CN106127160A - A kind of human eye method for rapidly positioning for iris identification - Google Patents

Abstract

The present invention relates to computer vision and image identification technical field, it is provided that a kind of human eye method for rapidly positioning for iris identification, including: step one, input testing image, use Face datection algorithm, extract facial image；Step 2, facial image a kind of for step is converted to gray level image, gray level image slightly extracts human eye area image；Step 3, according to the human eye area image obtained in step 2, calculate accurate human eye centre coordinate.The method slightly extracting human eye area image uses gray-level projection algorithm；The method of the accurate human eye centre coordinate calculated uses gradient algorithm.Present invention utilizes integral projection and two important characteristics of image of gradient, remain to normally work in the case of illumination condition is bad, there is stronger robustness；Have employed human eye coarse positioning and be accurately positioned the mode combined, it is not necessary to calculate all regions of image, reduce computation complexity, operate more rapid, human eye location is more accurate.

Description

A kind of human eye method for rapidly positioning for iris identification

Technical field

The present invention relates to computer vision and image identification technical field, particularly to a kind of human eye for iris identification Method for rapidly positioning.

Background technology

Along with high speed development and the arrival of information age of modern society, identity recognizing technology becomes more and more important, phase For the identification technology that some are traditional, living things feature recognition has the advantages such as safer, more convenient, more secrecy, common fingerprint Identifying, recognition of face broadly falls into living things feature recognition, but the iris identification to be belonged to that wherein safety is the highest, the coding of iris identification Space is enriched, and everyone iris lines is unique and easily distinguishes.Moreover, the iris in human body is not being shared the same light Can show different shapes, the most only living body iris under the conditions of according to can be by detection, and this has just stopped similar duplication The potential safety hazard of fingerprint etc.

Although iris identification safety is higher, however it is necessary that the spacing of user and equipment is close and just can be identified, Fine on Consumer's Experience, distant range iris identification be compare in recent years burning the hotest identification technology, user and equipment it Between have only to keep normal separation (between about 0.5 meter to 1 meter), equipment just can collect iris information, quickly identifies.

But, because the collection of iris information needs the equipment (typically 1080P and more than) of very high resolution, if Remote capture iris information, that will take for a lot of time because have a lot of image-regions include other parts of face and Background image is unable to bring useful information for iris identification, and this is also why iris identification closely identifies often Reason.

The method of human eye location has a lot at present, and algorithm in early days has Gray Projection integration, template matching, Hough circle to become Change, after have again Corner Detection, symmetry transformation method, isophote scheduling algorithm, in recent years, along with the development of machine learning algorithm, Application in terms of human eye location also gets more and more, as former with what Adaboost was sorted in Face datection according to Haar feature Reason, positions human eye also with this method, in addition with SVM (support vector machine), and KNN (k nearest neighbor) scheduling algorithm, but this A little algorithms or computation complexity is the highest, otherwise positioning precision is inadequate, it is impossible to meet the oculocentric demand of real-time positioning people.

Therefore, computer vision is badly in need of a kind of human eye for iris identification quickly side of location with image identification technical field Method, make use of integral projection and two important characteristics of image of gradient, in the case of illumination condition is bad, it is also possible to normal work Make, there is stronger robustness；Have employed human eye coarse positioning and be accurately positioned the mode combined, it is to avoid calculate all districts of image Territory, reduces computation complexity, and more quickly, easily, human eye location is more accurate in operation.

Summary of the invention

The present invention is to solve the problems referred to above, it is provided that a kind of human eye method for rapidly positioning for iris identification, technology Scheme is as follows:

A kind of human eye method for rapidly positioning for iris identification, comprises the steps:

Step one, input testing image, use Face datection algorithm, extract facial image；

Step 2, facial image a kind of for step is converted to gray level image, gray level image slightly extracts human eye district Area image；

Step 3, according to the human eye area image obtained in step 2, calculate accurate human eye centre coordinate.

Preferably, in above-mentioned a kind of convex mirror image-forming correction method, concretely comprising the following steps of gray-level projection algorithm:

Concretely comprising the following steps of gray-level projection algorithm:

A) testing image is carried out Face datection, i-th facial image is designated as?Middle know according to priori Know and carry out human eye location, i.e. draw left eye, right eye candidate region, be designated as respectivelyWith

B) calculation procedure A) in drawWithIntegration in the horizontal direction also obtains normalization result；

$h\left(y\right)=\frac{1}{y_{max}-y_{min}}{\∫}_{y_{\min }}^{y_{max}}I(x,y)dx$

Wherein, (x y) represents facial image (x, y) gray value put, y to I_max、y_minObtain according to priori, table Showing the scope of eye areas abscissa, h (y) representsOrNormalization projection function in the horizontal direction；

C) projection function h (y) is calculated respectively at the interval [x of human eye vertical coordinate_min,x_maxAverage in] and standard deviation, use Average deducts the standard deviation of certain multiple as dynamic threshold T (h (y))；Filter out all indulging less than dynamic threshold T (h (y)) Coordinate, i.e. gathered y | h (y) >=T (h (y)) }, y | the vertical coordinate of h (y) expression h (y), and according to vertical coordinate from small to large Being ranked up, in taking-up vertical coordinate sequence, the vertical coordinate of 1/4th quantiles is as with reference to vertical coordinate x, with this reference vertical coordinate Centered by x, 0.1 width as interval band, the i.e. Range Representation of human eye vertical coordinate be [x-0.1 × height, x+0.1 × Height, height represent the height of facial image, further, take [x, (ymax+ymin)/2] human eye as coarse positioning Centre coordinate；

D) by step C) in centered by the human eye centre coordinate obtained and priori coarse localization human eye area image D_iris, D_irisCoordinate range is: abscissa [y_max,y_min], vertical coordinate scope [x-0.1*height, x+0.1*height, slightly fixed Position human eye centre coordinate is [x, (ymax+ymin)/2].

Preferably, in above-mentioned a kind of convex mirror image-forming correction method, in the accurate human eye calculated in step 3 The heart is sat calibration method and is used gradient algorithm.

Preferably, in above-mentioned a kind of convex mirror image-forming correction method, concretely comprising the following steps of gradient algorithm:

1) for the human eye area image D of coarse positioning in step 2_iris, for the gray value I of gray level image in step one (x y), calculates GradConcrete formula is as follows:

$G_x(x,y)=\left[\begin{array}{ccc}-1& 0& 1\\ -2& 0& 2\\ -1& 0& 1\end{array}\right]\×I(x,y);$

$G_y(x,y)=\left[\begin{array}{ccc}1& 2& 1\\ 0& 0& 0\\ -1& -2& -1\end{array}\right]\×I(x,y);$

$\▿\·I(x,y)=\sqrt{G_x{(x,y)}^2+G_y{(x,y)}^2}$

Wherein, G_x(x, y) and G_y(x, y) represents x direction and the Grad in y direction respectively,

2) in order to try to achieve human eye centre coordinate c, then the formula of object function is:

$F\left(c\right)={\mathrm{\Σ}}_{i=1}^N({\stackrel{\→}{d}}_i\·{\stackrel{\→}{g}}_i)$

${\stackrel{\→}{d}}_i=\frac{x_i-c}{\left|\right|x_i-c|{|}_2}$

${\stackrel{\→}{g}}_i=\frac{{{\stackrel{\→}{g}}_i}^{\′}}{\left|\right|{{\stackrel{\→}{g}}_i}^{\′}|{|}_2}$

${\stackrel{\→}{g}}_i^{\′}=\▿\·I\left(x\right){|}_{(x=x_i)}$

Wherein, x₁,x₂,…x_i…,x_NRepresent the pixel in gray level image,Represent x_iThe gradient of point Value,Represent some x_iNormalized vector between a c 2,Represent some x_iPlace's gradient vector,Represent some x_iPlace gradient to The normalized value of amount, F (c) represents the object function of human eye central point c；

3) for object function F (c), it is optimized, is chosen the maximum of object function F (c) after optimization, this F The coordinate of c central point C that () maximum is corresponding as accurate human eye centre coordinate, the formula after optimization is:

$F\left(c\right)={\mathrm{\Σ}}_{i=1}^N\mathrm{\ω}\left(c\right)\·max({\stackrel{\→}{d}}_i\·{\stackrel{\→}{g}}_i,0),$

W (c)=(Max-I (x, y))^r

Wherein, w (c) represents the weighted value of human eye centre coordinate C, and Max represents the maximum of pixel in gray level image.

Beneficial effects of the present invention:

1, present invention employs human eye coarse positioning and be accurately positioned the mode combined, it is to avoid calculate all regions of image, Reduce computation complexity.

2, present invention utilizes integral projection and two important characteristics of image of gradient, in the situation that illumination condition is bad Under, it is also possible to normally work, there is stronger robustness.

3, the present invention is when calculating gradient object function, eliminates the gradient vector that part mould length is shorter, have employed segmentation Choose the mode of object function maximum, reduce computation complexity.

4, the present invention can combine with iris identification, by quickly human eye center, location, extracts image around iris, keeps away Exempt from iris identification and calculated face Zone Full at the very start, improve accuracy and the real-time of identification.

Accompanying drawing explanation

The present invention is described below in conjunction with the accompanying drawings in detail with detailed description of the invention:

Fig. 1 is the flow chart of a kind of human eye method for rapidly positioning for iris identification.

Fig. 2 is that C point is positioned at central point and the structural representation of non-central point.

Detailed description of the invention

For the measure making the technology of the present invention realize, creation characteristic, reach purpose and be easy to understand with effect, below will In conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is clearly and completely described, it is clear that Described embodiment is only a part of embodiment of the present invention rather than whole embodiments.Based on the enforcement in the present invention Example, the every other embodiment that those of ordinary skill in the art are obtained under not making creative work premise, broadly fall into The scope of protection of the invention.

Embodiment 1:

Fig. 1 is the flow chart of a kind of human eye method for rapidly positioning for iris identification.

As it is shown in figure 1, a kind of human eye method for rapidly positioning for iris identification, comprise the steps:

Step one, input testing image, use Face datection algorithm, extract facial image；

Step 2, facial image a kind of for step is converted to gray level image, gray level image slightly extracts human eye district Area image；

Step 3, according to the human eye area image obtained in step 2, calculate accurate human eye centre coordinate.

Embodiment 2:

Fig. 1 is the flow chart of a kind of human eye method for rapidly positioning for iris identification.

As it is shown in figure 1, a kind of human eye method for rapidly positioning for iris identification, comprise the steps:

Step one, input testing image, use Face datection algorithm, extract facial image；

Step 2, facial image a kind of for step is converted to gray level image, utilizes gray-level projection algorithm slightly to extract Go out human eye area image；

Concretely comprising the following steps of gray-level projection algorithm:

A) testing image is carried out Face datection, i-th facial image is designated as?Middle know according to priori Know and carry out human eye location, i.e. draw left eye, right eye candidate region, be designated as respectivelyWith

B) calculation procedure A) in drawWithIntegration in the horizontal direction also obtains normalization result；

$h\left(y\right)=\frac{1}{y_{max}-y_{min}}{\∫}_{y_{min}}^{y_{max}}I(x,y)dx$

Wherein, (x y) represents facial image (x, y) gray value put, y to I_max、y_minObtain according to priori, table Showing the scope of eye areas abscissa, h (y) representsOrNormalization projection function in the horizontal direction；

C) projection function h (y) is calculated respectively at the interval [x of human eye vertical coordinate_min,x_maxAverage in] and standard deviation, use Average deducts the standard deviation of certain multiple as dynamic threshold T (h (y))；Filter out all indulging less than dynamic threshold T (h (y)) Coordinate, i.e. gathered y | h (y) >=T (h (y)) }, y | the vertical coordinate of h (y) expression h (y), and according to vertical coordinate from small to large Being ranked up, in taking-up vertical coordinate sequence, the vertical coordinate of 1/4th quantiles is as with reference to vertical coordinate x, with this reference vertical coordinate Centered by x, 0.1 width as interval band, the i.e. Range Representation of human eye vertical coordinate be [x-0.1 × height, x+0.1 × Height, height represent the height of facial image, further, take [x, (ymax+ymin)/2] human eye as coarse positioning Centre coordinate；

D) by step C) in centered by the human eye centre coordinate obtained and priori coarse localization human eye area image D_iris, D_irisCoordinate range is: abscissa [y_max,y_min], vertical coordinate scope [x-0.1*height, x+0.1*height, slightly fixed Position human eye centre coordinate is [x, (ymax+ymin)/2]；

Step 3, according to the human eye area image D obtained in step 2_iris, utilize gradient algorithm to calculate accurate people Eye centre coordinate；

Concretely comprising the following steps of gradient algorithm:

1) for the human eye area image D of coarse positioning in step 2_iris, for the gray value I of gray level image in step one (x y), calculates GradConcrete formula is as follows:

$G_x(x,y)=\left[\begin{array}{ccc}-1& 0& 1\\ -2& 0& 2\\ -1& 0& 1\end{array}\right]\×I(x,y);$

$G_y(x,y)=\left[\begin{array}{ccc}1& 2& 1\\ 0& 0& 0\\ -1& -2& -1\end{array}\right]\×I(x,y);$

$\▿\·I(x,y)=\sqrt{G_x{(x,y)}^2+G_y{(x,y)}^2}$

Wherein, G_x(x, y) and G_y(x, y) represents x direction and the Grad in y direction respectively,

2) in order to try to achieve human eye centre coordinate c, then the formula of object function is:

$F\left(c\right)={\mathrm{\Σ}}_{i=1}^N({\stackrel{\→}{d}}_i\·{\stackrel{\→}{g}}_i)$

${\stackrel{\→}{d}}_i=\frac{x_i-c}{\left|\right|x_i-c|{|}_2}$

${\stackrel{\→}{g}}_i=\frac{{{\stackrel{\→}{g}}_i}^{\′}}{\left|\right|{{\stackrel{\→}{g}}_i}^{\′}|{|}_2}$

${\stackrel{\→}{g}}_i^{\′}=\▿\·I\left(x\right){|}_{(x=x_i)}$

Wherein, x₁,x₂,…x_i…,x_NRepresent the pixel in gray level image,Represent x_iThe gradient of point Value,Represent some x_iNormalized vector between a c 2,Represent some x_iPlace's gradient vector,Represent some x_iPlace gradient to The normalized value of amount, F (c) represents the object function of human eye central point c；

3) for object function F (c), it is optimized, is chosen the maximum of object function F (c) after optimization, this F The coordinate of c central point C that () maximum is corresponding as accurate human eye centre coordinate, the formula after optimization is:

$F\left(c\right)={\mathrm{\Σ}}_{i=1}^N\mathrm{\ω}\left(c\right)\·max({\stackrel{\→}{d}}_i\·{\stackrel{\→}{g}}_i,0),$

W (c)=(Max-I (x, y))^r

Wherein, w (c) represents the weighted value of human eye centre coordinate C, and Max represents the maximum of pixel in gray level image.

Below in conjunction with concrete data, the present invention is specifically described:

Step one, input testing image, use Face datection algorithm, extract facial image；

Step 2, according to the facial image obtained in step one, utilize gray-level projection algorithm to extract out human eye Approximate region；

Firstly, it is necessary to the non-face background parts of facial image is left out, find according to statistics, abscissa [0,0.1] ∪ The region of [0.9,1] is frequently not face, is deleted by the background image at this abscissa, and therefore, [0.1,0.9] is face Abscissa；

Further, left and right face respectively accounts for the general area of face, choose the most respectively abscissa [0.1,0.5] and [0.5, 0.9] respectively as left eye and the candidate region of right eye；

Further, as a example by left eye, left-eye candidate region is carried out horizontal direction integral projection, projection function h (y) For:

$h\left(y\right)=\frac{1}{0.5-0.1}{\∫}_{0.1}^{0.5}I(x,y)dx;$

Wherein, (x y) represents the gray value of facial image to I；The right eye abscissa above-mentioned formula of substitution is calculated by right eye ?；

By statistical analysis and priori, find the vertical coordinate interval of eyes generally [0.5,0.8], this vertical coordinate Interval eyebrow and eyes position are due to dark, and all integrated values are relatively low, and the present embodiment uses the method setting dynamic threshold Determining the vertical coordinate of human eye, concrete grammar is as follows:

A) projection function h (y) average in vertical coordinate interval [0.5,0.8] and standard deviation are calculated respectively；

B) average is used to deduct the standard deviation of suitable multiple as dynamic threshold；

C) filter out all vertical coordinates less than dynamic threshold, and sort from small to large according to vertical coordinate, take out wherein four / mono-quantile is as with reference to vertical coordinate；

D) will be centered by this reference coordinate, 0.1 is the interval band scope as human eye vertical coordinate of width；

After obtaining images of left and right eyes center vertical coordinate, it is assumed that right and left eyes abscissa is respectively 0.25 and 0.75, with two eyes it is Center, chooses the right and left eyes coarse positioning region of a length of 0.4, a width of 0.3.

Step 3, according to the human eye coarse positioning area image obtained in step 2, utilize gradient algorithm to calculate accurately Human eye centre coordinate, concretely comprises the following steps:

1) calculate gradient, first calculate facial image gray value I (x, Grad y)Used here as Sobel Operator carries out gradient calculation, and wherein the gradient in x direction and y direction is respectively G_x(x, y) and G_y(x, y), concrete formula is:

$G_x(x,y)=\left[\begin{array}{ccc}-1& 0& 1\\ -2& 0& 2\\ -1& 0& 1\end{array}\right]\×I(x,y);$

$G_y(x,y)=\left[\begin{array}{ccc}1& 2& 1\\ 0& 0& 0\\ -1& -2& -1\end{array}\right]\×I(x,y);$

$\▿\·I(x,y)=\sqrt{G_x{(x,y)}^2+G_y{(x,y)}^2}$

2) calculating object function F (c) of left eye central point C, concrete formula is:

$F\left(c\right)={\mathrm{\Σ}}_{i=1}^N\mathrm{\ω}\left(c\right)\·max({\stackrel{\→}{d}}_i\·{\stackrel{\→}{g}}_i,0);$

W (c)=(255-I (x, y))^r；

${\stackrel{\→}{d}}_i=\frac{x_i-c}{\left|\right|x_i-c|{|}_2}$

${\stackrel{\→}{g}}_i=\frac{{{\stackrel{\→}{g}}_i}^{\′}}{\left|\right|{{\stackrel{\→}{g}}_i}^{\′}|{|}_2}$

${\stackrel{\→}{g}}_i^{\′}=\▿\·I\left(x\right){|}_{(x=x_i)}$

Wherein, r takes 0.5 according to experiment effect, and Fig. 2 is that C point is positioned at central point and the structural representation of non-central point, according to The schematic diagram of Fig. 2 is it can be seen that only when c is in the center of circle, target function value is maximum；

3) choose a little in the maximum point of object function F (c) value, be pinpoint human eye center.

Present invention employs human eye coarse positioning and be accurately positioned the mode combined, it is to avoid calculate all regions of image, fall Low computation complexity.

Present invention utilizes integral projection and two important characteristics of image of gradient, in the case of illumination condition is bad, Can also normally work, there is stronger robustness.

The present invention, when calculating gradient object function, eliminates the gradient vector that part mould length is shorter, have employed segmentation choosing Take the mode of object function maximum, reduce computation complexity.

The present invention can combine with iris identification, by quickly human eye center, location, extracts image around iris, it is to avoid Iris identification calculates face Zone Full at the very start, improves accuracy and the real-time of identification.

The ultimate principle of the present invention, principal character and advantages of the present invention have more than been shown and described.The technology of the industry Personnel, it should be appreciated that the present invention is not restricted to the described embodiments, simply illustrating this described in above-described embodiment and description The principle of invention, the present invention also has various changes and modifications without departing from the spirit and scope of the present invention, and these become Change and improvement both falls within scope of the claimed invention.Claimed scope by appending claims and Equivalent defines.

Claims (5)

1. the human eye method for rapidly positioning for iris identification, it is characterised in that comprise the steps:

Step one, input testing image, use Face datection algorithm, extract facial image；

Step 2, facial image a kind of for step is converted to gray level image, gray level image slightly extracts human eye area figure Picture；

Step 3, according to the human eye area image obtained in step 2, calculate accurate human eye centre coordinate.

A kind of human eye method for rapidly positioning for iris identification the most according to claim 1, it is characterised in that described step The method slightly extracting human eye area image in rapid two uses gray-level projection algorithm.

A kind of human eye method for rapidly positioning for iris identification the most according to claim 2, it is characterised in that described ash Concretely comprising the following steps of degree integral projection algorithm:

A) testing image is carried out Face datection, i-th facial image is designated as?Middle carry out according to priori Human eye positions, and i.e. draws left eye, right eye candidate region, is designated as respectivelyWith

B) calculation procedure A) in drawWithIntegration in the horizontal direction also obtains normalization result；

$h\left(y\right)=\frac{1}{y_{max}-y_{min}}{\∫}_{y_{min}}^{y_{max}}I(x,y)dx$

Wherein, (x y) represents facial image (x, y) gray value put, y to I_max、y_minObtain according to priori, represent eye The scope of eyeball region abscissa, h (y) representsOrNormalization projection function in the horizontal direction；

C) projection function h (y) is calculated respectively at the interval [x of human eye vertical coordinate_min, x_maxAverage in] and standard deviation, use average Deduct the standard deviation of certain multiple as dynamic threshold T (h (y))；Filter out all vertical seats less than dynamic threshold T (h (y)) Mark, i.e. gathered y | h (y) >=T (h (y)) }, y | h (y) represents the vertical coordinate of h (y), and enters from small to large according to vertical coordinate Row sequence, in taking-up vertical coordinate sequence, the vertical coordinate of 1/4th quantiles is as with reference to vertical coordinate x, with this reference vertical coordinate x Centered by, 0.1 width as interval band, the i.e. Range Representation of human eye vertical coordinate be [x-0.1 × height, x+0.1 × Height, height represent the height of facial image, further, take [x, (ymax+ymin)/2] human eye as coarse positioning Centre coordinate；

D) by step C) in centered by the human eye centre coordinate obtained and priori coarse localization human eye area image D_iris, D_irisCoordinate range is: abscissa [y_max, y_min], vertical coordinate scope [x-0.1*height, x+0.1*height, coarse positioning people Eye centre coordinate is [x, (ymax+ymin)/2].

A kind of human eye method for rapidly positioning for iris identification the most according to claim 1, it is characterised in that described step The method of the accurate human eye centre coordinate calculated in rapid three uses gradient algorithm.

A kind of human eye method for rapidly positioning for iris identification the most according to claim 4, it is characterised in that described ladder Concretely comprising the following steps of degree algorithm:

1) for the human eye area image D of coarse positioning in step 2_iris, in step one gray level image gray value I (x, y), Calculate GradConcrete formula is as follows:

$G_x(x,y)=\left[\begin{array}{ccc}-1& 0& 1\\ -2& 0& 2\\ -1& 0& 1\end{array}\right]\×I(x,y);$

$G_y(x,y)=\left[\begin{array}{ccc}1& 2& 1\\ 0& 0& 0\\ -1& -2& -1\end{array}\right]\×I(x,y);$

$\▿\·I(x,y)=\sqrt{G_x{(x,y)}^2+G_y{(x,y)}^2}$

Wherein, G_x(x, y) and G_y(x y) represents x direction and the Grad in y direction respectively；

2) in order to try to achieve human eye centre coordinate c, then the formula of object function is:

$F\left(c\right)={\mathrm{\Σ}}_{i=1}^N({\stackrel{\→}{d}}_i\·{\stackrel{\→}{g}}_i)$

${\stackrel{\→}{d}}_i=\frac{x_i-c}{\left|\right|x_i-c|{|}_2}$

${\stackrel{\→}{g}}_i=\frac{{{\stackrel{\→}{g}}_i}^{\′}}{\left|\right|{{\stackrel{\→}{g}}_i}^{\′}|{|}_2}$

${\stackrel{\→}{g}}_i^{\′}=\▿\·I\left(x\right){|}_{\left(x=x_i\right)}$

Wherein, x₁,x₂,…x_i…,x_NRepresent the pixel in gray level image,Represent x_iThe Grad of point,Table Show an x_iNormalized vector between a c 2,Represent some x_iPlace's gradient vector,Represent some x_iReturning of place's gradient vector One change value, F (c) represents the object function of human eye central point c；

3) for object function F (c), being optimized, choose the maximum of object function F (c) after optimization, this F (c) is The coordinate of the central point C that big value is corresponding as accurate human eye centre coordinate, the formula after optimization is:

$F\left(c\right)={\mathrm{\Σ}}_{i=1}^N\mathrm{\ω}\left(c\right)\·max({\stackrel{\→}{d}}_i\·{\stackrel{\→}{g}}_i,0),$

W (c)=(Max-I (x, y))^r

Wherein, w (c) represents the weighted value of human eye centre coordinate C, and Max represents the maximum of pixel in gray level image.