restraining condition and confirms the iris upper eyelid based on the method for one-dimensional signal.For the accuracy rate that guarantees that iris is cut apart, the present invention's methods such as eyeball of also closing one's eyes through judgement, and use the inner histogram of iris to remove the inner high bright spot of iris at last.Guaranteed the accuracy rate that iris is cut apart.
Description of drawings:
The iris image synoptic diagram of Fig. 1 in the scene of complicacy, collecting;
Fig. 2 is an iris segmenting system synoptic diagram;
Fig. 3 is histogrammic adaptive quantizing interval division synoptic diagram;
Fig. 4 is for to confirm iris outline synoptic diagram based on the histogrammic K-Means clustering algorithm of symbiosis, (a) the rough human eye subimage (c) that obtains of pending iris image (b) human eye detection based on image segmentation (d) image segmentation of symbiosis histogram feature and K-Means cluster after the outline map (e) of human eye subimage confirm that based on oval Hough conversion the improved Hough conversion based on circle of edge image (g) of human eye subimage (f) elliptic region more accurately confirms that iris outline (h) confirms the pupil profile based on the integro differential operator of circle;
Fig. 5 is for judging the schematic flow sheet of the image of closing one's eyes, and (a) candidate's close one's eyes histogram and threshold value of image of image (b) candidate of closing one's eyes confirms that (c) candidate image binaryzation (d) of closing one's eyes carries out expansion process (e) upper part that back edge of image figure (f) uses edge linking algorithm to find edge the longest in the edge image (g) to get this edge horizontal parabola match of going forward side by side of expanding to binary image;
Fig. 6 confirms iris outline synoptic diagram for using colour of skin information, and the expand edge image (f) of back subimage of the image (e) after (a) the human eye subimage (d) cut apart of initial human eye subimage (c) colour of skin information that obtains of pending iris image (b) human eye detection is cut apart the back and carried out image expansion confirms that based on oval Hough conversion the improved Hough conversion based on circle of edge image (h) of human eye subimage (g) elliptic region more accurately confirms that iris outline (i) confirms the pupil profile based on the integral differential algorithm of circle;
Fig. 7 is for detecting the upper eyelid synoptic diagram, (a) provided the eyelid para-curve of marginal point (f) match that gradient (e) one-dimensional signal of one-dimensional signal (c) Gauss that the hunting zone (b) of one-dimensional signal is used to cut apart one-dimensional signal (d) one-dimensional signal after level and smooth detects;
Fig. 8 is for detecting the iris split image synoptic diagram of going up palpebra inferior;
Fig. 9 is for removing the inner high bright spot synoptic diagram of iris, and the iris image (b) of (a) having confirmed iris inside and outside contour and last palpebra inferior is cut apart the bianry image that the back inner histogram of iris (c) is removed iris after the high bright spot.
Embodiment:
Embodiment 1:
The present invention is a kind of iris splitting method that adapts to inferior quality iris image in the complex application context; Use is tentatively confirmed the human eye subimage based on the human eye detection device of AdaBoost algorithm; Further use then based on the image segmentation and the oval Hough conversion of symbiosis histogram and K-Means cluster and confirm the human eye subimage more accurately; Adopt improved Hough conversion location iris outline and export the integro differential operator value CID of testing result; If the not accurate enough of iris outline location judges whether it is the image of closing one's eyes through iris image being carried out binaryzation, rim detection and fitting of parabola; Again confirm the iris outline for the non-imagery exploitation colour of skin information of closing one's eyes; Adopt the para-curve integro differential operator of one-dimensional signal detection algorithm and belt restraining to detect the upper eyelid at last, adopt one-dimensional signal detection algorithm and RANSAC algorithm to detect palpebra inferior, seek threshold value through the histogram that calculates iris inside and remove high bright spot.
In conjunction with accompanying drawing, Fig. 2 has provided the process flow diagram of iris splitting method, and the practical implementation step is following:
Step 2, accurately confirm the human eye subimage based on the image segmentation of symbiosis histogram feature and K-Means cluster and oval Hough conversion;
The iris outline is confirmed in step 3, improved Hough conversion based on circle;
The judgement of step 4, the image of closing one's eyes;
Step 5, confirm the iris outline again based on colour of skin information;
Step 6, based on the circle integro differential operator confirm the pupil profile;
Step 7, detection upper eyelid;
Step 8, detection palpebra inferior;
Step 9, the inner high bright spot of removal iris.
Wherein the practical implementation step of step 1 is:
Train 16 layers left eye and right eye detecting device respectively based on Adaboost algorithm and rectangular characteristic.Obtain a large amount of eye images and non-eye image as training set through in the laboratory, gathering and collect two kinds of methods, obtain positive routine sample and counter-example sample with the method for artificial mark from the internet.In learning process, require this system to have near zero false drop rate.With this human eye detection device detect comprise eyes a pocket as subimage.
Wherein the practical implementation step of step 2 is:
1) gray level self-adaptation subdivision
In the subimage that step 1 obtains, calculate comprise 64 between homogeneity range grey level histogram and carry out smoothly with Gaussian function; Each peak in the histogram and the paddy of these both sides, peak continue subdivision as an end points with this sub-range with this flex point if this sub-interval contains flex point as the histogrammic sub-interval of symbiosis.Fig. 3 representes the example of a gray level self-adaptation subdivision, in this example, has finally confirmed 5 intervals (noticing that о representes flex point here).
2) calculate gray scale symbiosis histogram as proper vector
On gray level self-adaptation subdivision basis, adopt normalized gray scale symbiosis histogram as proper vector.For a certain pixel z
*=(x
*, y
*), I (z
*) represent the gray-scale value of this point.According to the 1st) go on foot gray-scale value I is quantized in m the interval.Make S
*Expression is with z
*Be the center of circle, with d
*Be the set of long upright square area interior pixel point, promptly
‖ ‖ wherein
∞Represent infinite norm, then gray scale symbiosis histogram is calculated as follows:
U wherein, v=1 ..., m, S
z=z ' | ‖ z '-z ‖
∞=d), N
dBe normaliztion constant, δ () expression Kronecker function.Gray scale symbiosis histogram P
d(u v) is the associating grey level distribution with spatial coherence, its expression satisfy pixel z that gray level is u and with its distance be the probability of v for the gray level of the pixel z ' of d.
3) based on the image segmentation and the rim detection of K-Means cluster
Four significantly zones are arranged in the human eye subimage: iris, sclera, skin, pupil and eyelash, therefore type of getting number is 4 in the K-Means algorithm.In the K-Means algorithm, calculate the corresponding symbiosis histogram of each pixel as proper vector, use the Bhatacharyya coefficient as distance measure.Further carry out rim detection in the image after cutting apart.Result after Fig. 4 (c), (d) represent respectively the human eye subimage cut apart and its edge image.
4) accurately confirm the human eye subimage based on the Hough conversion of ellipse
Because the shape of human eye can be used ellipse match well, so adopt the elliptic region of confirming to comprise human eye based on the Hough conversion of ellipse.The oval center of circle of order is (x
c, y
c), major axis is a, and minor axis is b, and a bit (x y) satisfies equation on the ellipse
Each four-tuple (x of scanning in parameter space
c, y
c, a, b), for each four-tuple, statistics satisfies the number of all marginal points of the corresponding ellipse of this four-tuple.The ellipse that the maximum four-tuple of the marginal point number of being added up is corresponding is as testing result, and Fig. 4 (e) has provided the elliptic region that comprises eyes.
Wherein the practical implementation step of step 3 is:
1) the Hough conversion of consideration gradient direction
In the elliptic region that step 2 obtains, use the Canny edge detection method to obtain outline map.Because the direction of iris outline coboundary point should be consistent with the normal direction of iris outline, therefore when carrying out the Hough conversion, with the ballot of this constraint marginal point.Specifically, make that volume coordinate is that (x, the gradient vector of pixel y) is [I
x, I
y]
T, its gradient direction is θ
e=arctan (I
y/ I
x), this point is along with (x
c, y
c, be θ r) for the round normal vector deflection of the ballot of parameter
c, θ then
eShould satisfy
|θ
e-θ
c|≤ε
θ
ε wherein
θThe expression threshold value.Utilize the inner product of vector to express, can be changed into through the derivation following formula:
In first formula, need to calculate inverse trigonometric function arctan, and second formula includes only the multiplication that adds of floating number, avoided inverse trigonometric function to calculate arctan, improved the arithmetic speed of algorithm greatly.
2) confirm optimum iris outline
Because the existence of noise; Generally exist some to detect inaccurate marginal point on the edge of in the image; And the circle that radius is bigger in the Hough conversion tends to have more marginal points, so the outline that the maximum circle of ballot may not real corresponding iris in the Hough conversion.In order to address this problem, select 10 maximum candidate's circles of ballot, utilize following integro differential operator that these candidate's circles are assessed once more:
Here * representes convolution operation, G
σ(r) the expression standard deviation is the Gaussian function of σ,
The expression circular arc.In 10 candidate's circles, selecting the maximum circle of CID is the outline of iris and the integro differential operator value CID that exports this circle.The detected iris outline of circle expression among Fig. 4 (g).
Wherein the practical implementation step of step 4 is:
1) confirms candidate's image of closing one's eyes
Iris outline circle should have bigger CID, if so CID＜ε
Iris, ε wherein
IrisBe threshold value, then this image possibly be the image or that the iris outline detects in this image is inaccurate of closing one's eyes.
2) candidate's image binaryzation of closing one's eyes
The histogram of calculated candidate image also uses Gaussian function to carry out smoothly, the lower eyelash of first peak corresponding grey scale value in histogram left side.Select the paddy on this right side, peak image to be carried out binaryzation, then this bianry image is carried out the morphology expansive working, like Fig. 5 (a) and (b) with (c) as threshold value.
3) judge whether to be the image of closing one's eyes
Bianry image is carried out rim detection, use edge linking algorithm to find out the longest in an edge image edge, get this edge upper part and carry out fitting of parabola.Because Open Side Down for the para-curve that under the situation of opening eyes, simulates, and the parabolical opening direction that under the situation of closing one's eyes, simulates upwards, therefore can judge according to the para-curve opening direction whether this image is the image of closing one's eyes.Fig. 5 (d), (e) and (f) represented the associative operation of step 3).
Wherein the practical implementation step of step 5 is:
This method detects skin pixel and non-skin pixel through skin color detector, obtains binary image.The edge is detected in the bianry image back of expanding, go out to comprise the elliptic region of eyes, the definite again iris outline of describing with step 3 afterwards of improved Hough conversion based on circle with the Hough change detection.Introduce the method for extracting the human eye subimage based on skin color segmentation below.
At first choose the colored iris image of 5000 width of cloth and be used for training.In training set, the positive routine sample (skin pixel) of iris image and counter-example sample (non-skin pixel, as: eyelid, iris, sclera etc.) be mark by hand.With p (I|skin) and p (I|non-skin) expression skin and noncutaneous histogram, these two histograms are respectively through positive routine sample and counter-example sample calculation.Colour of skin sorter is definite through likelihood ratio, for a certain pixel, if
This pixel is the colour of skin, otherwise is the non-colour of skin.Threshold value can obtain through balance detection rate and false drop rate in test set.
Wherein the practical implementation step of step 6 is:
Because the iris outline confirms that the pupil profile center of circle is limited in the very little square area.In addition, according to statistical knowledge biologically, the chances are 3: 1 for iris radius and pupil radius ratio.So the span of pupil radius is limited in interval [r/4,2r/3], r is an iris outline radius here.At last through confirming the pupil profile, shown in Fig. 4 (h) or Fig. 6 (i) based on the integro differential operator of circle.
Wherein the practical implementation step of step 7 is:
1) confirms the marginal point in upper eyelid based on the one-dimensional signal detection method
On detected iris outline basis, detect the marginal point in upper eyelid along following one dimension straight vertical line segment:
x＝d
y∈[y
c-2R，y
c+R/2]
D is a constant, and span is [x
c-3R, x
c-R/2] perhaps [x
c+ R/2, x
c+ 3R].At first from image, extract one-dimensional signal and use Gaussian function to carry out smoothly along the one dimension straight-line segment, the derivative of signal calculated then, then to surpass the point of some threshold epsilon e be marginal point to signal derivative.Fig. 7 (a) has provided the hunting zone of one-dimensional signal, Fig. 7 (b), (c) and (d) represent the one-dimensional signal and the derivative thereof of one dimension straight vertical line segment, extraction respectively, and Fig. 7 (e) has provided detected upper eyelid marginal point.
2) the para-curve integro differential operator of belt restraining is confirmed the iris upper eyelid
Use the upper eyelid of the para-curve modeling iris of following form:
y＝a(x-b)
2+c
Can confirm the variation range of these three parameters according to the implication of parabolical three parameters.For the upper eyelid, the para-curve of match should be that Open Side Down, and its curvature should less than the circle curvature, promptly the span of a should be (0,0.5/R).According to the geometric relationship of eyelid and iris, can further confirm the para-curve summit (then the upper eyelid detection model is for b, span c):
Here * represents convolution operation, G
σ(c) be that standard deviation is the Gaussian function of σ, Γ (a, b, c) and L (a, b c) represent para-curve and length thereof respectively.
Integro differential operator is appreciated that into a kind of para-curve edge detection operator, and this operator calculates the variation of the average gray value of whole para-curve segmental arc.When in the para-curve parameter space, searching for, adopt a kind of technology of the RANSAC of being similar to algorithm to judge a certain tlv triple (a, b; Whether the para-curve of c) confirming corresponding possible upper eyelid: if; Upper eyelid detection model above then continue using calculate f (a, b, c); Otherwise abandon this tlv triple, continue to search at parameter space.Specifically, and a given tlv triple (a, b, c), at all marginal point p that is detected
i, i=1 ..., N
eIn, statistics satisfies marginal point and arrives parabolical distance less than a certain threshold epsilon
dNumber be N
pIf N
pWith N
eRatio surpassed a certain threshold epsilon
N, i.e. N
p/ N
e>=ε
N, this tlv triple can be used as effective candidate's tlv triple so.This method has been dwindled the scope of search volume on the one hand, makes the efficient of algorithm higher; Constraint through marginal point makes algorithm more accurate on the other hand.The last upper eyelid of confirming of Fig. 7 (f) expression.
Wherein the practical implementation step of step 8 is:
Along following one dimension straight vertical line segment detected edge points:
x＝d
y
c≤y≤y
c+3R/2
Here d is a constant, and variation range is [x
c-R/2, x
c+ R/2].Because the marginal point that is detected has some to be inaccurate, therefore adopt the RANSAC algorithm parabola of fit of robust, as shown in Figure 8.
Wherein the practical implementation step of step 9 is:
On the basis of detected iris inside and outside contour and last palpebra inferior, the pixel and the background pixel segmentation that will belong to iris are come out.Calculate the histogram of iris interior zone and use Gaussian function to carry out smoothly; The corresponding high bright spot in peak of level and smooth back histogram low order end; Get the threshold value of the paddy corresponding gray scale value in this left side, peak as high brightness point, the iris internal pixel values is removed as the high brightness point greater than the point of this threshold value.Fig. 9 (a) and (b) with (c) provided the segmentation result of iris image, the histogram of iris region (among the figure * expression high brightness threshold point) respectively and removed the iris region bianry image behind the high-brightness region.
Claims (1)
1.
A kind of iris splitting method that adapts to inferior quality iris image in the complex application context is characterized in that, said method comprising the steps of:
Step 1: a zonule of using human eye detection device based on the AdaBoost algorithm in iris image, to detect to comprise single eyes is as subimage;
Step 2: in the subimage that step 1 obtains, calculate grey level histogram and carry out smoothly with Gaussian function; Each crest in the histogram and the trough of these crest both sides are as the histogrammic sub-interval of symbiosis; As an end points this sub-range is further divided with this flex point if flex point is contained in this sub-range, thereby obtained the symbiosis histogram between the self-adaptation dividing regions; Adopt normalized gray scale symbiosis histogram as proper vector; Based on said proper vector; With the K-Means clustering algorithm subimage that step 1 obtains is cut apart; Generate edge image with the Canny edge detection method in the image after cutting apart, thereby use Hough conversion match eye contour to produce the elliptic region that comprises eyes based on ellipse;
Step 3:1) adopts the Hough conversion location iris outline of considering gradient direction: in the elliptic region that step 2 obtains, use the Canny edge detection method to obtain outline map; The ballot that utilizes the direction condition consistent of iris outline coboundary point to come constrained edge point with the normal direction of iris outline; Specifically, make volume coordinate do
The gradient vector of pixel do
, its gradient direction does
, this pixel along with
The normal vector deflection round for the ballot of parameter does
, then
Should satisfy
Wherein
The expression threshold value utilizes the inner product of vector to express, and draws through derivation:
2) confirm optimum iris outline: select 10 maximum candidate's circles of ballot, utilize following integro differential operator that these candidate's circles are assessed once more:
Here * representes convolution operation,
The expression standard deviation does
Gaussian function,
The expression circular arc; In 10 candidate's circles, selecting the maximum circle of integro differential operator value is the outline of iris and the integro differential operator value of exporting this circle
Step 4:1) if
, wherein
Be threshold value, show that then this iris image is the image or that the iris outline detects in this iris image is inaccurate of closing one's eyes;
2) to satisfying
Iris image carry out binaryzation; Bianry image to obtaining carries out rim detection, uses edge linking algorithm to find out the longest in an edge image edge, to carrying out fitting of parabola in this edge; If parabolical opening direction upwards; Judge that then this candidate image of closing one's eyes is the image of closing one's eyes,, judge that then this candidate image of closing one's eyes is the non-image of closing one's eyes if parabolical opening direction is downward;
Step 5: for the non-image of closing one's eyes; Detect skin pixel and non-skin pixel through skin color detector; Obtain the binary image formed by skin pixel and non-skin pixel; The edge is detected in the bianry image of being made up of skin pixel and the non-skin pixel back of expanding, go out to comprise the elliptic region of eyes with the Hough change detection, with the definite again iris outline of the Hough conversion of the described consideration gradient direction of step 3;
Step 6: the span of confirming the pupil center of circle according to the center of circle of the said iris outline of confirming again; The ratio of utilizing profile in human eye iris outline and the iris retrains the radius of pupil, confirms the pupil profile in iris inside based on the integro differential operator of circle;
Step 7:1) confirms the marginal point in upper eyelid: adopt following one dimension straight vertical line segment to detect the marginal point in iris upper eyelid: x=d, y ∈ [y
c
-2R, y
c
+ R/2], d is a constant, span is [x
c
-3R, x
c
-R/2] perhaps [x
c
+ R/2, x
c
+ 3R], from iris image, extract one-dimensional signal and use Gaussian function to carry out smoothly along said one dimension straight vertical line segment, the derivative of signal calculated then, the point that the derivative of the signal that calculates surpasses some threshold values is the marginal point in upper eyelid;
2) with the para-curve in para-curve integro differential operator match iris upper eyelid;
Step 8: use one dimension straight vertical line segment: x=d, y ∈ [y
c
, y
c
+ 3R/2], d is a constant, span is [x
c
-R/2, x
c
+ R/2] detect the marginal point of iris palpebra inferior, use the para-curve of RANSAC algorithm match iris palpebra inferior.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2009100722308A CN101923645B (en) | 2009-06-09 | 2009-06-09 | Iris splitting method suitable for low-quality iris image in complex application context |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2009100722308A CN101923645B (en) | 2009-06-09 | 2009-06-09 | Iris splitting method suitable for low-quality iris image in complex application context |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN101923645A CN101923645A (en) | 2010-12-22 |
| CN101923645B true CN101923645B (en) | 2012-06-27 |
Family
ID=43338568
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2009100722308A Expired - Fee Related CN101923645B (en) | 2009-06-09 | 2009-06-09 | Iris splitting method suitable for low-quality iris image in complex application context |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101923645B (en) |
Families Citing this family (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103377478A (en) * | 2012-04-25 | 2013-10-30 | 鸿富锦精密工业（深圳）有限公司 | Moving object detecting system and method |
| CN103093215B (en) * | 2013-02-01 | 2016-12-28 | 北京天诚盛业科技有限公司 | Human-eye positioning method and device |
| CN103531094A (en) * | 2013-09-22 | 2014-01-22 | 明基材料有限公司 | Display system and method capable of adjusting image focal length automatically |
| CN103632137B (en) * | 2013-11-15 | 2016-08-24 | 长沙理工大学 | A kind of human eye iris segmentation method |
| CN105224285A (en) * | 2014-05-27 | 2016-01-06 | 北京三星通信技术研究有限公司 | Eyes open and-shut mode pick-up unit and method |
| CN104050472B (en) * | 2014-06-12 | 2018-02-27 | 浙江工业大学 | A kind of adaptive global threshold method of Binary Sketch of Grey Scale Image |
| US9704038B2 (en) * | 2015-01-07 | 2017-07-11 | Microsoft Technology Licensing, Llc | Eye tracking |
| CN105160306B (en) * | 2015-08-11 | 2019-05-07 | 北京眼神智能科技有限公司 | The method and apparatus of iris image fuzzy Judgment |
| CN105488803A (en) * | 2015-12-09 | 2016-04-13 | 重庆康华瑞明科技股份有限公司 | Human eye pupil image judgment method |
| CN106169079B (en) * | 2016-06-30 | 2019-04-30 | 浙江工业大学 | A kind of pressure vessel quantity recognition methods based on computer vision |
| CN106485210B (en) * | 2016-09-26 | 2019-09-20 | 成都通甲优博科技有限责任公司 | A kind of iris detection method based on ellipses detection |
| CN108009495A (en) * | 2017-11-30 | 2018-05-08 | 西安科锐盛创新科技有限公司 | Fatigue driving method for early warning |
| CN108596187B (en) * | 2018-03-30 | 2023-07-04 | 青岛海尔智能技术研发有限公司 | Commodity purity detection method and display cabinet |
| CN109993115B (en) * | 2019-03-29 | 2021-09-10 | 京东方科技集团股份有限公司 | Image processing method and device and wearable device |
| CN110929672B (en) * | 2019-11-28 | 2024-03-01 | 联想(北京)有限公司 | Pupil positioning method and electronic equipment |
| CN111160113A (en) * | 2019-12-10 | 2020-05-15 | 中山市奥珀金属制品有限公司 | Iris positioning method, device and storage medium |
| CN111739006B (en) * | 2020-06-22 | 2021-07-13 | 深圳企业云科技股份有限公司 | Elliptical image detection algorithm and system based on enclosed road integral |
| CN113190117B (en) * | 2021-04-29 | 2023-02-03 | 南昌虚拟现实研究院股份有限公司 | Pupil and light spot positioning method, data calculation method and related device |
| CN113688874B (en) * | 2021-07-29 | 2024-05-31 | 天津中科智能识别产业技术研究院有限公司 | Automatic iris region segmentation method and system in human eye iris image |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1885314A (en) * | 2006-07-11 | 2006-12-27 | 电子科技大学 | Pre-processing method for iris image |
| CN101317183A (en) * | 2006-01-11 | 2008-12-03 | 三菱电机株式会社 | Method for localizing pixels representing an iris in an image acquired of an eye |
| CN101339603A (en) * | 2008-08-07 | 2009-01-07 | 电子科技大学中山学院 | Method for selecting qualified iris image from video frequency stream |
-
2009
- 2009-06-09 CN CN2009100722308A patent/CN101923645B/en not_active Expired - Fee Related
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101317183A (en) * | 2006-01-11 | 2008-12-03 | 三菱电机株式会社 | Method for localizing pixels representing an iris in an image acquired of an eye |
| CN1885314A (en) * | 2006-07-11 | 2006-12-27 | 电子科技大学 | Pre-processing method for iris image |
| CN101339603A (en) * | 2008-08-07 | 2009-01-07 | 电子科技大学中山学院 | Method for selecting qualified iris image from video frequency stream |
Non-Patent Citations (4)
| Title |
|---|
| Peihua Li, Xiaomin Liu.An Incremental Method for Accurate Iris Segmentation.《19th International Conference on Pattern Recognition,2008. ICPR 2008》.2008,1-4. * |
| PeihuaLi Xiaomin Liu.An Incremental Method for Accurate Iris Segmentation.《19th International Conference on Pattern Recognition |
| 岳学东，刘洋.一种新的虹膜图像预处理眼睑定位算法.《光电工程》.2008,第35卷(第8期),66-70. * |
| 罗忠亮.改进的虹膜图像分割算法.《重庆文理学院学报(自然科学版)》.2009,第28卷(第1期),59-62. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN101923645A (en) | 2010-12-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN101923645B (en) | Iris splitting method suitable for low-quality iris image in complex application context | |
| CN100452081C (en) | Human eye positioning and human eye state recognition method | |
| CN104091147B (en) | A kind of near-infrared eyes positioning and eye state identification method | |
| CN103093215B (en) | Human-eye positioning method and device | |
| CN103136504B (en) | Face identification method and device | |
| CN100592322C (en) | An automatic computer authentication method for photographic faces and living faces | |
| CN103473571B (en) | Human detection method | |
| CN100458831C (en) | Human face model training module and method, human face real-time certification system and method | |
| CN102708361B (en) | Human face collecting method at a distance | |
| CN102663413A (en) | Multi-gesture and cross-age oriented face image authentication method | |
| CN103942577A (en) | Identity identification method based on self-established sample library and composite characters in video monitoring | |
| CN102902967A (en) | Method for positioning iris and pupil based on eye structure classification | |
| CN102270308B (en) | Facial feature location method based on five sense organs related AAM (Active Appearance Model) | |
| CN103632136A (en) | Method and device for locating human eyes | |
| CN102567744B (en) | Method for determining quality of iris image based on machine learning | |
| CN101710383A (en) | Method and device for identity authentication | |
| CN102096823A (en) | Face detection method based on Gaussian model and minimum mean-square deviation | |
| KR20050025927A (en) | The pupil detection method and shape descriptor extraction method for a iris recognition, iris feature extraction apparatus and method, and iris recognition system and method using its | |
| CN103632132A (en) | Face detection and recognition method based on skin color segmentation and template matching | |
| CN102262729B (en) | Fused face recognition method based on integrated learning | |
| CN103886589A (en) | Goal-oriented automatic high-precision edge extraction method | |
| CN103218610B (en) | The forming method of dog face detector and dog face detecting method | |
| CN105138968A (en) | Face authentication method and device | |
| CN106203375A (en) | A kind of based on face in facial image with the pupil positioning method of human eye detection | |
| CN105718866A (en) | Visual target detection and identification method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20120627 Termination date: 20150609 |
|
| EXPY | Termination of patent right or utility model |