CN104268527A - Iris locating method based on radial gradient detection - Google Patents

Abstract

The invention discloses an iris locating method based on radial gradient detection. The method comprises the steps of estimating the position of the center of a pupil, dividing an iris area to be searched for with the position as the reference point, and spreading the iris area to be searched for into a rectangular area with a specific size according to polar coordinates; locating the inner boundary of the iris area through the radial gradient detection method; locating the outer boundary of the iris area through the radial gradient detection method, and carrying out fine adjustment on the reference radius of the outer boundary according to the radial gradient in a neighborhood. According to the technical scheme, it is not needed to carry out pointwise three-dimensional space search on the iris boundaries within a large range or carry out self-adaptation determination on an edge detection and area segmentation complexity threshold value, disturbing influence caused by light source image points and uneven illumination is small, rapid and accurate iris locating can be achieved, and therefore the recognition speed of an iris recognition system can be increased, and the recognition accuracy of the iris recognition system can be improved.

Description

A kind of iris locating method detected based on radial gradient

Technical field

The present invention relates to a kind of image partition method.More specifically, a kind of quick iris locating method for iris biological recognition system is related to.

Background technology

Iris recognition has gradually become study hotspot and the development trend of field of biological recognition with its accuracy, stability, security and the significant advantage such as untouchable.In iris authentication system, iris preprocessing (comprising Iris Location, Interference Detection, normalization and image enhaucament etc.) is the prerequisite of iris recognition, and wherein Iris Location is crucial.Only have accurate positioning, more effectively iris feature could be extracted, thus realize accurately identifying.But in actual applications, due to factors such as the hardware constraints of collecting device and the illumination variation of collection environment, often there is the situations such as the fuzzy and contrast of noise in various degree, details is low in the iris image got, this just brings difficulty to Iris Location.

At present in iris authentication system, conventional Iris-orientation Algorithm is mainly divided three classes: the first kind is that the infinitesimal analysis loop truss device iteration parameter asked for corresponding to radial gradient circulation integral maximal value that utilizes proposed based on Daugman carries out the algorithm of Iris Location; Equations of The Second Kind is that the parameter obtained corresponding to the maximum circumference of ballot that utilizes in Hough loop truss transformation search parameter space proposed based on Wildes carries out the algorithm of Iris Location; 3rd class is based on edge gradient filtering and binaryzation rim detection and carries out the algorithm of Iris Location in conjunction with circumference matching.The speed of Iris Location and precision are conflicting, and in order to seek best balance between the real-time and accuracy of system, these algorithms require that the iris image of input must be high-quality.But in actual applications, the scene gathering iris image is complicated and changeable often, require that it is unpractical for getting high-quality iris image all the time.

More than conventional Iris-orientation Algorithm is all utilize in iris image to change obvious feature to sclera region transition position pixel grey scale again from pupil to iris, in the parameter space of iris image rectangular coordinate system, found the most optimized parameter on the corresponding inside and outside border of iris by the method for region segmentation, thus realize Iris Location.Although these algorithms can obtain good locating effect under certain condition, there is following significantly shortcoming:

The iterative detection of 1.Daugman, the Hough transform of Wildes and circumference matching scheduling algorithm all need three dimensions (the row, column coordinate in the circumference center of circle and the radius length) search in the certain limit in iris image, iris boundary parameter being carried out to pointwise, complexity is high, consuming time longer.If can not limit search scope effectively, the requirement of real-time meeting iris authentication system certainly will be difficult to.

2. these algorithms are all comparatively responsive on impacts such as the noise in real image and uneven illuminations, the problems such as the edges of regions that the local edge Gray Level Jump especially caused by light source picture point and eyelashes etc. and uneven illumination cause is fuzzy, cause iris boundary detection efficiency too low, thus locate unsuccessfully.If effective iris image quality criterion and noise detection algorithm can not be introduced, the accuracy requirement meeting iris authentication system certainly will be difficult to.

3. in Iris Location process, need to input rational threshold value and carry out rim detection, and choosing of threshold value is relevant with the intensity profile statistical property of iris image, the iris image statistical property difference that different system collects is obvious, and selected threshold is more difficult adaptively.In actual applications, often determine that several empirical values are as threshold value to be chosen according to the imaging characteristic of particular system and a large amount of tests.The threshold value of non-self-adapting chooses the complicacy that can increase system and algorithm design, certainly will be difficult to the practicality requirement meeting iris authentication system.

Therefore, need to provide a kind of quick iris locating method for iris biological recognition system.

Summary of the invention

The object of the present invention is to provide a kind of iris locating method detected based on radial gradient, solve Iris Location speed slow and accurately cannot locate the problem of iris due to local disturbances.

For achieving the above object, the present invention adopts following technical proposals:

Based on the iris locating method that radial gradient detects, the method step comprises:

S1, estimate the possible rectangular coordinate (xp of pupil center location based on the relative position of light source picture point and pupil _p, yp _p);

S2, with the possible rectangular coordinate of the pupil center location of described estimation for initial point, iris region to be searched is expanded into rectangular area by polar coordinates;

S3, in described rectangular area, radial gradient detection method is utilized to locate the inner boundary of iris region;

S4, in described rectangular area, radial gradient detection method is utilized to locate the outer boundary of iris region.

Preferably, step S1 comprises step further:

S11, change light source battle array spread pattern, regulate the relative position of light source and camera to make light source picture point be gathered in certain in pupil;

The position of light source picture point may be there is in S12, search;

The position of pupil region may be there is in S13, search;

S14, estimate the inner rectangular coordinate comprising the pupil region center of light source picture point based on the relative position of possible light source picture point and pupil.

Preferably, step S12 comprises step further:

S121, to the smoothing process of iris image, formula is as follows:

imgfil＝imfilter(eyeimage,H)；

In formula, eyeimage is the iris image collected, and H is smothing filtering operator;

S122, the position utilizing Threshold segmentation and connected domain detection light source for searching picture point to exist, formula is as follows:

$\left\{\begin{array}{c}\mathrm{BWL}=(\mathrm{imgfil}>=T_{\mathrm{light}});\\ [\mathrm{label}\_\mathrm{light},\mathrm{numl}]=\mathrm{bwlabel}(\mathrm{BWL},8);\end{array}\right.$

In formula, numl is the number of the connected region that possible there is light source picture point, T _lightfor light source picture point gray scale detection threshold value;

S123, estimate the rectangular coordinate of each possible light source picture point, formula is as follows:

$\left\{\begin{array}{c}[\mathrm{xi},\mathrm{yi}]=\mathrm{find}(\mathrm{label}\_\mathrm{light}==i),i=\mathrm{1,2},...\mathrm{numl};\\ {\mathrm{xl}}_i=\mathrm{round}((\max \left(\mathrm{xi}\right)+\min \left(\mathrm{xi}\right))/2);\\ {\mathrm{yl}}_i=\mathrm{round}((\max \left(\mathrm{yi}\right)+\min \left(\mathrm{yi}\right))/2);\\ \mathrm{loc}\_\mathrm{light}={\left\{({\mathrm{xl}}_i,{\mathrm{yl}}_i)\right\}}_{i=\mathrm{1,2},...\mathrm{numl}};\end{array}\right.$

In formula, loc_light is that the row, column coordinate of rectangular coordinate corresponding to possible light source picture point center is to vector.

Preferably, step S13 comprises step further:

S131, iris image carried out to local and fill, formula is as follows:

imgcom＝imcomplement(imfill(imcomplement(imgfil),'holes'))；

S132, the position utilizing Threshold segmentation and connected domain detection search pupil region to exist, formula is as follows:

$\left\{\begin{array}{c}\mathrm{BWP}=(\mathrm{imgcom}<T_{\mathrm{pupil}});\\ [\mathrm{label}\_\mathrm{pupil},\mathrm{nump}]=\mathrm{bwlabel}(\mathrm{BWP},8);\end{array}\right.$

In formula, nump is the number of the connected region that possible there is pupil, T _pupilfor pupil region gray scale detection threshold value;

S133, the row, column coordinate estimating each possible pupil and radius thereof, formula is as follows:

$\left\{\begin{array}{c}[\mathrm{xj},\mathrm{yj}]=\mathrm{find}(\mathrm{label}\_\mathrm{pupil}==j),j=\mathrm{1,2},...\mathrm{nump};\\ {\mathrm{xp}}_j=\mathrm{round}((\max \left(\mathrm{xj}\right)+\min \left(\mathrm{xj}\right))/2);\\ {\mathrm{yp}}_j=\mathrm{round}((\max \left(\mathrm{yj}\right)+\min \left(\mathrm{yj}\right))/2);\\ {\mathrm{rp}}_j=\mathrm{round}(((\max \left(\mathrm{xj}\right)-\min \left(\mathrm{xj}\right))+(\max \left(\mathrm{yj}\right)-\min \left(\mathrm{yj}\right)))/4);\\ \mathrm{loc}\_\mathrm{pupil}={\left\{({\mathrm{xp}}_j,{\mathrm{yp}}_j)\right\}}_{j=\mathrm{1,2},...\mathrm{nump}};\\ \mathrm{rad}\_\mathrm{pupil}={\left\{{\mathrm{rp}}_j\right\}}_{j=\mathrm{1,2},...\mathrm{nump}};\end{array}\right.$

In formula, loc_pupil be the row, column coordinate of the corresponding rectangular coordinate of possible pupil center to vector, rad_pupil is radius vectors corresponding to possible pupil region.

Preferably, step S14 comprises step further:

S141, the inner pupil region comprising light source picture point of search, the rectangular coordinate (xp of estimation pupil center location _a, yp _a), formula is as follows:

$\left\{\begin{array}{c}({\mathrm{xp}}_a,{\mathrm{yp}}_a)=(\mathrm{xp},\mathrm{yp}),(\mathrm{xp},\mathrm{yp})\&Element;\mathrm{loc}\_\mathrm{pupil}\\ s.t.\sqrt{{(\mathrm{xp}-\mathrm{xl})}^2+{(\mathrm{yp}-\mathrm{yl})}^2}\&le;\mathrm{rp},(\mathrm{xl},\mathrm{yl})\&Element;\mathrm{loc}\_\mathrm{light},\mathrm{rp}\&Element;\mathrm{rad}\_\mathrm{pupil}\end{array}\right.;$

S142, the rectangular coordinate (xp that statistics pupil center location is possible _p, yp _p), formula is as follows:

-3<xp _p-xp _a<3∩-3<yp _p-yp _a<3。

Preferably, step S2 comprises further:

With the center coordinate of eye pupil (xp that each is possible _p, yp _p) be initial point, by its N _h× N _wthe rectangular neighborhood of size expands into N by polar coordinates _r× N _θthe rectangular area of size, wherein N _hand N _wbe the height and width of rectangular area respectively, method of deploying is by iris region N _h× N _winterior every bit (x, y) is mapped to the rectangular area N represented by polar coordinates (r, θ) _r× N _θin, formula is as follows:

I(x(r,θ),y(r,θ))→I(r,θ),I(x,y)∈imgcom

In formula, radius r is in radial position, and scope is [1, R _m], wherein R _mbe the iris boundary radius maximal value limited, radial sampling number is N _r; Angle θ along angle direction, scope be [0 °, 360 °), it is N that angular samples is counted _θ.

Preferably, step S3 comprises further:

The inner boundary setting iris region as (xp, yp), radius be rp;

With the iris inner boundary central coordinate of circle (xp that each is possible _p, yp _p) be initial point, carry out the step S2 launching rectangular area by polar coordinates, obtain a N _r× N _θrectangularly-sampled battle array I (r, θ, the xp of size _p, yp _p), each I obtains along r direction its radial gradient vector Grad (r, xp _p, yp _p), formula is as follows:

Grad(r)＝sum(I(r+1,:))-sum(I(r,:))；

In formula, r=1,2 ..., N _r-1;

Add up each radial gradient vector Grad (r, xp _p, yp _p) peak value, three-dimensional parameter vector (r, xp that wherein peak-peak is corresponding _p, yp _p) be the parameter of iris inner boundary, formula is as follows:

$\left\{\begin{array}{c}(\mathrm{xp},\mathrm{yp})=\underset{({\mathrm{xp}}_p,{\mathrm{yp}}_p)}{\arg \max }\left(\underset{r}{\max }\left(\mathrm{Grad}(r,{\mathrm{xp}}_p,{\mathrm{yp}}_p)\right)\right),\\ \mathrm{rp}=\underset{r}{\arg \max }\left(\mathrm{Grad}(r,\mathrm{xp},\mathrm{yp})\right)\end{array}\right.$

In formula, (xp, yp) and rp are respectively central coordinate of circle and the radius of iris inner boundary.

Preferably, step S4 comprises further:

The center of circle setting outer boundary as (xo, yo), radius be ro;

The N of the center of circle (xo, yo) in the inner boundary center of circle (xp, yp) of outer boundary _o× N _oin neighborhood, formula is as follows:

In formula, (xo _p, yo _p) be the possible coordinate in the exterior iris boundary center of circle, N _o≤ 5;

With the exterior iris boundary central coordinate of circle (xo that each is possible _p, yo _p) be initial point, carry out the step S2 launching rectangular area by polar coordinates, obtain a N _r× N _θrectangularly-sampled battle array I (r, θ, the xo of size _p, yo _p), each I obtains along r direction its radial gradient vector Grad (r ', xo _p, yo _p), formula is as follows:

Grad(r')＝sum(I(r'+1,:))-sum(I(r',:))；

In formula, r '=rp+1, rp+2 ..., N _r-1;

Add up each radial gradient vector Grad (r ', xo _p, yo _p) peak value, three-dimensional parameter vector that wherein peak-peak is corresponding (r ', xo _p, yo _p) be the parameter of exterior iris boundary, formula is as follows:

$\left\{\begin{array}{c}(\mathrm{xo},\mathrm{yo})=\underset{({\mathrm{xo}}_p,{\mathrm{yo}}_p)}{\arg \max }\left(\underset{r^{\&prime;}}{\max }\left(\mathrm{Grad}(r^{\&prime;},{\mathrm{xo}}_p,{\mathrm{yo}}_p)\right)\right),\\ {\mathrm{ro}}_{\mathrm{ref}}=\underset{r^{\&prime;}}{\arg \max }\left(\mathrm{Grad}(r^{\&prime;},\mathrm{xo},\mathrm{yo})\right)\end{array}\right.$

In formula, (xo, yo) and ro _refbe respectively the center of circle and the reference radius of exterior iris boundary.

Preferably, the method for locating exterior iris boundary reference radius comprises further:

According to the radial gradient in iris region outer boundary neighborhood △ r to the reference radius ro of described outer boundary _reffinely tune, formula is as follows:

$\mathrm{ro}=\left\{\begin{array}{c}\mathrm{round}\left(\mathrm{mean}({\mathrm{ro}}_{\mathrm{ref}}+\mathrm{\&Sigma;}{r}_{\mathrm{\&Delta;}})\right),\&Exists;\mathrm{Grad}(r_{\mathrm{\&Delta;}},\mathrm{xo},\mathrm{yo})\&GreaterEqual;\mathrm{\&delta;Grad}({\mathrm{ro}}_{\mathrm{ref}},\mathrm{xo},\mathrm{yo})\\ {\mathrm{ro}}_{\mathrm{ref}},\mathrm{else}\end{array}\right.$

In formula, r _△∈ [ro _ref-△ r, ro _ref) ∪ (ro _ref, ro _ref+ △ r], δ ∈ (0,1) is weight factor, and ro is exterior iris boundary radius.

Beneficial effect of the present invention is as follows:

Technical scheme of the present invention, avoid the three dimensions search on a large scale, iris boundary parameter being carried out to pointwise, and edge detects and the self-adaptation of the complicated threshold value of region segmentation is determined, and the disturbing effects such as light source image point and uneven illumination are less, accurate Iris Location fast can be realized, thus contribute to the recognition speed and the accuracy rate that improve iris authentication system.

Accompanying drawing explanation

Below in conjunction with accompanying drawing, the specific embodiment of the present invention is described in further detail.

Fig. 1 illustrates a kind of process flow diagram of the iris locating method based on radial gradient detection.

Embodiment

In order to be illustrated more clearly in the present invention, below in conjunction with preferred embodiments and drawings, the present invention is described further.Parts similar in accompanying drawing represent with identical Reference numeral.It will be appreciated by those skilled in the art that specifically described content is illustrative and nonrestrictive, should not limit the scope of the invention with this below.

A kind of concrete steps of the iris locating method based on radial gradient detection are:

The first step, obtains iris image, estimates the position of pupil center and it can be used as reference point to divide iris region to be searched.

First, may there is the position of light source picture point in search.There is interference in various degree in the iris image collected in reality, in order to eliminate the impact of high frequency interference on iris boundary localization, to the smoothing process of iris image:

imgfil＝imfilter(eyeimage,H)； (1)

In formula, eyeimage is the iris image collected, and H is the Gassian low-pass filter template of 3 × 3.In iris image, the gray-scale value of light source picture point is far longer than the gray-scale value of other pixels, utilizes the position that Threshold segmentation and connected domain detection light source for searching picture point may exist:

$\left\{\begin{array}{c}\mathrm{BWL}=(\mathrm{imgfil}>=T_{\mathrm{light}});\\ [\mathrm{label}\_\mathrm{light},\mathrm{numl}]=\mathrm{bwlabel}(\mathrm{BWL},8);\end{array}\right.---\left(2\right)$

In formula, numl is the number of light source picture point connected region, if T _light=225 is light source picture point gray scale detection threshold value.Estimate the row, column coordinate of each possible light source picture point:

$\left\{\begin{array}{c}[\mathrm{xi},\mathrm{yi}]=\mathrm{find}(\mathrm{label}\_\mathrm{light}==i),i=\mathrm{1,2},...\mathrm{numl};\\ {\mathrm{xl}}_i=\mathrm{round}((\max \left(\mathrm{xi}\right)+\min \left(\mathrm{xi}\right))/2);\\ {\mathrm{yl}}_i=\mathrm{round}((\max \left(\mathrm{yi}\right)+\min \left(\mathrm{yi}\right))/2);\\ \mathrm{loc}\_\mathrm{light}={\left\{({\mathrm{xl}}_i,{\mathrm{yl}}_i)\right\}}_{i=\mathrm{1,2},...\mathrm{numl}};\end{array}\right.---\left(3\right)$

In formula, loc_light is that the center row, column coordinate of possible light source picture point rectangular coordinate is to vector.

Then, may there is the position of pupil region in search.Owing to there is light source picture point in real pupil region, in order to eliminate the interference of light source picture point, local being carried out to iris image and fills:

imgcom＝imcomplement(imfill(imcomplement(imgfil),'holes'))； (4)

In iris image, the gray-scale value of pupil region is far smaller than the gray-scale value of other pixels, utilizes Threshold segmentation and connected domain to detect the position searched for pupil region and may exist:

$\left\{\begin{array}{c}\mathrm{BWP}=(\mathrm{imgcom}<T_{\mathrm{pupil}});\\ [\mathrm{label}\_\mathrm{pupil},\mathrm{nump}]=\mathrm{bwlabel}(\mathrm{BWP},8);\end{array}\right.---\left(5\right)$

In formula, nump is the number of pupil center's connected region, if T _pupil=30 is pupil region gray scale detection threshold value.Estimate row, column coordinate and the radius thereof of each possible pupil:

$\left\{\begin{array}{c}[\mathrm{xj},\mathrm{yj}]=\mathrm{find}(\mathrm{label}\_\mathrm{pupil}==j),j=\mathrm{1,2},...\mathrm{nump};\\ {\mathrm{xp}}_j=\mathrm{round}((\max \left(\mathrm{xj}\right)+\min \left(\mathrm{xj}\right))/2);\\ {\mathrm{yp}}_j=\mathrm{round}((\max \left(\mathrm{yj}\right)+\min \left(\mathrm{yj}\right))/2);\\ {\mathrm{rp}}_j=\mathrm{round}(((\max \left(\mathrm{xj}\right)-\min \left(\mathrm{xj}\right))+(\max \left(\mathrm{yj}\right)-\min \left(\mathrm{yj}\right)))/4);\\ \mathrm{loc}\_\mathrm{pupil}={\left\{({\mathrm{xp}}_j,{\mathrm{yp}}_j)\right\}}_{j=\mathrm{1,2},...\mathrm{nump}};\\ \mathrm{rad}\_\mathrm{pupil}={\left\{{\mathrm{rp}}_j\right\}}_{j=\mathrm{1,2},...\mathrm{nump}};\end{array}\right.---\left(6\right)$

In formula, loc_pupil be the row, column coordinate of possible pupil center's rectangular coordinate to vector, rad_pupil is the radius vectors of its correspondence.

Finally, by detecting possible light source picture point and the relative position of pupil, finding out the pupil region that inside comprises light source picture point and being real pupil, estimating its center (xp _a, yp _a) be:

$\left\{\begin{array}{c}({\mathrm{xp}}_a,{\mathrm{yp}}_a)=(\mathrm{xp},\mathrm{yp}),(\mathrm{xp},\mathrm{yp})\&Element;\mathrm{loc}\_\mathrm{pupil}\\ s.t.\sqrt{{(\mathrm{xp}-\mathrm{xl})}^2+{(\mathrm{yp}-\mathrm{yl})}^2}\&le;\mathrm{rp},(\mathrm{xl},\mathrm{yl})\&Element;\mathrm{loc}\_\mathrm{light},\mathrm{rp}\&Element;\mathrm{rad}\_\mathrm{pupil}\end{array}\right.---\left(7\right)$

The horizontal or vertical deviation of the physical location of pupil center and estimation position generally within 3 pixels, the rectangular coordinate (xp that namely pupil center location is possible _p, yp _p) meet:

$\left\{\begin{array}{c}({\mathrm{xp}}_r,{\mathrm{yp}}_r)\&Element;\left\{({\mathrm{xp}}_p,{\mathrm{yp}}_p)\right\},\\ s.t.-3<{\mathrm{xp}}_p-{\mathrm{xp}}_a<3\&cap;-3<{\mathrm{yp}}_p-{\mathrm{yp}}_a<3\end{array}\right.---\left(8\right)$

In formula, (xp _r, yp _r) be actual pupil center location coordinate.

Second step, expands into a certain size rectangular area by polar coordinates by iris region to be searched.

If the resolution of iris image imgcom is S _h× S _w, the resolution of iris region to be searched is N _s× N _s, wherein N _s=min{S _h, S _w}/2.With the center coordinate of eye pupil (xp that each is possible _p, yp _p) be initial point, by its N _s× N _sthe rectangular neighborhood of size expands into (N by polar coordinates _s/ 2) rectangular area of × 360 sizes, namely by sampling on radius and angle direction, is mapped in the rectangular area represented by polar coordinates (r, θ) by the every bit (x, y) in iris region to be searched:

I(x(r,θ),y(r,θ))→I(r,θ),I(x,y)∈imgcom (9)

In formula, r is in radial position, and scope is [1, N _s/ 2], radius sampling number is N _s/ 2; θ along angle direction, scope be [0 °, 360 °), it is 360 that angular samples is counted.

3rd step, utilizes radial gradient detection method to locate the inner boundary of iris region.

If the center of circle of iris region inner boundary is (xp, yp), radius is rp, utilizes radial gradient detection method to locate the inner boundary (xp, yp, rp) of iris region.

Estimating the central coordinate of circle iris inner boundary is in a first step (xp _a, yp _a), if there is (xp, yp) in its 5 × 5 neighborhood.With the iris inner boundary central coordinate of circle (xp that each is possible _p, yp _p) be initial point, obtain (a N according to the method for second step _s/ 2) rectangularly-sampled battle array I (r, θ, the xp of × 360 sizes _p, yp _p), each I obtains along r (OK) direction its radial gradient vector Grad (r, xp _p, yp _p):

Grad(r)＝sum(I(r+1,:))-sum(I(r,:))； (10)

In formula, r=1,2 ..., N _r-1.

Add up each radial gradient vector Grad (r, xp _p, yp _p) peak value, three-dimensional parameter vector (r, xp that wherein peak-peak (radial gradient maximal value) is corresponding _p, yp _p) be the parameter of iris inner boundary:

$\left\{\begin{array}{c}(\mathrm{xp},\mathrm{yp})=\underset{({\mathrm{xp}}_p,{\mathrm{yp}}_p)}{\arg \max }\left(\underset{r}{\max }\left(\mathrm{Grad}(r,{\mathrm{xp}}_p,{\mathrm{yp}}_p)\right)\right),\\ \mathrm{rp}=\underset{r}{\arg \max }\left(\mathrm{Grad}(r,\mathrm{xp},\mathrm{yp})\right)\end{array}\right.---\left(11\right)$

In formula, (xp, yp) and rp are respectively central coordinate of circle and the radius length of iris inner boundary.

4th step, utilizes radial gradient detection method to locate the outer boundary of iris region.

If the center of circle of iris region outer boundary is (xo, yo), radius is ro, utilizes radial gradient detection method to locate the outer boundary (xo, yo, ro) of iris region.

Although the inside and outside border non-concentric of iris region, the N of the center of circle (xo, yo) in the inner boundary center of circle (xp, yp) of outer boundary _o× N _oin neighborhood, that is:

In formula, (xo _p, yo _p) be the possible coordinate in the exterior iris boundary center of circle.In reality, setting N _o=5.

With the exterior iris boundary central coordinate of circle (xo that each is possible _p, yo _p) be initial point, obtain (a N according to the method for second step _s/ 2) rectangularly-sampled battle array I (r, θ, the xo of × 360 sizes _p, yo _p), each I obtains along r (OK) direction its radial gradient vector Grad (r ', xo _p, yo _p):

Grad(r')＝sum(I(r'+1,:))-sum(I(r',:))； (13)

In formula, r '=rp+1, rp+2 ..., N _r-1.

Add up each radial gradient vector Grad (r ', xo _p, yo _p) peak value, three-dimensional parameter vector that wherein peak-peak (radial gradient maximal value) is corresponding (r ', xo _p, yo _p) be the parameter of exterior iris boundary:

$\left\{\begin{array}{c}(\mathrm{xo},\mathrm{yo})=\underset{({\mathrm{xo}}_p,{\mathrm{yo}}_p)}{\arg \max }\left(\underset{r^{\&prime;}}{\max }\left(\mathrm{Grad}(r^{\&prime;},{\mathrm{xo}}_p,{\mathrm{yo}}_p)\right)\right),\\ {\mathrm{ro}}_{\mathrm{ref}}=\underset{r^{\&prime;}}{\arg \max }\left(\mathrm{Grad}(r^{\&prime;},\mathrm{xo},\mathrm{yo})\right)\end{array}\right.---\left(14\right)$

In formula, the center of circle that (xo, yo) is exterior iris boundary, ro _reffor reference radius.

In the iris image of reality, the pixel grey scale change in iris region outer boundary neighborhood △ r is comparatively mild, and region segmentation boundary does not have inner boundary place so obvious, therefore needs according to the reference radius ro of the radial gradient in neighborhood △ r to outer boundary _reffinely tune:

$\mathrm{ro}=\left\{\begin{array}{c}\mathrm{round}\left(\mathrm{mean}({\mathrm{ro}}_{\mathrm{ref}}+\mathrm{\&Sigma;}{r}_{\mathrm{\&Delta;}})\right),\&Exists;\mathrm{Grad}(r_{\mathrm{\&Delta;}},\mathrm{xo},\mathrm{yo})\&GreaterEqual;\mathrm{\&delta;Grad}({\mathrm{ro}}_{\mathrm{ref}},\mathrm{xo},\mathrm{yo})\\ {\mathrm{ro}}_{\mathrm{ref}},\mathrm{else}\end{array}\right.---\left(15\right)$

In formula, r _△∈ [ro _ref-△ r, ro _ref) ∪ (ro _ref, ro _ref+ △ r], δ ∈ (0,1) is weight factor, and ro is exterior iris boundary radius.In reality, if △ is r=5, δ=0.75.

By above step, inside and outside boundary parameter (xp, yp, rp) and (xo, yo, the ro) of iris region can be tried to achieve, complete Iris Location.

In sum, technical scheme of the present invention avoids the three dimensions search on a large scale, iris boundary parameter being carried out to pointwise, and edge detects and the self-adaptation of the complicated threshold value of region segmentation is determined, and the disturbing effect such as light source image point and uneven illumination is less.Test for the image in Institute of Automation, CAS iris database CASIA 1.0, the Iris-orientation Algorithm based on Hough loop truss of Wildes classics is on average consuming time is 2.97s, and the above-mentioned Iris-orientation Algorithm detected based on radial gradient on average the 0.14s of being only consuming time (CPU of computing machine is Intel i5-3470s, 4G internal memory, system is Windows XPSP3, and testing software is MATLAB R2012a).Therefore, above-mentioned algorithm can realize accurate Iris Location fast, thus contributes to the recognition speed and the accuracy rate that improve iris authentication system.

Obviously; the above embodiment of the present invention is only for example of the present invention is clearly described; and be not the restriction to embodiments of the present invention; for those of ordinary skill in the field; can also make other changes in different forms on the basis of the above description; here cannot give exhaustive to all embodiments, every belong to technical scheme of the present invention the apparent change of extending out or variation be still in the row of protection scope of the present invention.

Claims (9)

1., based on the iris locating method that radial gradient detects, it is characterized in that, the method step comprises:

S1, estimate the possible rectangular coordinate (xp of pupil center location based on the relative position of light source picture point and pupil _p, yp _p);

S2, with the possible rectangular coordinate of the pupil center location of described estimation for initial point, iris region to be searched is expanded into rectangular area by polar coordinates;

S3, in described rectangular area, radial gradient detection method is utilized to locate the inner boundary of iris region;

S4, in described rectangular area, radial gradient detection method is utilized to locate the outer boundary of iris region.

2., according to claim 1 based on the iris locating method that radial gradient detects, it is characterized in that, described step S1 comprises step further:

S11, change light source battle array spread pattern, regulate the relative position of light source and camera to make light source picture point be gathered in certain in pupil;

The position of light source picture point may be there is in S12, search;

The position of pupil region may be there is in S13, search;

S14, estimate the inner rectangular coordinate comprising the pupil region center of light source picture point based on the relative position of possible light source picture point and pupil.

3., according to claim 2 based on the iris locating method that radial gradient detects, it is characterized in that, described step S12 comprises step further:

S121, to the smoothing process of iris image, formula is as follows:

imgfil＝imfilter(eyeimage,H)；

In formula, eyeimage is the iris image collected, and H is smothing filtering operator;

S122, the position utilizing Threshold segmentation and connected domain detection light source for searching picture point to exist, formula is as follows:

$\left\{\begin{array}{c}\mathrm{BWL}=(\mathrm{imgfil}>=T_{\mathrm{light}});\\ [\mathrm{label}\_\mathrm{light},\mathrm{numl}]=\mathrm{bwlabel}(\mathrm{BWL},8);\end{array}\right.$

In formula, numl is the number of the connected region that possible there is light source picture point, T _lightfor light source picture point gray scale detection threshold value;

S123, estimate the rectangular coordinate of each possible light source picture point, formula is as follows:

$\left\{\begin{array}{c}[\mathrm{xi},\mathrm{yi}]=\mathrm{find}(\mathrm{label}\_\mathrm{light}==i),i=\mathrm{1,2},...\mathrm{numl};\\ {\mathrm{xl}}_i=\mathrm{round}((\max \left(\mathrm{xi}\right)+\min \left(\mathrm{xi}\right))/2);\\ {\mathrm{yl}}_i=\mathrm{round}((\max \left(\mathrm{yi}\right)+\min \left(\mathrm{yi}\right))/2);\\ \mathrm{loc}\_\mathrm{light}={\left\{({\mathrm{xl}}_i,{\mathrm{yl}}_i)\right\}}_{i=\mathrm{1,2},...\mathrm{numl}};\end{array}\right.$

In formula, loc_light is that the row, column coordinate of rectangular coordinate corresponding to possible light source picture point center is to vector.

4., according to claim 2 based on the iris locating method that radial gradient detects, it is characterized in that, described step S13 comprises step further:

S131, iris image carried out to local and fill, formula is as follows:

imgcom＝imcomplement(imfill(imcomplement(imgfil),'holes'))；

S132, the position utilizing Threshold segmentation and connected domain detection search pupil region to exist, formula is as follows:

$\left\{\begin{array}{c}\mathrm{BWP}=(\mathrm{imgcom}<T_{\mathrm{pupil}});\\ [\mathrm{label}\_\mathrm{pupil},\mathrm{nump}]=\mathrm{bwlabel}(\mathrm{BWP},8);\end{array}\right.$

In formula, nump is the number of the connected region that possible there is pupil, T _pupilfor pupil region gray scale detection threshold value;

S133, the row, column coordinate estimating each possible pupil and radius thereof, formula is as follows:

$\left\{\begin{array}{c}[\mathrm{xj},\mathrm{yj}]=\mathrm{find}(\mathrm{label}\_\mathrm{pupil}==j),j=\mathrm{1,2},...\mathrm{nump};\\ {\mathrm{xp}}_j=\mathrm{round}((\max \left(\mathrm{xj}\right)+\min \left(\mathrm{xj}\right))/2);\\ {\mathrm{yp}}_j=\mathrm{round}((\max \left(\mathrm{yj}\right)+\min \left(\mathrm{yj}\right))/2);\\ {\mathrm{rp}}_j=\mathrm{round}(((\max \left(\mathrm{xj}\right)-\min \left(\mathrm{xj}\right))+(\max \left(\mathrm{yj}\right)-\min \left(\mathrm{yj}\right)))/4);\\ \mathrm{loc}\_\mathrm{pupil}={\left\{({\mathrm{xp}}_j,{\mathrm{yp}}_j)\right\}}_{j=\mathrm{1,2},...\mathrm{nump}};\\ \mathrm{rad}\_\mathrm{pupil}={\left\{{\mathrm{rp}}_j\right\}}_{j=\mathrm{1,2},...\mathrm{nump}};\end{array}\right.$

In formula, loc_pupil be the row, column coordinate of the corresponding rectangular coordinate of possible pupil center to vector, rad_pupil is radius vectors corresponding to possible pupil region.

5., according to claim 2 based on the iris locating method that radial gradient detects, it is characterized in that, described step S14 comprises step further:

S141, the inner pupil region comprising light source picture point of search, the rectangular coordinate (xp of estimation pupil center location _a, yp _a), formula is as follows:

$\left\{\begin{array}{c}({\mathrm{xp}}_a,{\mathrm{yp}}_a)=(\mathrm{xp},\mathrm{yp}),(\mathrm{xp},\mathrm{yp})\&Element;\mathrm{loc}\_\mathrm{pupil}\\ s.t.\sqrt{{(\mathrm{xp}-\mathrm{xl})}^2+{(\mathrm{yp}-\mathrm{yl})}^2}\&le;\mathrm{rp},(\mathrm{xl},\mathrm{yl})\&Element;\mathrm{loc}\_\mathrm{light},\mathrm{rp}\&Element;\mathrm{rad}\_\mathrm{pupil}\end{array}\right.;$

S142, the rectangular coordinate (xp that statistics pupil center location is possible _p, yp _p), formula is as follows:

-3<xp _p-xp _a<3∩-3<yp _p-yp _a<3。

6., according to claim 1 based on the iris locating method that radial gradient detects, it is characterized in that, described step S2 comprises further:

With the center coordinate of eye pupil (xp that each is possible _p, yp _p) be initial point, by its N _h× N _wthe rectangular neighborhood of size expands into N by polar coordinates _r× N _θthe rectangular area of size, wherein N _hand N _wbe the height and width of rectangular area respectively, method of deploying is by iris region N _h× N _winterior every bit (x, y) is mapped to the rectangular area N represented by polar coordinates (r, θ) _r× N _θin, formula is as follows:

I(x(r,θ),y(r,θ))→I(r,θ),I(x,y)∈imgcom

In formula, radius r is in radial position, and scope is [1, R _m], wherein R _mbe the iris boundary radius maximal value limited, radial sampling number is N _r; Angle θ along angle direction, scope be [0 °, 360 °), it is N that angular samples is counted _θ.

7., according to claim 1 based on the iris locating method that radial gradient detects, it is characterized in that, described step S3 comprises further:

The inner boundary setting iris region as (xp, yp), radius be rp;

With the iris inner boundary central coordinate of circle (xp that each is possible _p, yp _p) be initial point, carry out the step S2 launching rectangular area by polar coordinates, obtain a N _r× N _θrectangularly-sampled battle array I (r, θ, the xp of size _p, yp _p), each I obtains along r direction its radial gradient vector Grad (r, xp _p, yp _p), formula is as follows:

Grad(r)＝sum(I(r+1,:))-sum(I(r,:))；

In formula, r=1,2 ..., N _r-1;

Add up each radial gradient vector Grad (r, xp _p, yp _p) peak value, three-dimensional parameter vector (r, xp that wherein peak-peak is corresponding _p, yp _p) be the parameter of iris inner boundary, formula is as follows:

$\left\{\begin{array}{c}(\mathrm{xp},\mathrm{yp})=\underset{({\mathrm{xp}}_p,{\mathrm{yp}}_p)}{\arg \max }\left(\underset{r}{\max }\left(\mathrm{Grad}(r,{\mathrm{xp}}_p,{\mathrm{yp}}_p)\right)\right),\\ \mathrm{rp}=\underset{r}{\arg \max }\left(\mathrm{Grad}(r,\mathrm{xp},\mathrm{yp})\right)\end{array}\right.$

In formula, (xp, yp) and rp are respectively central coordinate of circle and the radius of iris inner boundary.

8., according to claim 1 based on the iris locating method that radial gradient detects, it is characterized in that, described step S4 comprises further:

The center of circle setting outer boundary as (xo, yo), radius be ro;

The N of the center of circle (xo, yo) in the inner boundary center of circle (xp, yp) of outer boundary _o× N _oin neighborhood, formula is as follows:

In formula, (xo _p, yo _p) be the possible coordinate in the exterior iris boundary center of circle, N _o≤ 5;

With the exterior iris boundary central coordinate of circle (xo that each is possible _p, yo _p) be initial point, carry out the step S2 launching rectangular area by polar coordinates, obtain a N _r× N _θrectangularly-sampled battle array I (r, θ, the xo of size _p, yo _p), each I obtains along r direction its radial gradient vector Grad (r ', xo _p, yo _p), formula is as follows:

Grad(r')＝sum(I(r'+1,:))-sum(I(r',:))；

In formula, r '=rp+1, rp+2 ..., N _r-1;

Add up each radial gradient vector Grad (r ', xo _p, yo _p) peak value, three-dimensional parameter vector that wherein peak-peak is corresponding (r ', xo _p, yo _p) be the parameter of exterior iris boundary, formula is as follows:

$\left\{\begin{array}{c}(\mathrm{xo},\mathrm{yo})=\underset{({\mathrm{xo}}_p,{\mathrm{yo}}_p)}{\arg \max }\left(\underset{r^{\&prime;}}{\max }\left(\mathrm{Grad}(r^{\&prime;},{\mathrm{xo}}_p,{\mathrm{yo}}_p)\right)\right),\\ {\mathrm{ro}}_{\mathrm{ref}}=\underset{r^{\&prime;}}{\arg \max }\left(\mathrm{Grad}(r^{\&prime;},\mathrm{xo},\mathrm{yo})\right)\end{array}\right.$

In formula, (xo, yo) and ro _refbe respectively the center of circle and the reference radius of exterior iris boundary.

9., according to claim 8 based on the iris locating method that radial gradient detects, it is characterized in that, the method for described location exterior iris boundary reference radius comprises further:

According to the radial gradient in iris region outer boundary neighborhood △ r to the reference radius ro of described outer boundary _reffinely tune, formula is as follows:

$\mathrm{ro}=\left\{\begin{array}{c}\mathrm{round}\left(\mathrm{mean}({\mathrm{ro}}_{\mathrm{ref}}+\mathrm{\&Sigma;}{r}_{\mathrm{\&Delta;}})\right),\&Exists;\mathrm{Grad}(r_{\mathrm{\&Delta;}},\mathrm{xo},\mathrm{yo})\&GreaterEqual;\mathrm{\&delta;Grad}({\mathrm{ro}}_{\mathrm{ref}},\mathrm{xo},\mathrm{yo})\\ {\mathrm{ro}}_{\mathrm{ref}},\mathrm{else}\end{array}\right.$

In formula, r _△∈ [ro _ref-△ r, ro _ref) ∪ (ro _ref, ro _ref+ △ r], δ ∈ (0,1) is weight factor, and ro is exterior iris boundary radius.