CN103824061A - Light-source-reflection-region-based iris positioning method for detecting and improving Hough conversion - Google Patents

Abstract

The invention provides a light-source-reflection-region-based iris positioning method for detecting and improving Hough conversion. The iris positioning method comprises the following steps: (1) carrying out a morphological algorithm on an original iris image to eliminate a light reflection region of a pupil region; before and after eliminating the light reflection region, carrying out binaryzation by using a pre-set threshold value; and determining a reflection region in a pupil and calculating a mean value, wherein a mean value point C is used as a standard point of an inner boundary; and (2) determining an area of interest of the inner boundary and a circular center radius range fitted by the inner boundary; carrying out improved Hough conversion on the area of interest of the inner boundary and finding out the inner boundary of an iris image; and finding out an outer boundary of the iris image on the basis of the inner boundary. The method sufficiently utilizes the characteristics of the iris image; in the positioning process of the inner and outer boundaries, a standard point C is used for setting and fitting the area of interest and the circular center radius range of the image for edge detection; the calculating amount of a fitting algorithm is greatly reduced; the influences on the fitting by noisy points including eyelids, eyelashes and the like are reduced.

Description

Detect and improve the iris locating method of Hough conversion based on light reflector region

Technical field

The present invention relates to a kind of iris image localization method that detects and improve Hough conversion based on light reflector region, belong to biometrics identification technology field.

Background technology

Living things feature recognition algorithm (Biometrics) utilizes mankind itself's physiology or behavioural characteristic to carry out person identification.The more bio-identification of application has the physiological characteristics such as face, iris, fingerprint, sound, vein and the behavioural characteristics such as action, gait of signing at present.These features vary with each individual, easy to carry and there is suitable stability.Living things feature recognition algorithm relies on these unique advantages, is applied widely in fields such as information security, financial transaction, social safety, health cares.

Compared with other biological spy, iris has the advantages such as very high stability, uniqueness and protection against trespasser, has obtained showing great attention to of scientific circles and industry member, and has obtained application in fields such as security protection, mining industry, finance.In the recognition system of iris, generally comprise pre-service, the feature extraction of iris and the characteristic matching of iris of iris.Wherein the pre-service of iris is the key of whole iris authentication system, and it provides effective information for follow-up feature extraction and cataloged procedure.In the pre-service the inside of iris image, emphasis is exactly the location of iris, and the location of iris is exactly the inner and outer boundary of definite iris in essence, thus inner and outer boundary correctness directly affect the accuracy of iris recognition.The basic step of Iris preprocessing is: (1) initialization pupil center; (2) determine iris inner boundary; (3) determine iris outer boundary.Iris boundary localization method is the earliest proposed by Daugman, i.e. classical integration/differentiating operator (Integro-differential operator); The algorithm that Wildes proposes combines rim detection with Hough conversion; The image zero crossing that the employing one dimension cubic spline wavelets such as Boles extract is as feature.Subsequently, people have introduced multiple types of tools in the iris segmentation stage, as least square fitting, movable contour model, Gabor wave filter etc.

There is following difficult point in iris boundary localization: the impact of 1 light: for example on eyes, occur retroreflective regions.2 block: the blocking of eyelashes, eyelid etc., eyes are almost closed.3: iris self gray scale is inhomogeneous, especially iris is abundanter near pupil part details.Therefore iris image quality greatly reduces, and very large difficulty has been brought in this accurate location that is iris boundary.For this reason, people have proposed diverse ways.For example

deng the figure such as average fuzzy clustering, Pundlik cut method, He etc. and proposed snake model and the angular integral sciagraphy of chord length equalization methods method, Jarjes etc.But the general calculated amount of these algorithms is very large, committed memory is many, and locating accuracy is not high.

Summary of the invention

The problems such as the calculated amount that exists for existing iris boundary localization technology is large, committed memory is many, locating accuracy is not high, the present invention proposes a kind of iris locating method that detects and improve Hough conversion based on light reflector region, the method can comparatively fast realize the accurate location on inside and outside border, has also suppressed to a certain extent the impact of burrs on edges point and part eyelashes point, eyelid.

The iris locating method that detects and improve Hough conversion based on light reflector region of the present invention, comprises the following steps:

(1) determine reference point: utilize (collecting device generally adopts symmetrical infrared light supply) in iris capturing process can in pupil, form this characteristic of reflective spot, original iris image is carried out to morphology operations, eliminate the retroreflective regions of pupil region, eliminating retroreflective regions front and back, carry out binaryzation by the threshold value of setting respectively, determine the interior retroreflective regions of pupil and calculate average, the reference point using average point C as inner boundary;

The detailed process of step (1) is as follows:

1. the iris image I (m, n) (m≤MI, n≤NI) of input is carried out to gaussian pyramid decomposition, MI and NI are respectively total line number and total columns of iris image, obtain general picture image I c;

2. according to the grey value characteristics of retroreflective regions, setting threshold Th, by general picture image I c binaryzation.Gray-scale value lower than the pixel of setting threshold Th is set to 0, obtains low-light level L region; Otherwise put 1, obtain high brightness H region, obtain the binary image Ic_b of general picture image I c;

$\left\{\begin{array}{c}\mathrm{Ic}\_b(i,j)=\left\{\begin{array}{c}1,\mathrm{ifIc}(i,j)\&GreaterEqual;\mathrm{Th}\\ 0,\mathrm{ifIc}(i,j)<\mathrm{Th}\end{array}\right.\\ \mathrm{Th}=\frac{1}{N}\underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}\mathrm{Ic}\_\max \left(j\right)\end{array}\right.,$

Wherein Ic_max (j) represents the pixel maximal value of j row, and N represents total columns of general picture image I c, and Th represents the average of Ic_max (j);

3. general picture image I c is carried out to morphology opening operation, the image I c_pupil after the inner retroreflective regions of the pupil that is eliminated;

4. according to the grey value characteristics of pupil region, setting threshold Tl, by the image I c_pupil binaryzation of eliminating after the inner retroreflective regions of pupil, is set to 0 lower than the gray-scale value of the pixel of setting threshold Tl, obtains low-light level L1 region; Otherwise put 1, obtain high brightness H1 region, the image I c_b1(after the image binaryzation being eliminated after retroreflective regions is the image after new binaryzation);

$\left\{\begin{array}{c}\mathrm{Ic}\_b1(i,j)=\left\{\begin{array}{c}1,\mathrm{ifIc}\_\mathrm{pupil}(i,j)\&GreaterEqual;\mathrm{Tl}\\ 0,\mathrm{ifIc}\_\mathrm{pupil}(i,j)<\mathrm{Tl}\end{array}\right.\\ \mathrm{Tl}=\frac{1}{N}\underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}\mathrm{Ic}\_\mathrm{pupil}\_\min \left(j\right)\end{array}\right.,$

Wherein Ic_pupil_min (j) represents the pixel minimum of j row, (total columns of image I c_pupil is identical with total columns of general picture image I c for total columns of N presentation video Ic_pupil, so also represent with N), Tl represents the average of Ic_pupil_min (j);

5. owing to only having retroreflective regions in pupil to belong to high-brightness region in the binary image Ic_b of general picture image and the low brightness area in image I c_b1 simultaneously, as can be seen here, in pupil, the average point C of reflective spot is just in pupil, average point C just can be served as the reference point of inner boundary location, C=(C _x, C _y).

(2) to iris inner boundary and outer boundary location: utilize reference point to determine the area-of-interest of inner boundary and the center of circle radius of inner boundary matching, the area-of-interest of inner boundary is improved to Hough conversion, find out iris image inner boundary; On the basis of inner boundary, determine the area-of-interest of outer boundary and the center of circle radius of outer boundary matching, the area-of-interest of outer boundary is carried out to same improvement Hough conversion, find out iris image outer boundary;

The detailed process of step (2) is as follows:

1. rim detection: general picture image I c is carried out to rim detection operation with canny operator, Canny operator asks marginal point specific algorithm step as follows: (a) use Gaussian filter smoothed image, (b) by single order local derviation finite difference compute gradient amplitude and direction, (c) gradient magnitude is carried out to non-maximum value inhibition method and come refinement border, after processing, obtain image I c_nmax, (d) finally use hysteresis threshold (dual threshold) binaryzation, high threshold TH_H is for detection of the large strong edge of gradient magnitude, low threshold value TH_L detects the weak edge being connected with strong edge and is connected edge, obtain image I c_icanny,

$\left\{\begin{array}{c}\mathrm{TH}\_H=\frac{1}{N}\underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}\mathrm{Ic}\_n\max \left(j\right)\\ \mathrm{TH}\_L=0.5*\mathrm{TH}\_H\end{array}\right.;$

2. determine inner boundary area-of-interest: set pupil radius [R _minr _max], and range of interest using this scope as radius parameter, make △=0.5* (R _max-R _min), △ is less than 5, sets row [2* △+C _x2* △+C _x], row [2* △+C _y2* △+C _y] Microcell between be the scope in the iris inner boundary center of circle, this scope is exactly between the region of interest of center of circle parameter; Scope to Hough conversion matching limits, reference point C=(C _x, C _y) in pupil inside, make D_interior=0.5* (R _min+ R _max), cut-away view is as the behavior [2*D_interior+C of Ic_icanny thus _x2*D_interior+C _x], classify [2*D_interior+C as _y2*D_interior+C _y] rectangular area as the area-of-interest of inner boundary;

3. determine inner boundary by improving Hough conversion:

Iris inner boundary can be expressed as equation of a circle, and due to not necessarily complete circle of rim detection inner boundary out, a large amount of frontier points are not on circle, but in inside and outside two or three pixel coverages of circle, so need to improve Hough conversion, then area-of-interest is carried out to matching, determine inner boundary; Concrete formula is as follows:

$H(x_c,y_c,r)=\underset{i=-2*D\_\mathrm{interior}+C_x}{\overset{2*D\_\mathrm{interior}+C_x}{\mathrm{\&Sigma;}}}\underset{j=-2*D\_\mathrm{interior}+C_y}{\overset{2*D\_\mathrm{interior}+C_y}{\mathrm{\&Sigma;}}}h(x_j,y_j,x_c,y_c,r),$

Wherein,

$\left\{\begin{array}{c}h(x_j,y_j,x_c,y_c,r)=\left\{\begin{array}{c}1,\mathrm{if}-10\&le;g(x_j,y_j,x_c,y_c,r)\&le;10\\ 0,\mathrm{else}\end{array}\right.\\ g(x_j,y_j,x_c,y_c,r)={(x_j-x_c)}^2+{(y_j-y_c)}^2-r^2\\ R_{\min }\&le;r\&le;R_{\max }\\ |x_c-C_x|\&le;2*\mathrm{\&Delta;}\\ |y_c-C_y|\&le;2*\mathrm{\&Delta;}\end{array}\right.,$

Wherein (x _j, y _j) expression marginal point coordinate, (x _c, y _c), r represents respectively the center of circle, radius parameter, g (x _j, y _j, x _c, y _c, r) represent marginal point (x _j, y _j) distance parameter is to being (x _c, y _c, the distance of equation of a circle r), h (x _j, y _j, x _c, y _c, r) judge whether marginal point satisfies condition, H (x _c, y _c, r) represent qualified number of edge points;

According to ballot method, for each marginal point (x _j, y _j), if can make-10≤g of this marginal point is (x _j, y _j, x _c, y _c, r)≤10, this point in parameter to being (x _c, y _c, circumference r) encloses; Therefore, by interative computation, the value of H is maximized, mean that the frontier point enclosing at this circumference is maximum, now parameter is to (x _c, y _c, the r) parameter of iris inner boundary namely;

4. improve hysteresis threshold:

In order to reduce the impact of inner boundary and the external edge fitting of part noise point minutiae point, need first these points to be carried out to zero setting; Specific practice is as follows:

The image I c_nmax on the border after canny operator maximum value is suppressed carries out denoising, and, higher than the value zero setting of threshold value T0, formula is as follows;

$\left\{\begin{array}{c}\mathrm{mask}1(i,j)=\left\{\begin{array}{cc}0,& \mathrm{ifIc}\_n\max (i,j)\&GreaterEqual;T0\\ \mathrm{Ic}\_n\max (i,j),& \mathrm{else}\end{array}\right.\\ T0=\frac{1}{N}\underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}\mathrm{Ic}\_n\max \_\max \left(j\right)\end{array}\right.,$

Wherein Ic_nmax_max (j) represents the pixel maximal value of j row, (total columns of image I c_nmax is identical with total columns of general picture image I c for total columns of N presentation video Ic_nmax, so also represent with N), T0 represents the average of Ic_nmax_max (j), by hysteresis threshold, mask1 is carried out to binaryzation again, obtain the image I c_ocanny after binaryzation;

5. determine outer boundary area-of-interest: establishing iris outer boundary radius is D_out, gets [D_out-△ D_out+ △] scope as outer boundary radius, set row [x _c-△ x _c+ △], row [y _c-△ y _c+ △] Microcell between be the scope in the iris outer boundary center of circle, this scope is exactly between the region of interest of center of circle parameter; Cut-away view is as Ic_ocanny behavior [0.6*D_out+x _cd_out+x _c], classify [1.2*D_out+y as _c1.2*D_iout+y _c] image section as effective outer boundary region;

6. determine outer boundary by improving Hough conversion: by step 3. described process improve Hough conversion at outer boundary area-of-interest, then outer boundary area-of-interest is carried out to matching, determine outer boundary.

Beneficial effect of the present invention is as follows:

(1) make full use of the feature of iris image.Because the infrared light supply of collecting device is all symmetric design, the reflective spot that infrared light supply forms in pupil also can have certain symmetry.So, the average point C(of these points is reference point), be bound to drop on the inside of pupil, due to the symmetry of light source, C point approaches pupil center conventionally.There is this reference point C, just can set the scope of the inside and outside edge fitting center of circle, radius in step 2, greatly saved the calculated amount of Hough conversion.

(2) in inside and outside boundary alignment process, utilize the image setting matching area-of-interest of reference point C edge detection and the scope of center of circle radius, greatly reduce fitting algorithm calculated amount, and reduced the impact of the noise such as eyelid, eyelashes on matching; In this interested region, adopt improvement Hough conversion to carry out, eliminated the impact of smeared out boundary on matching, realized accurate location internal, outer boundary.

Accompanying drawing explanation

Fig. 1 is general picture image I c.

Fig. 2 is the image I c_b after general picture image I c binaryzation.

Fig. 3 is image I c_pupil after general picture image I c opening operation.

Fig. 4 is the image I c_b1 of binaryzation again after general picture image I c opening operation.

Fig. 5 is the reference point schematic diagram of pupil inner boundary location.

Fig. 6 is the image after maximum value inhibition in edge detection process

Fig. 7 is hysteresis threshold image after treatment in edge detection process.

Fig. 8 is the marginal information schematic diagram in inner boundary area-of-interest.

Fig. 9 is the matching schematic diagram of iris inner boundary.

Figure 10 carries out the edge fitting schematic diagram after binaryzation by hysteresis threshold to mask1.

Figure 11 is the marginal information schematic diagram in outer boundary area-of-interest.

Figure 12 is the matching schematic diagram of outer boundary.

Embodiment

The present invention is based on light reflector region and detect and improve the iris locating method of Hough conversion, comprise two steps, detailed process is as follows.

(1) determine reference point: utilize iris capturing equipment generally to adopt symmetrical infrared light supply, this characteristic of reflective spot can be formed in pupil, original iris image is carried out to morphology operations, eliminate the retroreflective regions of pupil region.Eliminating retroreflective regions front and back, carry out binaryzation by suitable threshold value respectively, determine the interior retroreflective regions of pupil and calculate average, the reference point using average point as inside and outside boundary alignment.

The detailed process of step (1) is as follows:

1. the iris image I (m, n) (m≤MI, n≤NI) to input, MI, NI are respectively the row, column numbers of iris image, carry out gaussian pyramid decomposition, and this process is with matrix representation, and formula is as follows;

$\mathrm{Ic}=\underset{n=-2}{\overset{2}{\mathrm{\&Sigma;}}}\underset{m=-2}{\overset{2}{\mathrm{\&Sigma;}}}w(m,n)I(2i+m,2j+n),j\&le;\frac{\mathrm{NI}}{2},i\&le;\frac{\mathrm{MI}}{2}---\left(1\right)$

Wherein w (m, n) is a gauss low frequency filter, it can guarantee low-pass characteristic keep reducing and expansion simultaneously after brightness level and smooth, typical w (m, n) 5*5 window is as follows:

$w(m,n)=\frac{1}{256}\left[\begin{array}{ccccc}1& 4& 6& 4& 1\\ 4& 16& 24& 16& 4\\ 6& 24& 36& 24& 6\\ 4& 16& 24& 16& 4\\ 1& 4& 6& 4& 1\end{array}\right]---\left(2\right)$

The general picture image of Ic presentation video I after gaussian pyramid decomposes, as shown in Figure 1.

2. the gray-scale value of considering retroreflective regions is higher, sets higher thresholds Th, by image I c binaryzation, is set to 0 lower than the gray-scale value of the pixel of this threshold value, obtains L region; Otherwise put 1, obtain H region, obtain the binary image Ic_b of general picture image;

$\left\{\begin{array}{c}\mathrm{Ic}\_b(i,j)=\left\{\begin{array}{c}1,\mathrm{ifIc}(i,j)\&GreaterEqual;\mathrm{Th}\\ 0,\mathrm{ifIc}(i,j)<\mathrm{Th}\end{array}\right.\\ \mathrm{Th}=\frac{1}{N}\underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}\mathrm{Ic}\_\max \left(j\right)\end{array}---\left(3\right)\right.$

Wherein Ic_max (j) represents the pixel maximal value of j row, total columns of N presentation video Ic, and Th represents the average of Ic_max (j).The object of this binaryzation operation is to obtain retroreflective regions, as shown in Figure 2;

3. general picture image I c is gone to reflective operation.The method of taking is morphology opening operation, i.e. corrosion+expand, and the object of opening operation operation is mainly to eliminate the reflective spot of small scale.The closed operation construction operator adopting is 5 × 5 rectangles, represents with b.

$\mathrm{Ic}\_\mathrm{pupil}=\left(\mathrm{L\&Theta;b}\right)\&CirclePlus;b$

$b=\left[\begin{array}{ccccc}1& 1& 1& 1& 1\\ 1& 1& 1& 1& 1\\ 1& 1& 1& 1& 1\\ 1& 1& 1& 1& 1\\ 1& 1& 1& 1& 1\end{array}\right]---\left(4\right)$

Wherein

Θ represents respectively to expand, erosion operation.The formula that burn into expands is as follows respectively:

f1=(fΘb)(s,t)=min{f(s+x,t+y)-b(x,y)|(s+x),(t+y)∈D _f;(x,y)∈D _b} (5)

$\mathrm{Ic}\_\mathrm{pupil}=(f1\&CirclePlus;b)(s,t)=\max \{f1(s-x,t-y)+b(x,y)|(s-x),(t-y)\&Element;D_{f1};(x,y)\&Element;D_b---\left(6\right)$

D _f, D _f1, D _bit is respectively the field of definition of f, f1, b;

After closed operation is processed, obtain image I c_pupil, eliminate the image after the inner retroreflective regions of pupil, as shown in Figure 3:

4. the gray-scale value of considering pupil region is lower, sets lower threshold value Tl, by the image I c_pupil binaryzation after opening operation, is set to 0 lower than the gray-scale value of the pixel of this threshold value, obtains low-light level L1 region; Otherwise put 1, obtain high brightness H1 region, obtain new binary image Ic_b1;

$\left\{\begin{array}{c}\mathrm{Ic}\_b1(i,j)=\left\{\begin{array}{c}1,\mathrm{ifIc}\_\mathrm{pupil}(i,j)\&GreaterEqual;\mathrm{Tl}\\ 0,\mathrm{ifIc}\_\mathrm{pupil}(i,j)<\mathrm{Tl}\end{array}\right.\\ \mathrm{Tl}=\frac{1}{N}\underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}\mathrm{Ic}\_\mathrm{pupil}\_\min \left(j\right)\end{array}\right.---\left(7\right)$

Wherein Ic_pupil_min (j) represents the little value of pixel of j row, total columns of N presentation video Ic_pupil, and Tl represents the average of Ic_pupil_min (j).The object of this binaryzation operation is to obtain pupil region, as shown in Figure 4;

5. owing to only having the retroreflective regions in pupil to belong to the L1 region in H region and the image I c_b1 in image I c_b simultaneously.As can be seen here, the average C of retroreflective regions is also just in pupil in pupil, and this C point just can be used as the reference point that inner boundary is located, as shown in Figure 5.

Find out that in image I c_b, in H region and image I c_b1 the overlapping region in L region, i.e. the retroreflective regions of pupil inside, calculates the barycenter C in this region,

$C=(C_x,C_y)=(\frac{C_{\mathrm{xsum}}}{N_{\mathrm{pupil}}},\frac{C_{\mathrm{Ysum}}}{N_{\mathrm{pupil}}}),---\left(8\right)$

Wherein, $\left\{\begin{array}{cc}{C}_{\mathrm{xsum}}=\mathrm{\&Sigma;i},& \\ C_{\mathrm{ysum}}=\mathrm{\&Sigma;j},& \mathrm{ifIc}\_b(i,j)=1,\mathrm{Ic}\_b1(i,j)=0,\\ N_{\mathrm{pupil}}=\mathrm{\&Sigma;}1,& \end{array}\right.$

Wherein (i, j) is image pixel point coordinate, N _pupilfor the number of pixels of retroreflective regions.C _xsum, C _ysumbe respectively the horizontal ordinate satisfying condition, the aggregate-value of ordinate, C _x, C _ybe respectively x, the y axial coordinate of reference point.

(2) to the inside and outside boundary alignment of iris: utilize reference point to determine area-of-interest, area-of-interest is improved to Hough conversion, find out iris image inner boundary; On the basis of inner boundary, determine the area-of-interest of outer boundary, then adopt similar method to find out iris image outer boundary.Detailed process is as follows.

Described step (2) specifically comprises the following steps:

1. rim detection: general picture image I c is carried out to rim detection operation with canny operator, Canny operator asks marginal point specific algorithm step as follows: (a) by Gaussian filter smoothed image (b) single order local derviation finite difference compute gradient amplitude and direction (c), gradient magnitude is carried out to non-maximum value inhibition method and come refinement border, after processing, obtain image I c_nmax, (d) finally use hysteresis threshold (dual threshold) binaryzation, high threshold is for detection of the large strong edge of gradient magnitude, the weak edge that low threshold test is connected with strong edge and be connected edge.Obtain image I c_icanny.As shown in Figure 6 and Figure 7.

2. inner boundary area-of-interest:

When image I c_icanny is directly carried out to the matching of Hough conversion circle, hunting zone is excessive, and can be subject to the noise interference such as eyelid, eyelashes, background.Be large with calculated amount, real-time is poor, and error rate is high.For improving accuracy rate, increase real-time, need to reduce matching and count, dwindle the parameter area of the center of circle, radius.Implementation method is as follows:

According to the logical light characteristic of pupil, pupil size can change along with the power of light, but the variation of pupil size always within the specific limits.Thus, when inner boundary is done to the matching of Hough conversion circle, the variation range that radius is set is [R _minr _max], and range of interest using this scope as radius parameter.

In order further to improve real-time, need to set the hunting zone in the center of circle.Because reference point C approaches the center of pupil, can make △=0.5* (R _max-R _min), general △ is less than 5, sets row [2* △+C of image I c_icanny _x2* △+C _x], row [2* △+C _y2* △+C _y] Microcell between be the scope in the iris inner boundary center of circle, this scope is exactly between the region of interest of center of circle parameter.

Finally, the scope of Hough conversion matching is limited.Reference point C=(C _x, C _y) in pupil inside, make D_interior=0.5* (R _min+ R _max), cut-away view is as the behavior [2*D_interior+C of Ic_icanny thus _x2*D_interior+C _x], classify [2*D_interior+C as _y2*D_interior+C _y] rectangular area as the area-of-interest of inner boundary, after intercepting as shown in Figure 8.As can be seen, in area-of-interest, the noise such as eyelashes, eyelid border is eliminated substantially, can not cause interference to the matching of inner boundary.

3. improve Hough conversion and determine inner boundary:

Iris inner boundary can be expressed as equation of a circle conventionally, and due to not necessarily complete circle of Boundary Detection inner boundary out, a large amount of frontier points are not on circle, but in circumference encloses two or three pixel coverages.In order to improve the robustness of matching, to Hough, suitable improvement has been done in conversion, then carries out matching at area-of-interest.Concrete formula is as follows:

$H(x_c,y_c,r)=\underset{i=-2*D\_\mathrm{interior}+C_x}{\overset{2*D\_\mathrm{interior}+C_x}{\mathrm{\&Sigma;}}}\underset{j=-2*D\_\mathrm{interior}+C_y}{\overset{2*D\_\mathrm{interior}+C_y}{\mathrm{\&Sigma;}}}h(x_j,y_j,x_c,y_c,r),---\left(9\right)$

Wherein,

$\left\{\begin{array}{c}h(x_j,y_j,x_c,y_c,r)=\left\{\begin{array}{c}1,\mathrm{if}-10\&le;g(x_j,y_j,x_c,y_c,r)\&le;10\\ 0,\mathrm{else}\end{array}\right.\\ g(x_j,y_j,x_c,y_c,r)={(x_j-x_c)}^2+{(y_j-y_c)}^2-r^2\\ R_{\min }\&le;r\&le;R_{\max }\\ |x_c-C_x|\&le;2*\mathrm{\&Delta;}\\ |y_c-C_y|\&le;2*\mathrm{\&Delta;}\end{array}\right.,$

According to ballot method, for each marginal point (x _j, y _j), if can make-10≤g of this marginal point is (x _j, y _j, x _c, y _c, r)≤10, this point in parameter to being (x _c, y _c, circumference r) encloses, and only needs the value maximum of H, means that the frontier point enclosing at this circumference is maximum, and this parameter is to (x _c, y _c, the r) parameter of iris inner boundary namely.The inner boundary of matching as shown in Figure 9.

4. improve hysteresis threshold:

Because the transition band width of outer boundary is wider, there is blocking of eyelid eyelashes toward contact, so outer boundary is fuzzyyer.In image I c_nmax, outer boundary gray-scale value is lower, and inner boundary gray-scale value and part noise point minutiae point pixel value are higher, in order to reduce the impact of inner boundary and the external edge fitting of part noise point minutiae point, needs first these points to be carried out to zero setting.

Specific practice is as follows: the image I c_nmax on the border after canny operator maximum value is suppressed carries out denoising higher than the value zero setting of threshold value T0, and formula is as follows;

$\left\{\begin{array}{c}\mathrm{mask}1(i,j)=\left\{\begin{array}{cc}0,& \mathrm{ifIc}\_n\max (i,j)\&GreaterEqual;T0\\ \mathrm{Ic}\_n\max (i,j),& \mathrm{else}\end{array}\right.\\ T0=\frac{1}{N}\underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}\mathrm{Ic}\_n\max \_\max \left(j\right)\end{array}\right.,---\left(10\right)$

Wherein Ic_nmax_max (j) represents the pixel maximal value of j row, total columns of N presentation video Ic_nmax, and T0 represents the average of Ic_nmax_max (j).By hysteresis threshold, mask1 is carried out to binaryzation again, the same inner boundary of disposal route, obtains the image I c_ocanny after binaryzation, as shown in figure 10.

5. determine outer boundary area-of-interest:

According to human eye feature, the size variation of the iris of human eye is limited, establishes iris outer boundary radius size D_out, considers the faint difference of different people's iris existence, gets [D_out-△ D_out+ △] scope as outer boundary radius.In addition, inner boundary is definite, and parameter is to being (x _c, y _c, r), the center of circle on the inside and outside border of circle matching is generally more approaching, overlaps even, sets row [x _c-△ x _c+ △], row [y _c-△ y _c+ △] Microcell between be the scope in the iris outer boundary center of circle, this scope is exactly between the region of interest of center of circle parameter.

As Fig. 9, the upper and lower two parts iris region that is positioned at pupil is often easily subject to blocking of eyelid, and the iris that is positioned at part iris region on pupil is more easily subject to the impact of eyelashes.This part iris region noise is more, can impact the matching of iris outer boundary.The iris portion that is positioned at pupil both sides generally can not blocked by eyelid, can select the iris region of this part as area-of-interest.Cut-away view is as the behavior [0.6*D_out+x of Ic_ocanny thus _cd_out+x _c], classify [1.2*D_out+y as _c1.2*D_iout+y _c] image section as effective outer boundary region, intercept after as shown in figure 11.

6. improve Hough conversion and determine outer boundary:

Iris outer boundary also can be expressed as equation of a circle conventionally, and method is identical with inner boundary, and concrete formula is as follows:

$H(x_{\mathrm{oc}},y_{\mathrm{oc}},r_o)=\underset{i=-0.6*D\_\mathrm{out}+x_c}{\overset{D\_\mathrm{out}+x_c}{\mathrm{\&Sigma;}}}\underset{j=-1.2*D\_\mathrm{out}+y_c}{\overset{1.2*D\_\mathrm{out}+y_c}{\mathrm{\&Sigma;}}}h(x_j,y_j,x_{\mathrm{oc}},y_{\mathrm{oc}},r_o),---\left(11\right)$

Wherein,

$\left\{\begin{array}{c}h(x_j,y_j,x_{\mathrm{oc}},y_{\mathrm{oc}},r_o)=\begin{array}{c}\left\{\begin{array}{c}1,\mathrm{if}-10\&le;g(x_j,y_j,x_{\mathrm{oc}},y_{\mathrm{oc}},r_o)\&le;10\\ 0,\mathrm{else}\end{array}\right.\end{array}\\ g(x_j,y_j,x_{\mathrm{oc}},y_{\mathrm{oc}},r_o)={(x_j-x_{\mathrm{oc}})}^2+{(y_j-y_{\mathrm{oc}})}^2-{r_o}^2\\ |r_o-D\_\mathrm{out}|\&le;\mathrm{\&Delta;}\\ |x_{\mathrm{oc}}-x_c|\&le;\mathrm{\&Delta;}\\ |y_{\mathrm{oc}}-y_c|\&le;\mathrm{\&Delta;}\end{array}\right.,$

In outer boundary area-of-interest, adopt equally ballot method, obtain the parameter of iris outer boundary Hough circle matching to (x _oc, y _oc, r _o).Draw outer boundary as shown in figure 12.

Claims (1)

1. an iris locating method that detects and improve Hough conversion based on light reflector region, is characterized in that: comprise the following steps:

(1) determine reference point: utilize in iris capturing process and can in pupil, form this characteristic of reflective spot, original iris image is carried out to morphology operations, eliminate the retroreflective regions of pupil region, eliminating retroreflective regions front and back, carry out binaryzation by the threshold value of setting respectively, determine the interior retroreflective regions of pupil and calculate average, the reference point using average point C as inner boundary;

The detailed process of step (1) is as follows:

1. the iris image I (m, n) to input, m≤MI, n≤NI, carries out gaussian pyramid decomposition, and MI and NI are respectively total line number and total columns of iris image, obtain general picture image I c;

2. according to the gray-scale value characteristic of retroreflective regions, setting threshold Th, by general picture image I c binaryzation.Gray-scale value lower than the pixel of setting threshold Th is set to 0, obtains low-light level L region; Otherwise put 1, obtain high brightness H region, obtain the binary image Ic_b of general picture image I c;

$\left\{\begin{array}{c}\mathrm{Ic}\_b(i,j)=\left\{\begin{array}{c}1,\mathrm{ifIc}(i,j)\&GreaterEqual;\mathrm{Th}\\ 0,\mathrm{ifIc}(i,j)<\mathrm{Th}\end{array}\right.\\ \mathrm{Th}=\frac{1}{N}\underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}\mathrm{Ic}\_\max \left(j\right)\end{array}\right.,$

Wherein Ic_max (j) represents the pixel maximal value of j row, and N represents total columns of general picture image I c, and Th represents the average of Ic_max (j);

3. general picture image I c is carried out to morphology opening operation, the image I c_pupil after the inner retroreflective regions of the pupil that is eliminated;

4. according to the gray-scale value characteristic of pupil region, setting threshold Tl, by the image I c_pupil binaryzation of eliminating after the inner retroreflective regions of pupil, is set to 0 lower than the gray-scale value of the pixel of setting threshold Tl, obtains low-light level L1 region; Otherwise put 1, obtain high brightness H1 region, the image I c_b1 after the image binaryzation being eliminated after retroreflective regions;

$\left\{\begin{array}{c}\mathrm{Ic}\_b1(i,j)=\left\{\begin{array}{c}1,\mathrm{ifIc}\_\mathrm{pupil}(i,j)\&GreaterEqual;\mathrm{Tl}\\ 0,\mathrm{ifIc}\_\mathrm{pupil}(i,j)<\mathrm{Tl}\end{array}\right.\\ \mathrm{Tl}=\frac{1}{N}\underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}\mathrm{Ic}\_\mathrm{pupil}\_\min \left(j\right)\end{array}\right.,$

Wherein Ic_pupil_min (j) represents the pixel minimum of j row, total columns of N presentation video Ic_pupil, and Tl represents the average of Ic_pupil_min (j);

5. owing to only having retroreflective regions in pupil to belong to high-brightness region in the binary image Ic_b of general picture image and the low brightness area in image I c_b1 simultaneously, as can be seen here, in pupil, the average point C of reflective spot is just in pupil, average point C just can be served as the reference point of inner boundary location, C=(C _x, C _y).

(2) to iris inner boundary and outer boundary location: utilize reference point to determine the area-of-interest of inner boundary and the center of circle radius of inner boundary matching, the area-of-interest of inner boundary is improved to Hough conversion, find out iris image inner boundary; On the basis of inner boundary, determine the area-of-interest of outer boundary and the center of circle radius of outer boundary matching, the area-of-interest of outer boundary is carried out to same improvement Hough conversion, find out iris image outer boundary;

The detailed process of step (2) is as follows:

1. rim detection: general picture image I c is carried out to rim detection operation with canny operator, Canny operator asks marginal point specific algorithm step as follows: (a) use Gaussian filter smoothed image, (b) by single order local derviation finite difference compute gradient amplitude and direction, (c) gradient magnitude is carried out to non-maximum value inhibition method and come refinement border, after processing, obtain image I c_nmax, (d) finally use hysteresis threshold binaryzation, high threshold TH_H is for detection of the large strong edge of gradient magnitude, low threshold value TH_L detects the weak edge being connected with strong edge and is connected edge, obtain image I c_icanny,

$\left\{\begin{array}{c}\mathrm{TH}\_H=\frac{1}{N}\underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}\mathrm{Ic}\_n\max \left(j\right)\\ \mathrm{TH}\_L=0.5*\mathrm{TH}\_H\end{array}\right.;$

2. determine inner boundary area-of-interest: set pupil radius [R _minr _max], and range of interest using this scope as radius parameter, make △=0.5* (R _max-R _min), △ is less than 5, sets row [2* △+C _x2* △+C _x], row [2* △+C _y2* △+C _y] Microcell between be the scope in the iris inner boundary center of circle, this scope is exactly between the region of interest of center of circle parameter; Scope to Hough conversion matching limits, reference point C=(C _x, C _y) in pupil inside, make D_interior=0.5* (R _min+ R _max), cut-away view is as the behavior [2*D_interior+C of Ic_icanny thus _x2*D_interior+C _x], classify [2*D_interior+C as _y2*D_interior+C _y] rectangular area as the area-of-interest of inner boundary;

3. determine inner boundary by improving Hough conversion:

Iris inner boundary can be expressed as equation of a circle, and due to not necessarily complete circle of rim detection inner boundary out, a large amount of frontier points are not on circle, but in inside and outside two or three pixel coverages of circle, so need to improve Hough conversion, then area-of-interest is carried out to matching, determine inner boundary; Concrete formula is as follows:

$H(x_c,y_c,r)=\underset{i=-2*D\_\mathrm{interior}+C_x}{\overset{2*D\_\mathrm{interior}+C_x}{\mathrm{\&Sigma;}}}\underset{j=-2*D\_\mathrm{interior}+C_y}{\overset{2*D\_\mathrm{interior}+C_y}{\mathrm{\&Sigma;}}}h(x_j,y_j,x_c,y_c,r),$

Wherein,

$\left\{\begin{array}{c}h(x_j,y_j,x_c,y_c,r)=\left\{\begin{array}{c}1,\mathrm{if}-10\&le;g(x_j,y_j,x_c,y_c,r)\&le;10\\ 0,\mathrm{else}\end{array}\right.\\ g(x_j,y_j,x_c,y_c,r)={(x_j-x_c)}^2+{(y_j-y_c)}^2-r^2\\ R_{\min }\&le;r\&le;R_{\max }\\ |x_c-C_x|\&le;2*\mathrm{\&Delta;}\\ |y_c-C_y|\&le;2*\mathrm{\&Delta;}\end{array}\right.,$

Wherein (x _j, y _j) expression marginal point coordinate, (x _c, y _c), r represents respectively the center of circle, radius parameter, g (x _j, y _j, x _c, y _c, r) represent marginal point (x _j, y _j) distance parameter is to being (x _c, y _c, the distance of equation of a circle r), h (x _j, y _j, x _c, y _c, r) judge whether marginal point satisfies condition, H (x _c, y _c, r) represent qualified number of edge points;

According to ballot method, for each marginal point (x _j, y _j), if can make-10≤g of this marginal point is (x _j, y _j, x _c, y _c, r)≤10, this point in parameter to being (x _c, y _c, circumference r) encloses; Therefore, by interative computation, the value of H is maximized, mean that the frontier point enclosing at this circumference is maximum, now parameter is to (x _c, y _c, the r) parameter of iris inner boundary namely;

4. improve hysteresis threshold:

In order to reduce the impact of inner boundary and the external edge fitting of part noise point minutiae point, need first these points to be carried out to zero setting; Specific practice is as follows:

The image I c_nmax on the border after canny operator maximum value is suppressed carries out denoising, and, higher than the value zero setting of threshold value T0, formula is as follows;

$\left\{\begin{array}{c}\mathrm{mask}1(i,j)=\left\{\begin{array}{cc}0,& \mathrm{ifIc}\_n\max (i,j)\&GreaterEqual;T0\\ \mathrm{Ic}\_n\max (i,j),& \mathrm{else}\end{array}\right.\\ T0=\frac{1}{N}\underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}\mathrm{Ic}\_n\max \_\max \left(j\right)\end{array}\right.,$

Wherein Ic_nmax_max (j) represents the pixel maximal value of j row, total columns of N presentation video Ic_nmax, T0 represents the average of Ic_nmax_max (j), then by hysteresis threshold, mask1 is carried out to binaryzation, obtains the image I c_ocanny after binaryzation;

5. determine outer boundary area-of-interest: establishing iris outer boundary radius is D_out, gets [D_out-△ D_out+ △] scope as outer boundary radius, set row [x _c-△ x _c+ △], row [y _c-△ y _c+ △] Microcell between be the scope in the iris outer boundary center of circle, this scope is exactly between the region of interest of center of circle parameter; Cut-away view is as Ic_ocanny behavior [0.6*D_out+x _cd_out+x _c], classify [1.2*D_out+y as _c1.2*D_iout+y _c] image section as effective outer boundary region;

6. determine outer boundary by improving Hough conversion: by step 3. described process improve Hough conversion at outer boundary area-of-interest, then outer boundary area-of-interest is carried out to matching, determine outer boundary.

Images