CN103020953A - Segmenting method of fingerprint image - Google Patents

Abstract

The invention discloses a segmenting method of a fingerprint image. The segmenting method specifically comprises the following steps: a, reading in the fingerprint image; b, quickly cutting the fingerprint image; c, transforming in reverse colour; d, equalizing; e, carrying out top-hat transformation; f, calculating the characteristic quantity in blockst; g, segmenting in blocks; h, processing the image according to morphology; and i, obtaining the segmented fingerprint image. The segmenting method of the finger image is applicable to the fingerprint image poorer in quality, can correctly segment the fingerprint, and is higher in reliability; the ISODATA (Iterative Self-organizing Data Analysis Techniques Algorithm) clustering algorithm is adopted and used for segmenting in blocks, and the clustering speed is higher; the quick cutting way is carried out, thereby the processing time is reduced; the block segmenting mode and pixel segmenting mode are combined, thus the segmented fingerprint is relatively smooth in contour; and the image equalizing mode and the top-hat transformation mode are adopted to enhance the image, and as a result, the segmenting is more efficient.

Description

A Segmentation Method of Fingerprint Image

technical field

The invention relates to a fingerprint image preprocessing method, in particular to a fingerprint image segmentation method. the

Background technique

Fingerprint image recognition is the earliest application and the cheapest branch of human biometric identification technology. It is widely used in criminal investigation, housing security, identity confirmation of banks, securities, insurance and other financial institutions, access control in important areas, staff or Membership management and other fields have broad application prospects. the

Fingerprint image segmentation is an important step in the preprocessing stage of fingerprint recognition. The main purpose is to remove non-fingerprint areas and fingerprint areas that are more noisy and difficult to distinguish. Effective segmentation can reduce the time of subsequent processing, reduce the extraction of false feature points, and improve the accuracy of identification. Rate. the

At present, the commonly used fingerprint segmentation methods are: a. According to the segmentation method of image grayscale characteristics, the fingerprint image is segmented by using the average value and variance of the fingerprint image grayscale. There are global threshold segmentation and adaptive threshold segmentation. The global threshold segmentation depends on the image grayscale. The bimodal nature of the degree distribution is good. If the bimodality is not obvious or the gray level is multimodal, the segmentation effect is not ideal. The adaptive threshold segmentation will segment the easy-to-restore areas with low contrast and strong directionality, and the segmentation There is block effect in the final fingerprint image; b. According to the segmentation method of fingerprint image direction and frequency characteristics, this method is more complicated, especially the calculation of point direction or point frequency. Difficult to calculate accurately; c According to the segmentation method of grayscale characteristics and directional characteristics, using the local standard deviation (or variance) and consistency features of the image, the linear classification method is used for segmentation. This method takes the directional characteristics and grayscale characteristics into account. Fusion of multiple characteristics, but the selection of its coefficients is very critical, and the setting of the threshold is difficult; d segmentation method based on hidden Markov model; e segmentation method based on transformation and energy field; method d and method e consider There are many factors, but the calculation complexity of the algorithm is relatively large, and the processing efficiency of low-quality fingerprint images is low, and it cannot be accurately segmented. the

Contents of the invention

The object of the present invention is to provide a kind of segmentation method of fingerprint image, this method applies ISODATA clustering algorithm (iterative self-organization analysis algorithm) and morphological image processing method in the fingerprint image segmentation, first fingerprint image is cut out, if fingerprint If the image contrast is low, equalize the image, then perform top-hat transformation on the image to compensate for the uneven background brightness, and finally use ISODATA clustering to segment the image into blocks and perform morphological image processing. the

The specific steps are:

(1) Read in the fingerprint image, determine the block size W according to the image resolution, if resolution>600dpi, then W=30, if resolution<400 dpi, then W=8, otherwise W is equal to 16 by default, and convert the fingerprint image It is a double image img.

(2) Quickly crop the double class image img obtained in step (1):

m为图像的行数，i≤m，i用floor函数取整；  a. Starting from line 1, every x lines, calculate the difference diff(i) between the maximum gray value and the minimum gray value of line i; where i=1+kx, k is a non-negative integer,

m is the number of rows of the image, i≤m, and i is rounded by the floor function;

b. Find the maximum value diffmax of the difference diff(i) between the maximum gray value and the minimum gray value;

c. Find the calculation row diffd with a large difference between the maximum gray value and the minimum gray value (take the row whose difference between the maximum gray value and the minimum gray value is greater than diffmax/3);

d. Find the start line mb with a larger difference, starting from the first calculation line found in step (2) step c, if the difference between two consecutive calculation lines is large, determine that the start line is the calculation line The previous calculation line (not less than 1), otherwise look backward;

e. Find the ending line me with a large difference, starting from the last calculation line found in step (2) c, if the difference between two consecutive calculation lines is large, then the end line is this calculation line The next calculation row (not greater than m), otherwise look forward;

f. According to the method of steps a~e, find the start column nb and the end column ne with a large difference;

g. Adjust the end row me and the end column ne, so that the size of the cropped image is an integer multiple of the block size W, and crop the original image img to obtain the cropped image img1, the size of the cropped image is m1*n1, Where m1=me-mb, n1=ne-nb. the

(3) Inverse color transformation:

a. Calculate the average gray value avgbound of the pixels around the image img1, that is, calculate the average gray value of all pixels in the first row, the last row, the first column and the last column;

b. If the background is bright, that is, avgbound>0.5, perform inverse color transformation img1=1-img1, otherwise skip to step (4). the

(4) Equalization;

a. Calculate the average gray value of the image img1 $\mathrm{avg}1=\underset{i=1}{\overset{m1}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{\mathrm{no}1}{\mathrm{\&Sigma;}}}\mathrm{img}1(i,j)---\left(1\right)$

b. Calculate the standard deviation of the image $\mathrm{std}1=\sqrt{\frac{\underset{i=1}{\overset{m1}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{\mathrm{no}1}{\mathrm{\&Sigma;}}}{(\mathrm{imgl}(i,j)-\mathrm{avg}1)}^2}{m1*\mathrm{no}1-1}}---\left(2\right)$

c. Image histogram equalization, if std1<T1 (T1 takes 0.11), perform image histogram equalization, otherwise skip to step (5). the

(5) Top hat transformation;

a. Set the structural element to be a circle with a radius of W/2;

b. Perform top-hat transformation on image img1 to obtain image img2. the

(6) Calculate the feature quantity in blocks;

a. Divide the image img2 into non-overlapping small blocks imgb of W*W size, and calculate the feature quantity;

b. Calculate the block gray average value avgb: $\mathrm{avgb}=\underset{i=1}{\overset{w}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{w}{\mathrm{\&Sigma;}}}\mathrm{imgb}(i,j)---\left(3\right)$

c. Calculate block standard deviation stdb: $\mathrm{stdb}=\sqrt{\frac{\underset{i=1}{\overset{w}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{w}{\mathrm{\&Sigma;}}}{(\mathrm{imgb}(i,j)-\mathrm{avgb})}^2}{w*w-1}}---\left(4\right)$

d. Calculate block gray contrast zb:

In the formula, n1 is the number of points in the block whose gray value is greater than or equal to the average gray value of the block, n2 is the number of points in the block whose gray value is smaller than the average gray value of the block, and t1 is the number of points in the block whose gray value is greater than or equal to the gray value of the block The sum of the gray values of all points with the mean value avgb, t2 is the sum of the gray values of all points in the block whose gray value is less than the block gray mean value avgb;

e. Calculate block direction consistency cohb: $\mathrm{cohb}=\frac{\sqrt{{(G_{\mathrm{xx}}-G_{\mathrm{yy}})}^2+4{G_{\mathrm{xy}}}^2}}{G_{\mathrm{xx}}+G_{\mathrm{yy}}}---\left(6\right)$

${\mathrm{<span\; class="notranslate">\; G</span>}}_{\mathrm{<span\; class="notranslate">\; xx</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; w</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\underset{\mathrm{<span\; class="notranslate">\; v</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; w</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{<span\; class="notranslate">\; \&dtri;</span>}_{\mathrm{<span\; class="notranslate">\; x</span>}}{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; v</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}$

${\mathrm{<span\; class="notranslate">\; G</span>}}_{\mathrm{<span\; class="notranslate">\; yy</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; w</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\underset{\mathrm{<span\; class="notranslate">\; v</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; w</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{<span\; class="notranslate">\; \&dtri;</span>}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}}{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; v</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}$

In the formula, $G_{\mathrm{yy}}=\underset{u=1}{\overset{w}{\mathrm{\&Sigma;}}}\underset{v=1}{\overset{w}{\mathrm{\&Sigma;}}}{\&dtri;}_x(i+u,j+v)\&Center\; Dot;{\&dtri;}_{\mathrm{the\; y}}(i+u,j+v)---\left(7\right)$

${<span\; class="notranslate">\; \&dtri;</span>}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 3</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\underset{\mathrm{<span\; class="notranslate">\; v</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 3</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; v</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&Center\; Dot;</span>\mathrm{<span\; class="notranslate">\; imgb</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; v</span>}<span\; class="notranslate">\; )</span>$

${<span\; class="notranslate">\; \&dtri;</span>}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 3</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\underset{\mathrm{<span\; class="notranslate">\; v</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 3</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; v</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&CenterDot;</span>\mathrm{<span\; class="notranslate">\; imgb</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; v</span>}<span\; class="notranslate">\; )</span>$

S _x , S _y are Sobel operators.

(7) block segmentation;

a. Determine the upper and lower limits of the four feature quantities up_feature and low_feature respectively;

up_feature = min{0.9, maxfeature-difffeature}

low_feature = min{0.1, minfeature+difffeature} (8)

In the formula, difffeature=(maxfeature-minfeature)/10, maxfeature is the maximum value of all block feature quantities, and minfeature is the minimum value of all block feature quantities;

b. Determine the initial background block class and the center of the initial background block class. When a feature quantity is less than its corresponding lower limit, it is the initial background block. The block number of the initial background block is greater than or equal to 1. The initial background class center is a four-dimensional vector composed of the average value of all initial background block feature quantities;

c. Determine the initial foreground block class and the center of the initial foreground block class. When all feature quantities are greater than their corresponding upper limit, it is the initial foreground block. The initial foreground block class center is a four-dimensional vector composed of the average value of all initial foreground block feature quantities. If the initial foreground block is not found, that is, the block number of the initial foreground block is 0, then the initial foreground block class center is a four-dimensional vector composed of four feature quantity upper limits;

d. For each non-initial foreground block and the final block of non-initial background block, calculate the distance of four feature quantities to the four components of the foreground block class center and the background block class center respectively, namely the Euclidean distance;

e. If the distances from the four feature quantities to the center of the foreground block class are greater than twice the distance to the center of the background block class, then the block is a background block; if the distances from the four feature quantities to the center of the background block class are greater than to the foreground Twice the distance from the center of the block class, the block is a foreground block; in other cases, it is a block that cannot be determined temporarily;

f. Calculate the center of the foreground block class and the background block class respectively. The class center of the background block is a four-dimensional vector composed of the average value of all background block features. The foreground block class center is a four-dimensional vector composed of the average value of all foreground block features. If no foreground block is found, the center of the foreground block class is a four-dimensional vector composed of four feature quantity upper limits;

g. Compare the new cluster center with the old cluster center, if the distance is less than the threshold T2 (set to 0.01), stop, otherwise, take the new cluster center as the benchmark, and return to step d.

(8) Morphological image processing:

a. For the image img1, calculate the average gray level avgback of the background block segmented by the block. If the background is darker, that is, avgback<0.5, perform inverse color transformation, that is, y=1-img1, and change the gray value of the background Replaced by the average gray value of the background. ;

b. The number of processing times n is initialized to 1;

c. Corrode the image y with structural elements of 5*5 size, enhance (square), use the OTSU method (maximum inter-class variance method) to find the area mask msk, negate msk, and remove smaller objects;

d. Corrode the region mask msk with structural elements of 7*7 size, fill holes, and increase the number of processing n by 1, if n≤4, return to step c, otherwise execute step e;

e. In order to prevent excessive boundary segmentation, the outer boundary of the segmentation result is enlarged by W/2 to obtain the final segmentation result msk.

(9) Get the segmented fingerprint image:

The function of the step (2) is to remove the non-fingerprint area, reduce the time for subsequent processing, and use intervals of several rows for calculation to speed up the processing speed.

The step (3) is only implemented on the image with a brighter background, and its effect is to convert the fingerprint image into an image with a dark background and a bright foreground, so that step (5) implements top-hat transformation. the

The step (4) is only implemented for images with low contrast, and its effect is to widen the gray scale interval of the image, make the gray scale distribution uniform, increase the contrast, and make the image details clear. the

The function of the step (5) is to compensate the uneven background brightness. the

The step (6) calculates four feature quantities such as block gray mean value, block standard deviation, block gray contrast and block direction consistency in blocks, as the basis for step (7) block segmentation. the

Described step (7) adopts ISODATA clustering algorithm, sets initial foreground block and initial background block according to the value of step (6) feature quantity, carries out clustering with the mean value of initial foreground block and initial background block as initial clustering center, There are three types of clustering results: foreground blocks, background blocks, and temporarily undetermined blocks. During clustering, only undetermined blocks are calculated and judged, and the clustering speed is relatively fast. the

Said step (8) uses the morphological processing method to divide by pixel, corrodes and enhances the image, carries out binarization to the region mask with the method of maximum variance between classes, deletes small objects and holes, and for preventing excessive segmentation of the boundary, the The outer boundary of the segmentation result is expanded by W/2 to obtain the final segmentation result msk, and this boundary does not affect subsequent processing. the

Different from existing methods, the segmentation method of fingerprint image of the present invention has the following advantages:

(1) The present invention is also applicable to fingerprint images with poor quality, can be correctly segmented, and has higher reliability;

(2) The present invention adopts the ISODATA clustering algorithm to carry out block segmentation, and the clustering speed is fast; adopting fast cutting reduces the processing time;

(3) The present invention adopts the combination of block segmentation and pixel segmentation, and the fingerprint outline that is segmented is relatively smooth;

(4) The present invention uses image equalization and top-hat transformation to enhance the image to make the segmentation more effective.

Description of drawings

Fig. 1 is a flowchart of the fingerprint segmentation method of the present invention. the

Fig. 2 is a fingerprint image of an embodiment of the present invention. the

In the figure: (a) is the original image FVC2004db3 103_5.GIF; (b) is the image after (a) fast cropping; (c) is the image after (b) top-hat transformation; (d) is (c) block segmentation As a result, black is the background block; (e) is the image obtained by (b) inverse color transformation and replacing the background with the average gray value of the background according to (d); (f) is the image obtained by (e) morphological processing Mask; (g) is the image after (f) is enlarged to the original image size and (a) and operated. the

Fig. 3 is the feature quantity of the block of the present invention, which is a decimal between 0 and 1, wherein the darkest is 0, and the whitest is 1. the

In the figure: (a) block gray average avgb; (b) block standard deviation stdb; (c) block gray contrast zb; (d) block direction consistency cohb. the

Detailed ways

Example:

Method of the present invention is to run under Windows XP environment, realizes with Matlab language, and its step is:

(1) Use the imread function to read in the fingerprint image FVC2004db3 103_5.GIF (the type of collector is thermal scratch), use the imfinfo function to obtain the image resolution, determine the block size W=16, and convert the image into a double image img , the image size is 480*300, m=480, n=300, as shown in Figure 2(a).

(2) Quickly crop the image img, x=21.9089, respectively calculate the maximum gray value and minimum gray value of the 1st, 22nd, 44th, 66th, 88th, 110th, 132th, 154th, 176th, 198th, 220...461 lines The difference between degree values diff(i), the maximum value diffmax=0.3137, mb=44, me=454, nb=18, ne=300, and adjust the size to an integer multiple of W, me=443, ne=289, get The cropped image img1 has an image size of 400*272, as shown in Figure 2(b). the

(3) Calculate the average value of pixels around image img1 avgbound=0.4009, avgbound<0.5, skip to step (4). the

(4) Calculate the standard deviation std1=0.3201 of the image img1, std1>0.11, skip to step (5). the

(5) Use the function strel to create a morphological structure element of a circle with a radius of W/2=8, and use the function imtophat to perform top-hat transformation on the image img1 to obtain the image img2, as shown in Figure 2(c). the

(6) Calculate block gray average avgb, block standard deviation stdb, block gray contrast zb, and block direction consistency cohb for image img2, as shown in Figure 3. the

(7) Determine the upper and lower limits of the block gray mean value 0.5018, 0.0558, the upper and lower limits of the standard deviation 0.4248, 0.0472, the upper and lower limits of the block gray contrast 0.8380, 0.0931, the upper and lower limits of the block direction consistency 0.8684, 0.0965, the initial background block class center is ( 0.0343, 0.0260, 0.0439, 0.3199), the initial foreground block class center is (0.4483, 0.4464, 0.8646, 0.9166), use the ISODATA clustering algorithm to find some background blocks, and obtain the block segmentation result mskback, as shown in Figure 2(d) The middle black block is part of the background block found. the

(8) For the image img1, according to mskback, calculate the average value of the background gray level avgback=0.3090 from the block segmentation, perform inverse color transformation on the image, and replace the gray value of the background with 1-avgback to obtain the image y, As shown in Figure 2(e); use the function imerode to corrode and square the image y with a structure element of 5*5 size, use the OTSU method to find the area mask msk, negate it, and remove small objects; use a structure of 7*7 size The element corrodes the area mask msk, fills the holes with the function imfill, and then expands W/2=8 to obtain the final mask msk, as shown in Figure 2(f). the

(9) Expand the mask to the original image size of 480*300, and output the segmentation effect, as shown in Figure 2(g), black is the segmented background. the

Claims (1)

1. the dividing method of a fingerprint image is characterized in that concrete steps are:

(1) read in fingerprint image, determine a minute block size W according to image resolution ratio resolution, if resolution 600dpi, if W=30 then is resolution＜400dpi, then W=8, otherwise the W acquiescence equals 16, and fingerprint image is converted to double class image img;

(2) the double class image img of step (1) gained carried out Fast trim:

A. since the 1st row, capable every x, calculate the difference diff (i) of the capable maximum gradation value of i and minimum gradation value, i=1+kx wherein, k is nonnegative integer,

M is the line number of image, i≤m, and i rounds with the floor function;

B. obtain the maximal value diffmax of the difference diff (i) of all maximum gradation value and minimum gradation value;

C. find out the larger calculating row diffd of difference of maximum gradation value and minimum gradation value---get the difference of maximum gradation value and minimum gradation value greater than the row of diffmax/3;

D. find the larger begin column mb of difference, from the first calculating row beginning that step (2) c found out the step, if the difference of continuous two calculating row is all larger, determine that begin column is last calculating row of this calculating row, is not less than 1, otherwise seeks backward;

E. find out the larger end line me of difference, last that find out from step (2) c go on foot calculates row to begin, if the difference that continuous two calculating are gone is all larger, then end line is that this rear calculating of calculating row is gone, and is not more than m, otherwise forward searching;

F. the method by step a ~ e finds the larger begin column nb of difference and end column ne;

G. end line me and end column ne are adjusted, make cutting after image be of a size of minute integral multiple of block size W, original image img is carried out cutting, obtain image img1 after the cutting, the image size is m1*n1 after the cutting, m1=me-mb wherein, n1=ne-nb;

(3) inverse conversion:

A. the average gray avgbound of pixel around the computed image img1 namely calculates the first row, last column, and first row is listed as the average gray of all pixels with last;

If b. background is brighter, i.e. avgbound〉0.5, then carry out inverse conversion img1=1-img1, otherwise jump to step (4);

(4) equalization:

A. the average gray of computed image img1

B. the standard deviation of computed image

C. image histogram equalization, if std1＜T1, T1 gets 0.11, then carries out the image histogram equalization, otherwise jumps to step (5);

(5) top cap conversion:

A., it is that radius is the circle of W/2 that structural element is set;

B. image img1 is carried out top cap conversion, obtain image img2;

(6) piecemeal calculated characteristics amount:

A. the fritter imgb that image img2 is divided into the non-overlapping copies of W*W size, the calculated characteristics amount;

B. computing block gray average avgb:

C. computing block standard deviation stdb:

D. computing block grey-scale contrast zb:

In the formula, n1 is gray-scale value counting more than or equal to piece gray average avgb in the piece, n2 is gray-scale value counting less than piece gray average avgb in the piece, t1 is the had some gray-scale value sum of gray-scale value in the piece more than or equal to piece gray average avgb, and t2 is the had some gray-scale value sum of gray-scale value in the piece less than piece gray average avgb;

E. computing block direction consistance cohb:

In the formula,

S _x, S _yBe the Sobel operator;

(7) piecemeal is cut apart:

A. determine respectively four characteristic quantity bound up_feature and low_feature;

up_feature=min{0.9，maxfeature-difffeature}

low_feature=min{0.1，minfeature+difffeature} （8）

In the formula, difffeature=(maxfeature-minfeature)/10, maxfeature are the maximal value of all block feature amounts, and minfeature is the minimum value of all block feature amounts;

B. determine initial background piece class and initial background piece class center, when certain characteristic quantity is the initial background piece less than its corresponding lower prescribing a time limit, the piece number of initial background piece is more than or equal to 1, and initial background class center is the four-dimensional vector that the mean value of all initial background block feature amounts forms;

C. determine initial foreground blocks class and initial foreground blocks class center, when all characteristic quantities all are initial foreground blocks greater than its corresponding upper prescribing a time limit, the four-dimensional vector that the mean value that initial foreground blocks class center is all initial foreground blocks characteristic quantities forms, if do not find initial foreground blocks, the piece number that is initial foreground blocks is 0, and so initial foreground blocks class center is the four-dimensional vector that four characteristic quantity upper limits form;

D. piece is determined at the end of each non-initial foreground blocks and non-initial background piece, calculated respectively four characteristic quantities to the distance of four components at foreground blocks class center and background piece class center, i.e. Euclidean distance;

If e. all greater than the twice to the distance at background piece class center, then this piece is the background piece to the distance at foreground blocks class center for four characteristic quantities; If all greater than the twice to the distance at foreground blocks class center, then this piece is foreground blocks to the distance at background piece class center for four characteristic quantities; The piece of other situation for temporarily determining;

F. calculate respectively the center of foreground blocks class and background piece class, background piece class center is four-dimensional vectorial for that have powerful connections, block feature amount mean value formed, foreground blocks class center is the four-dimensional vector that the mean value of all foreground blocks characteristic quantities forms, if do not find foreground blocks, then foreground blocks class center is the four-dimensional vector that four characteristic quantity upper limits form;

G. with new cluster centre and old cluster centre relatively, if distance less than threshold value T2 then stop, T2 is made as 0.01, otherwise, take new cluster centre as benchmark, return steps d;

(8) morphological images is processed:

A. to image img1, calculate the average gray avgback at the background piece place that piecemeal is partitioned into, if background is darker, namely the inverse conversion is carried out in avgback＜0.5, namely y=1-img1 replaces background place gray-scale value with background place average gray;

B. number of processes n is initialized as 1;

C. the structural element with the 5*5 size corrodes image y, strengthens---square, use the OTSU method, namely maximum variance between clusters finds regional mask msk, and msk is negated, and removes less object;

D. the structural element with the 7*7 size corrodes regional mask msk, fills hole, and number of processes n adds 1, if step c is returned in n≤4, otherwise execution in step e;

E. excessive for preventing boundary segmentation, the segmentation result outer boundary is enlarged W/2, obtain final segmentation result msk;

(9) fingerprint image after obtaining cutting apart.

Images