CN104809464A - Fingerprint information processing method - Google Patents

Abstract

The invention provides a fingerprint information processing method, which comprises the following steps of uniformly carrying out normalization processing on images and carrying out image segmentation; carrying out image enhancement through image smoothing and median filtering; converting gray level images into binaryzed images by utilizing a dynamic threshold value; carrying out fining processing on the images, and deleting pseudo characteristic points. The invention provides the fingerprint information processing method; a user is enabled to intuitively know image quality; moreover, the identification success rate is improved; the fingerprint information processing method is easy to realize; moreover, the execution efficiency is higher.

Description

A kind of finger print information disposal route

Technical field

The present invention relates to image procossing, particularly a kind of finger print information disposal route.

Background technology

Fingerprint is widely used in person identification means in worldwide.Fingerprint recognition has become a gordian technique of process individual affair and information security.In fingerprint image quality is evaluated, prior art evaluates fingerprint image quality by the contrast of each sub-block of the image that takes the fingerprint and curvature characteristic mass, but this method is just analyzed from partial fingerprint image texture, is not enough to reflect fingerprint image global information; Can not obtain the directional diagram of fingerprint image very well when picture noise is larger, the pseudo-random numbers generation elimination result in addition for the fingerprint image of non-homogeneous collection is not good enough, thus affects last finger print information extraction and identify.

Summary of the invention

For solving the problem existing for above-mentioned prior art, the present invention proposes a kind of finger print information disposal route, comprising:

Step one, carries out normalization to image unification and processes and carry out Iamge Segmentation;

Step 2, carries out image enhaucament by image smoothing and medium filtering;

Step 3, utilizes dynamic threshold to be binary image by greyscale image transitions;

Step 4, performs thinning processing to image, deletes pseudo-random numbers generation.

Preferably, described step one comprises further:

(1) original image gray average E (G) and gray scale mean square deviation V (G) is calculated:

$E\left(G\right)=\frac{1}{M\×N}\underset{i=0}{\overset{M-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{N-1}{\mathrm{\Σ}}}G(i,j)$

$V\left(G\right)=\frac{1}{M\×N}\underset{i=0}{\overset{M-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{N-1}{\mathrm{\Σ}}}{[G(i,j)-E\left(G\right)]}^2$

Obtain the image intensity value after normalization:

$G^{\′}(i,j)=\left\{\begin{array}{cc}{E}_0+\sqrt{\frac{V_0\×{(G(i,j)-E\left(G\right))}^2}{V\left(G\right)}}& G(i,j)>E\left(G\right)\\ E_0-\sqrt{\frac{V_0\×{(G(i,j)-E\left(G\right))}^2}{V\left(G\right)}}& G(i,j)\≤E\left(G\right)\end{array}\right\}$

Wherein, G (i, j) represents the gray-scale value of original fingerprint image at (i, j) place pixel, and M, N are the height and the width of fingerprint image, E ₀, V ₀for gray average and the gray scale mean square deviation of expectation; G'(i, j) represent the gray-scale value of the fingerprint image after regularization at (i, j) place pixel;

(3) in conjunction with fingerprint image gray scale mean square deviation and directional information, by fingerprint image piecemeal, Sobel operator is utilized to calculate each pixel gradient respectively, and obtain block gradient average and block gradient mean square deviation, get the block gradient standard deviation sum of all directions as block eigenvalue, again the segmentation threshold that average finds out block gradient is got to block eigenvalue, be greater than described threshold portion as fingerprint image prospect, be less than threshold value as fingerprint image background.

Preferably, described step 4 comprises further:

Adopt mathematics to table look-up refinement, the crestal line pixel after refinement is divided into isolated point, end points, interior point, bifurcation, chooses and adopts statistics to be averaging, automatic selected threshold to threshold value; According to following known judgement eight neighborhood central point attribute: $Y=1/2\underset{i=0}{\overset{7}{\mathrm{\Σ}}}|P^{\′}\left(i\right)-P^{\′}(i+1)|$

Wherein, P'(i) represent the value of i-th neighborhood territory pixel point in eight neighborhood, if pixel is white point, then P'(i)=1; If stain, P'(i)=0, P'(8)=P'(0), above-mentioned Y characterizes the attribute that crestal line is put, Y=0,1, the central point of 2 corresponding eight neighborhood is respectively isolated point, end points, interior point, if Y >=3, then central point is bifurcation;

Complete pseudo-random numbers generation as follows to delete:

(1) in conjunction with the partial structurtes information of unique point, judge the attribute of fingerprint image black pixel point, after all Edge Feature Points of filtering, mark all accurate unique points;

(2) judge whether accurate unique point belongs to isolated point, end points, bifurcation, directly deletes for isolated point successively; For end points, judge whether the pixel number in this connected domain is greater than threshold value M ₁if be greater than threshold value M ₁then store, otherwise delete; For bifurcation, judge whether 3 adjacent crestal lines are greater than threshold value M successively ₁if be all greater than threshold value M ₁then store, otherwise, if having one to mark crestal line pixel number be not more than threshold value, then delete and be not more than threshold value M ₁gauge point; If have 2 to mark crestal line pixel number be not more than threshold value M ₁, judge that these 2 crestal lines are in tangential direction corresponding to central spot, and contrast with the tangential direction of the 3rd article of crestal line, retain the crestal line that tangential direction is close, another crestal line is filled;

(3) by crestal line directional information and correlativity, the comparatively large and unique point pixel distance of filtering relevance is not more than M ₁unique point, be left for obtain real features point.

The present invention compared to existing technology, has the following advantages:

The present invention proposes a kind of finger print information disposal route, allow user intuitively understand picture quality, and improve recognition success rate, be easy to realize, and execution efficiency is higher.

Accompanying drawing explanation

Fig. 1 is the process flow diagram of the finger print information disposal route according to the embodiment of the present invention.

Embodiment

Detailed description to one or more embodiment of the present invention is hereafter provided together with the accompanying drawing of the diagram principle of the invention.Describe the present invention in conjunction with such embodiment, but the invention is not restricted to any embodiment.Scope of the present invention is only defined by the claims, and the present invention contain many substitute, amendment and equivalent.Set forth many details in the following description to provide thorough understanding of the present invention.These details are provided for exemplary purposes, and also can realize the present invention according to claims without some in these details or all details.

An aspect of of the present present invention provides a kind of finger print information disposal route.Fig. 1 is the finger print information process flow figure according to the embodiment of the present invention.First the present invention adopts the fingerprint identification device with information displaying, gather fingerprint and quality evaluation result is outputted to screen, achieve and improve fingerprint recognition success ratio, certain link allowing user intuitively understand fingerprint collecting goes wrong, and how to revise.Fingerprint identification device comprises controller, and controller connects LED screen, fingerprint sensor, JTAG debugging interface, reset circuit, outside SDRAM data-carrier store, serial ports and USB interface respectively with different interfaces; Power management module provides power management for the chip in each part mentioned above and circuit.Wherein, screen is used for showing the result that fingerprint image initial mass is evaluated, the placement location of prompting user finger; Fingerprint sensor is used for obtaining the information in fingerprint pointed; JTAG debugging interface is used for being connected with host computer debugging; Reset circuit is used for initialization fingerprint identification device; FLASH program storage be used for store fingerprint identification device run program; Outside SDRAM data-carrier store is used for storing the ephemeral data produced in fingerprint identification device operational process; Serial ports and USB interface are used for communicating with the connection of host computer; Controller is the core of fingerprint identification device, controls the operation of fingerprint identification device.

Controller is the chip of embedded fingerprint identification technology, the image acquisition of fingerprint, feature extraction, chip that aspect ratio is right can be realized on sheet, performance history is made to become simple, developer can realize the function of fingerprint recognition easily, the preferred controller of the present invention, adopt 32 risc processor kernels, built-in special DSP instruction set.There is SEA/RSA accelerating engine, tailor-made algorithm software, built-in 128KB high speed static random access memory, embed 1MB high-capacity FLASH, clock, symmetry algorithm engine accelerator, RSA encryption and decryption engine, tandom number generator on 64kB ROM and 4kB OTP ROM, and possess abundant external interface: 3 groups USART interfaces, intelligent card interface, sheet.

Fingerprint sensor is made up of sensor array, and each array is a metal electrode.Be placed on the corresponding point of the finger on sensitive face then as an other pole, its principle of work is the capacitance type sensor changing polar plate spacing, whole sensor reads by reading the instruction of inductor, and the size of pickup area is by the decision of register XSHIFT and YSHIFT value.

In fingerprint collecting input process, due to the reason such as fingerprint quality, riding position of finger, all possibly correctly finger print information cannot be identified.For improving fingerprint recognition efficiency, first quality assessment is carried out to the finger print information gathered.Safety governor carries out the sampling of fingerprint grayscale image dot interlace to fingerprint image, and fingerprint image point directional image calculates, and before fingerprint grayscale image, Background is separated, image processing process such as Fingerprint Image Quality Analysis, and showing evaluation result.If quality assessment is defective, according to display information, Resurvey information in fingerprint.

Sample to fingerprint grayscale image by the mode of dot interlace, dot interlace obtains original fingerprint gray level image, and the basis not changing fingerprint character code is reduced data acquisition amount.The local grain trend at each pixel place in described dot interlace original fingerprint gray level image is represented with point directional image, specific as follows:

Calculated by fingerprint image point directional image, each image block in fingerprint image is divided into foreground blocks or background block.Adopt 7 × 7 templates, reference point is positioned at template center, from horizontal level, determine a direction every π/4, definition I=1, and 2,3,4, corresponding 0, π/4,2 π/4,3 π/4, π four direction.Calculate the grey scale change D of all directions _i, compare D _i, find minimum value, just represent the direction of this point:

D _I＝Σ|f’ _I(i,j)-f _I(i _k,j _k)|

In formula, f ' _i(i, j) is the gray average along I direction is put, f _i(i _k, j _k) be the gray-scale value that I direction is put.The foreground blocks of image is the image block being distributed with fingerprint ridge line, and remainder is background block.Foreground blocks is set to 1, and background block is set to 0, before realizing fingerprint grayscale image, Background be separated.Specific as follows:

1) judge whether to meet: 6f _i(i, j)+S _min+ S _max>0.75 Σ S _i

Wherein, f (i, j) is the gray-scale value that (i, j) puts; S _i=Σ f _i(i _k, j _k) adding up for gray-scale value on I direction; S _maxfor the upper limit of accumulated value; S _minfor the lower limit of accumulated value.If meet the condition of formula, then current point is foreground point; Otherwise be background dot.

2) according to the ratio of background dot in fritter, judge that each image block is foreground blocks or background block.If the quantity of background dot exceedes threshold value T in fritter _b, then think that this image block belongs to background block, otherwise be foreground blocks.

By picture quality Rule of judgment, compare quality assessment parameter Q and threshold value T _q.If Q≤T _q, illustrate that picture quality does not reach requirement, judge whether finger position is placed correctly.

Specific as follows:

1) whether belong to a certain specific direction according to each point in this block, judge whether a foreground image block has directivity characteristics;

2) calculate the direction of each image block, obtain the direction histogram of each piece.If the number of pixels with some direction D exceedes preset value T ₁, then the characteristic direction of this block is marked as D;

3) ratio that the quality of fingerprint image can account for all fingerprint foreground pictures by calculating continuous print characteristic direction region is described.Take method of weighting, the image block that distance reference point is far away, the information that it comprises is more reliable, and its weights are also higher;

4) for the arbitrary image block x in fingerprint foreground picture _i, its picture quality Q can be determined with the weights with continuous characteristic direction block with the ratio of the weights of all fingerprint foreground blocks;

5) by lower limit T that fingerprint image quality parameter Q and fingerprint image quality evaluate _qcompare, if Q≤T _q, then Image semantic classification is carried out; Otherwise departing from of finger position is analyzed, and shows corresponding suggestion content.

Whether the no matter fingerprint of which kind of type, exist a complete crestal line determine whether finger departs from by analyzing fingerprint image central area.The present invention uses the tracking based on directional diagram to judge, and whether the placement location pointed is correct.Specific as follows:

1) with the barycenter of foreground picture for initial point, build coordinate system;

2) at the left half axle of x-axis, a characteristic direction is selected not to be that the image block of 0 is as start reference image block;

3) whether belong to according to each point in block the characteristic direction that a certain certain party always judges image block.If block feature direction is 0, then reselect start reference image block; If block feature direction is not 0, then according to the direction of current image block, search for next image block to the right;

4) judge that the direction of this image block and previous image block changes.If the direction of image block and previous image block changes more than 90 °, show to undergo mutation in the direction of current image block, according to the continuity of crestal line, the direction of current image block is substituted the direction of previous image block, search for next image block on this basis;

5) if the direction of image block is no more than 90 °, then judge whether to search a complete crestal line.

If find complete crestal line, the fingerprint image of collection is correct, correct in onscreen cue input, terminates fingerprint image search; Otherwise, illustrate that current image block does not have enough close to the positive axis of x-axis, continue the next image block of search;

6) judge whether the negative semiaxis of x-axis completes search.If do not search for complete, then continue search; If the negative semiaxis of x-axis completes search, the positive axis of search x-axis;

7) positive axis of x-axis is searched in the opposite direction similarly.

If search a complete crestal line, then correct in onscreen cue input, terminate fingerprint image search; Otherwise continue search.If a complete crestal line all cannot be determined from the positive and negative semiaxis of x-axis, then show that this fingerprint image departs from, according to the position indicating user Resurvey fingerprint of barycenter.

According to the position of selected barycenter and the result of judgement, show on screen respectively.

Before carrying out feature extraction and matching, pre-service must be carried out to fingerprint image, recover dactylotype, so that the fingerprint characteristic that reliable extraction is correct.The complexity that direct effect characteristics extracts and mates by result, is related to the discrimination of whole algorithm.The present invention is directed to the problem that conventional fingerprint recognizer complexity is higher, computing is consuming time, consider system hardware platform features, existing algorithm is optimized.The fingerprint recognition optimized comprises fingerprint image preprocessing and feature point extraction and mates.Wherein pre-service comprises: the deletion of Iamge Segmentation, image enhaucament, binaryzation and aftertreatment, refinement and pseudo-random numbers generation.Because picture quality is different, can not directly split.

First adopt 3 × 3 Gaussian template filtering, remove the partial noise in image, make the texture of fingerprint image more level and smooth.Different for picture quality, to the process of fingerprint image unified specification, make all fingerprint images all have unified average and mean square deviation, reduce the gray difference between the crestal line of fingerprint and valley line simultaneously.Normalization and Iamge Segmentation flow process as follows:

(1) original image gray average and gray scale mean square deviation is obtained.

$E\left(G\right)=\frac{1}{M\×N}\underset{i=0}{\overset{M-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{N-1}{\mathrm{\Σ}}}G(i,j)$

$V\left(G\right)=\frac{1}{M\×N}\underset{i=0}{\overset{M-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{N-1}{\mathrm{\Σ}}}{[G(i,j)-E\left(G\right)]}^2$

Obtain the image intensity value after normalization.

$G^{\′}(i,j)=\left\{\begin{array}{cc}{E}_0+\sqrt{\frac{V_0\×{(G(i,j)-E\left(G\right))}^2}{V\left(G\right)}}& G(i,j)>E\left(G\right)\\ E_0-\sqrt{\frac{V_0\×{(G(i,j)-E\left(G\right))}^2}{V\left(G\right)}}& G(i,j)\≤E\left(G\right)\end{array}\right\}$

Wherein, G (i, j) represents that original fingerprint image is at (i, j) gray-scale value of place's pixel, M, N are the height and the width of fingerprint image, the gray average that E (G) is former fingerprint image, the gray scale mean square deviation that V (G) is former fingerprint image, E ₀, V ₀for gray average and the gray scale mean square deviation of expectation, generally experimentally choose moderate expectation value; G'(i, j) represent the gray-scale value of the fingerprint image after regularization at (i, j) place pixel.The present invention preferably chooses E ₀=100, V ₀=100.

(3) Sobel operator asks gradient and block gradient segmentation.

In conjunction with fingerprint image gray scale mean square deviation and directional information, by fingerprint image piecemeal, utilize Sobel operator to calculate each pixel gradient respectively, and obtain block gradient average and block gradient mean square deviation.Get the block gradient standard deviation sum of all directions as block eigenvalue, then the segmentation threshold that average finds out block gradient is got to block eigenvalue, be greater than threshold portion as fingerprint image prospect, be less than threshold value as fingerprint image background.

Image enhancement processes comprises: image smoothing, directional diagram and discretize and filtering.

Irregular noise extracts the accuracy of Fingerprint diretion and has a great impact, and in order to the field of direction that takes the fingerprint more accurately, first to picture smooth treatment, removes irregular noise.G (i, j) is the pixel value at (i, j) place.Adopt 3 × 3 templates to pixel (i, j) eight neighborhood obtains mean value E'(i, j), if | G'(i, j)-E'(i, j) | be greater than predetermined threshold value, then think that this place's pixel value is noise pixel, with E'(i, j) replace G'(i, j) as the pixel value at (i, j) place.Sobel operator is adopted to level and smooth fingerprint image, obtains the transverse gradients at pixel (i, j) place and longitudinal gradient, and fingerprint ridge direction span is defined as between [0 °, 180 °].Get a direction every 22.5 °, amount to 8 general orientation.Calculate the general direction of pixel crestal line, Choose filtering window be 5 × 5 this directional diagram of low-pass filter filter correction obtain smoothly discrete directional diagram.

Filtering Template according to structure construction eight directions of linear space carries out medium filtering.To remove in image noise region little compared with filtering size through medium filtering, filter function be expressed as:

$g_{\mathrm{uv}}(x,y)=\frac{k^2}{{\mathrm{\σ}}^2}\exp [-\frac{k^2(x^2+y^2)}{2{\mathrm{\σ}}^2}]\·[\exp (\mathrm{ik}\·\left(\begin{array}{c}x\\ y\end{array}\right))-\exp (-\frac{{\mathrm{\σ}}^2}{2})]$

$\begin{array}{cc}k=\left(\begin{array}{c}{k}_v\cos \mathrm{\α}\\ k_v\sin \mathrm{\α}\end{array}\right)& k_v=2^{-\frac{v+2}{2}}\mathrm{\π}\end{array}$

α＝uπ/k

The value of v determines the wavelength of filtering, and the value of u represents the direction of kernel function, and K represents total direction number.Parameter σ/k determines the size of Gauss's window, gets here 4 frequencies (v=0,1,2,3) are got in program, 8 directions (i.e. K=8, u=0,1 ..., 7), totally 32 kernel functions.

In order to remove edge fog effect as much as possible, it is Hx=0.1 (1,0,2,0,4,0 that the present invention sets tangential direction Filtering Template parameter, 2,0,1), normal direction Filtering Template is Hy=1/3 (-1,0 ,-2,0,9,0 ,-2,0 ,-1), rectangular filter process is adopted.Rectangular filter is estimated with tangential direction and normal direction 2 one-dimensional filtering devices.In conjunction with medium filtering, first normal direction sharpening is carried out to fingerprint image, then it is level and smooth to carry out tangential direction.

In image binaryzation process, comprise the following steps:

(1) fingerprint image after segmentation is divided into the wicket that n pixel is identical, window pixel number is W.The all pixel summations of each window are averaged as p _i, getting empirical value is ω, and dynamic window threshold value T is met:

T＝p _i+ω

p _i＝G' ₁+G' ₂+…+G' _W/W

(2) some assignment pixel values all in window being greater than T is 1, and the some assignment being less than T is 0.

(3) white point in searching image, searches for and adds up this white point four neighborhood stain number N ₁points N black in eight neighborhood ₂if, N ₂>=7 or N ₁>=3, N ₂>=5, then this eight connected region white point is filled to stain, otherwise the number TR of white point in statistics white point eight connected region R, it is W that threshold value is analyzed in setting ₁if, TR≤W ₁, to connected domain R, add up and judge the pixel distance S of 2 stains adjacent with a certain row (column) white point respectively _x(S _y), if all S _x(S _y)≤W2, is filled to stain by this eight connected region white point.Otherwise, the white point that mark connected domain is corresponding.

(4) until when there is not unlabelled white point in repetitive operation, and image negate look, returns (3); After processing completely, negate obtains final image again.

Thinning process adopts mathematics to table look-up refinement.Crestal line pixel after refinement can be divided into isolated point, end points, interior point, bifurcation according to eight neighborhood feature.In order to delete pseudo-random numbers generation better, threshold value being chosen and adopts statistics to be averaging, automatic selected threshold.

As follows for judging the formula of eight neighborhood central point attribute:

Wherein, P'(8)=P'(0), P'(i) represent the value of i-th neighborhood territory pixel point in eight neighborhood, if pixel is white point, then P'(i)=1; If stain, P'(i)=0.Y characterizes the attribute that crestal line is put, Y=0,1, the central point of 2 corresponding eight neighborhood is respectively isolated point, end points, interior point, Y >=3, central point is bifurcation.The present invention, in conjunction with the directional information of fingerprint and threshold information, provides pseudo-random numbers generation delet method.Concrete steps are as follows:

(1) in conjunction with the partial structurtes information of unique point, judge the attribute of fingerprint image black pixel point, after all Edge Feature Points of filtering, mark all accurate unique points.

(2) judge that accurate unique point belongs to isolated point, end points, bifurcation successively.Isolated point is directly deleted; For end points, judge whether the pixel number in this connected domain is greater than threshold value M ₁if be greater than threshold value, store, otherwise delete; Bifurcation is judged whether 3 adjacent crestal lines are greater than threshold value M successively ₁if be all greater than threshold value M ₁then store, otherwise, if having one to mark crestal line pixel number be not more than threshold value, then delete and be not more than threshold value M ₁gauge point; If have 2 to mark crestal line pixel number be not more than threshold value M ₁, judge that these 2 crestal lines are in tangential direction corresponding to central spot, and contrast with the tangential direction of the 3rd article of crestal line, retain the crestal line that tangential direction is close, another is filled.

(3) by crestal line directional information and correlativity, the comparatively large and unique point pixel distance of filtering relevance is not more than M ₁unique point, be left for obtain real features point.

In sum, the present invention proposes a kind of finger print information disposal route, allow user intuitively understand picture quality, and improve recognition success rate, be easy to realize, and execution efficiency is higher.

Obviously, it should be appreciated by those skilled in the art, above-mentioned of the present invention each module or each step can realize with general computing system, they can concentrate on single computing system, or be distributed on network that multiple computing system forms, alternatively, they can realize with the executable program code of computing system, thus, they can be stored and be performed by computing system within the storage system.Like this, the present invention is not restricted to any specific hardware and software combination.

Should be understood that, above-mentioned embodiment of the present invention only for exemplary illustration or explain principle of the present invention, and is not construed as limiting the invention.Therefore, any amendment made when without departing from the spirit and scope of the present invention, equivalent replacement, improvement etc., all should be included within protection scope of the present invention.In addition, claims of the present invention be intended to contain fall into claims scope and border or this scope and border equivalents in whole change and modification.

Claims (3)

1. a finger print information disposal route, before feature extraction and matching is carried out to fingerprint, pre-service is carried out to fingerprint image, it is characterized in that, comprising:

Step one, carries out normalization to image unification and processes and carry out Iamge Segmentation;

Step 2, carries out image enhaucament by image smoothing and medium filtering;

Step 3, utilizes dynamic threshold to be binary image by greyscale image transitions;

Step 4, performs thinning processing to image, deletes pseudo-random numbers generation.

2. method according to claim 1, is characterized in that, described step one comprises further:

(1) original image gray average E (G) and gray scale mean square deviation V (G) is calculated:

$E\left(G\right)=\frac{1}{M\×N}\underset{i=0}{\overset{M-1}{\mathrm{\Σ}}}\underset{J=0}{\overset{N-1}{\mathrm{\Σ}}}G(i,j)$

$V\left(G\right)=\frac{1}{M\×N}\underset{i=0}{\overset{M-1}{\mathrm{\Σ}}}\underset{J=0}{\overset{N-1}{\mathrm{\Σ}}}{[G(i,j)-E\left(G\right)]}^2$

Obtain the image intensity value after normalization:

$G^{,(i,j)=}\left\{\begin{array}{cc}{E}_0+\sqrt{\frac{V_0\×{(G(i,j)-E\left(G\right))}^2}{V\left(G\right)}}& G(i,j)>E\left(G\right)\\ E_0-\sqrt{\frac{V_0\×{(G(i,j)-E\left(G\right))}^2}{V\left(G\right)}}& G(i,j)\≤E\left(G\right)\end{array}\right\}$

Wherein, G (i, j) represents the gray-scale value of original fingerprint image at (i, j) place pixel, and M, N are the height and the width of fingerprint image, E ₀, V ₀for gray average and the gray scale mean square deviation of expectation; G'(i, j) represent the gray-scale value of the fingerprint image after regularization at (i, j) place pixel;

(3) in conjunction with fingerprint image gray scale mean square deviation and directional information, by fingerprint image piecemeal, Sobel operator is utilized to calculate each pixel gradient respectively, and obtain block gradient average and block gradient mean square deviation, get the block gradient standard deviation sum of all directions as block eigenvalue, again the segmentation threshold that average finds out block gradient is got to block eigenvalue, be greater than described threshold portion as fingerprint image prospect, be less than threshold value as fingerprint image background.

3. method according to claim 2, is characterized in that, described step 4 comprises further:

Adopt mathematics to table look-up refinement, the crestal line pixel after refinement is divided into isolated point, end points, interior point, bifurcation, chooses and adopts statistics to be averaging, automatic selected threshold to threshold value; According to following known judgement eight neighborhood central point attribute: $Y=1/2\underset{i=0}{\overset{7}{\mathrm{\Σ}}}|P^{\′}\left(i\right)-P^{\′}(i+1)|$

Wherein, P'(i) represent the value of i-th neighborhood territory pixel point in eight neighborhood, if pixel is white point, then P'(i)=1; If stain, P'(i)=0, P'(8)=P'(0), above-mentioned Y characterizes the attribute that crestal line is put, Y=0,1, the central point of 2 corresponding eight neighborhood is respectively isolated point, end points, interior point, if Y >=3, then central point is bifurcation;

Complete pseudo-random numbers generation as follows to delete:

(1) in conjunction with the partial structurtes information of unique point, judge the attribute of fingerprint image black pixel point, after all Edge Feature Points of filtering, mark all accurate unique points;

(2) judge whether accurate unique point belongs to isolated point, end points, bifurcation, directly deletes for isolated point successively; For end points, judge whether the pixel number in this connected domain is greater than threshold value M ₁if be greater than threshold value M ₁then store, otherwise delete; For bifurcation, judge whether 3 adjacent crestal lines are greater than threshold value M successively ₁if be all greater than threshold value M ₁then store, otherwise, if having one to mark crestal line pixel number be not more than threshold value, then delete and be not more than threshold value M ₁gauge point; If have 2 to mark crestal line pixel number be not more than threshold value M ₁, judge that these 2 crestal lines are in tangential direction corresponding to central spot, and contrast with the tangential direction of the 3rd article of crestal line, retain the crestal line that tangential direction is close, another crestal line is filled;

(3) by crestal line directional information and correlativity, the comparatively large and unique point pixel distance of filtering relevance is not more than M ₁unique point, be left for obtain real features point.