CN104680130A - Chinese character recognition method for identification cards - Google Patents

Abstract

The invention provides a Chinese character recognition method for identification cards. The Chinese character recognition method comprises the following steps: acquiring a classifier training file; transforming a color image to be recognized into a gray scale image; accurately positioning a Chinese character region of each identification card; carrying out character recognition on each Chinese character region; judging whether the current Chinese character region is the last Chinese character region. According to the Chinese character recognition method provided by the invention, positioning and segmentation of Chinese characters of the identification cards can be carried out by adopting an image processing technology; character recognition is carried out on a Chinese character image on the basis of a tessract library and font structure quality enhancement; the recognition accuracy is higher, no specific equipment is needed, the cost is low and information of multiple identification cards can be simultaneously read.

Description

A kind of I.D. Chinese characters recognition method

Technical field

The present invention relates to technical field of image processing, specifically a kind of I.D. Chinese characters recognition method.

Background technology

I.D. is the unique perfect instrument proving citizen's legal identity, that citizen participates in various politics, economy, the necessary certificate of social activities, such as: enter movable meeting-place, handle bank card, move in hotel, airplane train, Internet bar online, handle all kinds of qualification certificates etc.Can say, I.D. has goed deep into the various aspects of our life.Therefore, how fast and accurately reading identity card information becomes more and more important.

At present, the technical method of conventional reading identity card information has following several:

(1) identity-based card card reader reads information, and these class methods have the feature realization of type built-in contactless IC-card intelligent chip based on China's second-generation resident identification card.The advantage of these class methods is: the precision reading information is very high; For stain and the wearing and tearing of I.D., there is very strong resistivity; Do not rely on visible ray, can use under the rugged surroundings such as dust; But also there is obvious shortcoming in it: need special card-reading apparatus, equipment cost is higher.

(2) read information based on optical character recognition, these class methods are by I.D. coherent element position, image processing techniques location, and using forestland recognition technology carries out character training and identification.The advantage of these class methods is that accuracy is higher, and do not need particular device, hardware cost is lower, for the strong adaptability of various environment, can carry out many ID (identity number) card information and read simultaneously; Its shortcoming is the numerical portion mainly for I.D., and for Chinese character part, accuracy is lower, therefore which also limits these class methods and applies widely.

Chinese characters in common use due to China have more than 3,000, therefore, conventional machine learning method cannot be directly applied for I.D. Chinese Character Recognition, at present, most popular Chinese Character Recognition storehouse of increasing income is the tessract storehouse of google, but this storehouse requires very high for the picture quality of Chinese character and character form structure, if be directly used for identifying I.D. Chinese character, accuracy rate only has 60%, cannot meet actual application demand.

Summary of the invention

The object of the invention is to for the low shortcoming of optical character recognition reading identity card Chinese character information accuracy rate, a kind of I.D. Chinese characters recognition method is provided, make full use of the characteristics of image of I.D., Chinese character is extracted and strengthens, utilize tessract storehouse to carry out Chinese Character Recognition on this basis, accuracy is higher.

Technical scheme of the present invention is:

A kind of I.D. Chinese characters recognition method, comprises the following steps:

(1) sorter training file is obtained;

(2) coloured image to be identified is transformed into gray level image;

(3) the Chinese character region of I.D. is accurately located;

(4) character recognition is carried out to each Chinese character region;

(5) judge whether current Chinese character region is last Chinese character region, if so, then exports recognition result, if not, then return step (4).

Described I.D. Chinese characters recognition method, in step (1), described acquisition sorter training file, specifically comprises:

(11) existing Face datection sorter file is loaded;

(12) tessract Chinese Character Recognition storehouse is loaded;

(13) based on harr characteristic sum adaboost algorithm, the sorter file about I.D. national emblem is trained;

(14) template of the numeral of training I.D. Chinese character region to contain.

Described I.D. Chinese characters recognition method, in step (3), the described Chinese character region to I.D. is accurately located, and specifically comprises:

(31) based on adaboost sorter, the face location in I.D. front and the national emblem position at the I.D. back side is detected;

(32) according to eye position, slant correction is carried out to I.D. direct picture, according to five-pointed star position in national emblem, slant correction is carried out to I.D. back side image;

(33) based on face location and national emblem position, select the Chinese character of I.D. front and back to detect effective coverage respectively, carry out inverse process simultaneously;

(34) Fuzzy Processing is carried out to image, remove background patterns interference;

(35) the two-value vertical edge characteristic pattern of the gray level image after Fuzzy Processing is obtained;

(36) morphology operations is carried out to two-value vertical edge characteristic pattern, obtain connected region;

(37) according to area and position feature, corresponding connected region alternatively region is selected;

(38) based on character pitch feature in candidate region, each Chinese character region of I.D. front and back is accurately located;

(39) judge whether current face position or national emblem position are last face location or national emblem position, if so, then export corresponding Chinese character zone location result, if not, then continue to perform step (32) to step (39).

Described I.D. Chinese characters recognition method, in step (4), describedly carries out character recognition to each Chinese character region, specifically comprises:

(41) Accurate Segmentation Chinese character;

(42) judge that single character is Chinese character or numeral by the width of character, if Chinese character, then perform step (43), if digital, then perform step (45);

(43) Chinese character pattern structure is strengthened;

(44) Chinese Character Recognition is carried out based on tessract storehouse;

(45) carry out numeral based on nearest neighbor algorithm to identify;

(46) judge that whether current character is last character in current Chinese character region, if so, then export Chinese Character Recognition result, if not, then enter character late, continue to perform step (42) to step (46).

Described I.D. Chinese characters recognition method, in step (32), describedly carries out slant correction according to eye position to I.D. direct picture, specifically comprises:

(a1) based on five, three front yard layout rule, coarse positioning is carried out to left and right two;

(a2) central point of eyes is accurately oriented;

(a3) according to the calculating angle of inclination, center of two;

(a4) Sloped rotating correction is carried out to I.D. direct picture;

Describedly according to five-pointed star position in national emblem, slant correction is carried out to I.D. back side image, specifically comprises:

(b1) central point of five stars in national emblem is accurately oriented;

(b2) based on maximum star, the relative position of four starlets is oriented respectively;

(b3) according to the calculating angle of inclination, center of two stars in outside;

(b4) Sloped rotating correction is carried out to I.D. back side image.

Described I.D. Chinese characters recognition method, in step (36), describedly carries out morphology operations to two-value vertical edge characteristic pattern, obtains connected region, specifically comprises:

A () Height value data to vertical connection edges all in two-value vertical edge characteristic pattern is added up;

B the Height value data counted sorts by () from big to small, and obtain the mean value of the altitude information come within 1/3rd positions above, as the average height at vertical connection edge in two-value vertical edge characteristic pattern

C () utilizes structural element template, carry out two-value vertical edge characteristic pattern secondary morphological dilations computing, wherein, expression is not more than maximum integer;

D () utilizes structural element template, to process the two-value vertical edge characteristic pattern of secondary morphological dilations computing carries out 2 closing operation of mathematical morphology;

E () utilizes structural element template, carry out the two-value vertical edge characteristic pattern through 2 closing operation of mathematical morphology secondary morphological erosion computing.

Described I.D. Chinese characters recognition method, in step (41), described Accurate Segmentation Chinese character, specifically comprises:

A () adopts following formula, carry out local binarization to gray scale Chinese character region:

$g(x,y)=\left\{\begin{array}{cc}255& f(x,y)\&GreaterEqual;T\\ 0& f(x,y)<T\end{array}\right.$

$T=\frac{1}{M*N}\underset{m=1}{\overset{M}{\mathrm{\&Sigma;}}}\underset{n=1}{\overset{N}{\mathrm{\&Sigma;}}}f(x_m,y_n)+\mathrm{offset}$

Wherein, the gray-scale value at pixel (x, the y) place that g (x, y) is corresponding after representing binaryzation, f (x, y) represents the gray-scale value at binaryzation preceding pixel (x, y) place, f (x _m, y _n) represent pixel (x in the front M*N neighborhood centered by pixel (x, y) of binaryzation _m, y _n) gray-scale value at place, M, N represent width and the height of neighborhood respectively, and offset represents side-play amount;

B (), according to size, filters interference connected region;

C () carries out vertical projection, obtain the foreground target height of each row;

D (), by the position relationship of crest and trough, carries out Character segmentation;

E () carries out secondary splitting to adhesion character;

F () carries out merging treatment to fracture character.

Described I.D. Chinese characters recognition method, in step (43), described enhancing Chinese character pattern structure, specifically comprises:

A () adds up the connected region number of single character, judge whether to be greater than 1, if so, then perform step (b), if not, then directly performs step (44);

B () connects various piece connected region, specifically comprise:

(b1) select two connected regions arbitrarily, calculate minor increment therebetween;

(b2) in minimum distance, draw straight line, connect two connected regions, form new total connected region;

(b3) connect remaining connected region and total connected region successively, form final connected region.

Described I.D. Chinese characters recognition method, in step (45), described based on nearest neighbor algorithm carry out numeral identify, specifically comprise:

A () adopts following formula, calculate the characteristic distance between character to be identified and all template characters:

dis _i＝ΣΣs(x，y)

$s(x,y)=\left\{\begin{array}{cc}1& f(x,y)\&NotEqual;m(x,y)\\ 0& f(x,y)=m(x,y)\end{array}\right.$

Wherein, dis _irepresent the characteristic distance between character to be identified and i-th template character, f (x, y) represents the pixel (x of character to be identified, y) gray-scale value at place, m (x, y) represents the gray-scale value at pixel (x, the y) place of template character;

B template character that () selects minimal characteristic distance corresponding is as recognition result.

The present invention adopts image processing techniques, carries out the location of I.D. Chinese character, segmentation, and the Chinese character image strengthened based on tessract storehouse and character form structure quality carries out character recognition, reads technology and compares, have following characteristics with existing I.D.:

(1) Chinese Character Recognition accuracy is higher;

(2) do not need specific equipment, cost is low;

(3) many ID (identity number) card information can be carried out to read simultaneously.

Accompanying drawing explanation

Fig. 1 is the logical flow chart of the method for the invention;

Fig. 2 is I.D. front Chinese character zone location process flow diagram;

Fig. 3 is I.D. Chinese character zone location process flow diagram;

Fig. 4 is I.D. gray-scale map, and wherein (a) figure represents I.D. front correlogram, and (b) figure represents I.D. back side correlogram; If no special instructions, following Fig. 5 ~ Figure 12 all represents equivalent;

Fig. 5 is slant correction design sketch;

Fig. 6 selects effective surveyed area and design sketch after inverse process;

Fig. 7 is two-value vertical edge characteristic pattern;

Fig. 8 obtains candidate's connected region design sketch;

Fig. 9 is I.D. Chinese character region fine positioning design sketch;

Figure 10 is I.D. Chinese character segmentation effect figure;

Figure 11 is that Chinese character pattern strengthens design sketch, wherein (a) figure is former figure, b () figure strengthens design sketch, although (a) figure more meets reading custom, but for tessract storehouse, easily be identified as two characters, therefore adopt (b) figure strengthened, be conducive to the overall accuracy rate improving Chinese character;

Figure 12 is I.D. Chinese Character Recognition result figure.

Embodiment

Below, the present invention is further illustrated with specific embodiment by reference to the accompanying drawings.

As shown in Figure 1, a kind of I.D. Chinese characters recognition method, comprises the following steps:

Step 101, acquisition sorter training file, concrete steps are as follows:

(1) existing Face datection sorter file is loaded;

(2) tessract Chinese Character Recognition storehouse is loaded;

(3) based on harr characteristic sum adaboost algorithm, the sorter file about I.D. national emblem is trained;

(4) template of the numeral of training I.D. Chinese character region to contain.

Step 102, according to formula [1], coloured image is transformed into gray level image, effect as shown in Figure 4:

Formula [1]:

f＝0.299R+0.587G+0.114B

Wherein, f is grayscale image values, and R, G, B are the three-channel value of red, green, blue of respective pixel respectively.

The Chinese character regional location of step 103, accurately location I.D., the Chinese character region of I.D. comprises issuing authority's part at the name in front, sex, nationality, address portion and the back side, location algorithm is also corresponding is divided into front Chinese character zone location and back side Chinese character zone location two parts, specific as follows:

Situation 1, I.D. front Chinese character zone location algorithm, as shown in Figure 2, concrete steps are as follows:

Step 201, based on adaboost algorithm, locating human face position.

Step 202, carry out image inclination correction according to eye position, effect is as shown in Fig. 5 (a), and concrete steps are as follows:

(1) based on " three five, front yards " layout rule of face, the position that coarse positioning is left and right two;

(2) central point of eyes is accurately oriented;

(3) according to the calculating angle of inclination, center of two;

(4) Sloped rotating correction is carried out to image.

Step 203, based on face location, select left field as I.D. Chinese character detect effective coverage, carry out inverse process, effect is as shown in Fig. 6 (a) simultaneously.

Step 204, according to convolution mask formula [2], carry out image blurring process, remove background patterns interference:

Formula [2]:

$K=\frac{1}{w*h}\left[\begin{array}{cccccc}1& 1& 1& ...& 1& 1\\ 1& 1& 1& ...& 1& 1\\ 1& 1& 1& ...& 1& 1\\ .& .& .& .& .& .\\ .& .& .& .& .& .\\ .& .& .& .& .& .\\ 1& 1& 1& ...& 1& 1\\ 1& 1& 1& ...& 1& 1\end{array}\right]$

Wherein, w is the width of convolution kernel, and h is the height of convolution kernel.

Step 205, use vertical edge detective operators, obtain the two-value vertical edge characteristic pattern of gray level image, effect is as shown in Fig. 7 (a), and concrete steps are as follows:

(1) based on sobel edge detection operator, as shown in formula [3], vertical edge characteristic pattern is obtained by convolution algorithm;

Formula [3]:

$K=\left[\begin{array}{ccc}-1& 0& 1\\ -2& 0& 2\\ -1& 0& 1\end{array}\right]$

(2) based on niblack local binarization algorithm, two-value vertical edge characteristic pattern is obtained;

(3) isolated marginal point is removed.

Step 206, morphology operations, obtain image connectivity region, concrete steps are as follows:

(1) the height h at each vertical connection edge in two-value vertical edge characteristic pattern is added up _i, then sort from big to small, the altitude information come within 1/3rd positions be above averaging, as the average height at vertical connection edge in two-value vertical edge characteristic pattern

(2) structure based element template, as shown in formula [4], carries out secondary morphological dilations computing, expression is not more than maximum integer:

Formula [4]:

$K=\left[\begin{array}{ccccccc}0& 0& 0& 0& 0& 0& 0\\ 1& 1& 1& 1& 1& 1& 1\\ 0& 0& 0& 0& 0& 0& 0\end{array}\right]$

(3) structure based element template, as shown in formula [5], carries out 2 closing operation of mathematical morphology:

Formula [5]:

$K=\left[\begin{array}{ccc}0& 1& 0\\ 0& 1& 0\\ 0& 1& 0\end{array}\right]$

(4) structure based element template, as shown in formula [4], carries out secondary morphological erosion computing, expression is not more than maximum integer.

Step 207, based on area features, remove the connected region compared with small size, effect is as shown in Fig. 8 (a).

The position of step 208, accurately each Chinese character region, I.D. front, location, effect is as shown in Fig. 9 (a), and concrete steps are as follows:

(1) select uppermost connected region as name candidate region;

(2) two connected regions below selection name candidate region are as sex and national candidate region;

(3) select connected region below sex and national candidate region as birth candidate region;

(4) based on character pitch feature in candidate region, the accurately each Chinese character regional location in location.

Step 209, judge whether current face position is last face location, if so, export front Chinese character zone location result, otherwise, enter next face location, continue to perform step 202 to step 209.

Situation 2, I.D. back side Chinese character zone location algorithm, concrete steps are as follows:

(1) based on adaboost sorter, national emblem detection is carried out;

(2) based on the position of five stars in national emblem, carry out slant correction, effect is as shown in Fig. 5 (b), and concrete steps are as follows:

A () accurately orients the central point of five stars in national emblem;

B (), based on maximum star, orients the relative position of four starlets respectively;

C () is according to the calculating angle of inclination, center of two stars in outside;

D () carries out slant correction to image;

(3) based on national emblem position, select region, lower right to detect effective coverage as Chinese character, carry out inverse process, effect is as shown in Fig. 6 (b) simultaneously;

(4) similar step 204, carries out image blurring process, removes background patterns interference;

(5) similar step 205, uses vertical edge detective operators, and obtain the two-value vertical edge characteristic pattern of gray level image, effect is as shown in Fig. 7 (b);

(6) similar step 206, morphology operations, obtains image connectivity region;

(7) select uppermost connected region as Chinese character candidate region, the back side, effect is as shown in Fig. 8 (b);

(8) based on character pitch feature in candidate region, the accurately back side, location Chinese character regional location, effect is as shown in Fig. 9 (b);

(9) judge whether current national emblem position is last national emblem position, if so, export back side Chinese character zone location result, otherwise, enter next national emblem position, continue to perform step (2) to step (9).

Step 104, carry out character recognition to each Chinese character region, traditional way is that tessract storehouse is directly sent in gray scale Chinese character region, and the image processing function using tessract storehouse to carry carries out Chinese character location and identification, and recognition accuracy is lower; Therefore, character recognition is carried out in the Chinese character image that the present embodiment adopts quality to strengthen, tessract storehouse, digital template storehouse, and as shown in Figure 3, concrete steps are as follows:

Step 301, Accurate Segmentation Chinese character, as shown in Figure 10, concrete steps are as follows for effect:

(1) according to formula [6] and formula [7], local binarization is carried out to gray scale Chinese character region:

Formula [6]:

$g(x,y)=\left\{\begin{array}{cc}255& f(x,y)\&GreaterEqual;T\\ 0& f(x,y)<T\end{array}\right.$

Formula [7]:

$T=\frac{1}{M*N}\underset{m=1}{\overset{M}{\mathrm{\&Sigma;}}}\underset{n=1}{\overset{N}{\mathrm{\&Sigma;}}}f(x_m,y_n)+\mathrm{offset}$

Wherein, g (x, y) is the gray-scale value at pixel (x, y) place corresponding after binaryzation, and f (x, y) is the gray-scale value at binaryzation preceding pixel (x, y) place, f (x _m, y _n) be pixel (x in the front M*N neighborhood centered by pixel (x, y) of binaryzation _m, y _n) gray-scale value at place, M, N are width and the height of neighborhood respectively, and offset is side-play amount, are generally positive constants.

(2) according to size, less connected region interference is removed;

(3) carry out vertical projection, obtain the foreground target height of each row;

(4) by the position relationship of crest and trough, Character segmentation is carried out;

(5) secondary splitting is carried out to adhesion character;

(6) merging treatment is carried out to fracture character.

Step 302, judge that single character belongs to Chinese character or numeral by the width of character, if belong to Chinese character, enter step 303, otherwise, enter step 305.

Step 303, enhancing Chinese character pattern structure, concrete steps are as follows:

(1) add up the connected region number of single character, if be greater than 1, enter next step, otherwise, directly enter step 304;

(2) connect various piece connected region, as shown in figure 11, concrete steps are as follows for effect:

A () selects two connected regions arbitrarily, calculate minor increment therebetween;

B (), in minimum distance, is drawn straight line, is connected two connected regions, form new total connected region;

C () connects remaining connected region and total connected region successively, form final connected region.

Step 304, carry out Chinese Character Recognition based on tessract storehouse.

Step 305, to carry out numeral based on nearest neighbor algorithm and identify:

(1) according to formula [8] and formula [9], the characteristic distance dis between character to be identified and all template characters is calculated _i:

Formula [8]:

dis _i＝ΣΣs(x，y)

Formula [9]:

$s(x,y)=\left\{\begin{array}{cc}1& f(x,y)\&NotEqual;m(x,y)\\ 0& f(x,y)=m(x,y)\end{array}\right.$

Wherein, f (x, y) is the gray-scale value at pixel (x, the y) place of character to be identified, and m (x, y) is the gray-scale value at pixel (x, the y) place of template character;

(2) the template character selecting minimal characteristic distance corresponding is as recognition result.

Step 306, judge that whether current character is last character in current Chinese character region, if so, export Chinese Character Recognition result, otherwise, enter character late, continue to perform step 302 to step 306.

Step 105, judge whether current Chinese character region is last Chinese character region, if so, export I.D. recognition result, otherwise, enter next Chinese character region, continue to perform step 104 to step 105; Final Chinese Character Recognition result as shown in figure 12.

The above embodiment is only be described the preferred embodiment of the present invention; not scope of the present invention is limited; under not departing from the present invention and designing the prerequisite of spirit; the various distortion that those of ordinary skill in the art make technical scheme of the present invention and improvement, all should fall in protection domain that claims of the present invention determine.

Claims (9)

1. an I.D. Chinese characters recognition method, is characterized in that, comprises the following steps:

(1) sorter training file is obtained;

(2) coloured image to be identified is transformed into gray level image;

(3) the Chinese character region of I.D. is accurately located;

(4) character recognition is carried out to each Chinese character region;

(5) judge whether current Chinese character region is last Chinese character region, if so, then exports recognition result, if not, then return step (4).

2. I.D. Chinese characters recognition method according to claim 1, is characterized in that, in step (1), described acquisition sorter training file, specifically comprises:

(11) existing Face datection sorter file is loaded;

(12) tessract Chinese Character Recognition storehouse is loaded;

(13) based on harr characteristic sum adaboost algorithm, the sorter file about I.D. national emblem is trained;

(14) template of the numeral of training I.D. Chinese character region to contain.

3. I.D. Chinese characters recognition method according to claim 1, is characterized in that, in step (3), the described Chinese character region to I.D. is accurately located, and specifically comprises:

(31) based on adaboost sorter, the face location in I.D. front and the national emblem position at the I.D. back side is detected;

(32) according to eye position, slant correction is carried out to I.D. direct picture, according to five-pointed star position in national emblem, slant correction is carried out to I.D. back side image;

(33) based on face location and national emblem position, select the Chinese character of I.D. front and back to detect effective coverage respectively, carry out inverse process simultaneously;

(34) Fuzzy Processing is carried out to image, remove background patterns interference;

(35) the two-value vertical edge characteristic pattern of the gray level image after Fuzzy Processing is obtained;

(36) morphology operations is carried out to two-value vertical edge characteristic pattern, obtain connected region;

(37) according to area and position feature, corresponding connected region alternatively region is selected;

(38) based on character pitch feature in candidate region, each Chinese character region of I.D. front and back is accurately located;

(39) judge whether current face position or national emblem position are last face location or national emblem position, if so, then export corresponding Chinese character zone location result, if not, then continue to perform step (32) to step (39).

4. I.D. Chinese characters recognition method according to claim 1, is characterized in that, in step (4), describedly carries out character recognition to each Chinese character region, specifically comprises:

(41) Accurate Segmentation Chinese character;

(42) judge that single character is Chinese character or numeral by the width of character, if Chinese character, then perform step (43), if digital, then perform step (45);

(43) Chinese character pattern structure is strengthened;

(44) Chinese Character Recognition is carried out based on tessract storehouse;

(45) carry out numeral based on nearest neighbor algorithm to identify;

(46) judge that whether current character is last character in current Chinese character region, if so, then export Chinese Character Recognition result, if not, then enter character late, continue to perform step (42) to step (46).

5. I.D. Chinese characters recognition method according to claim 3, is characterized in that, in step (32), describedly carries out slant correction according to eye position to I.D. direct picture, specifically comprises:

(a1) based on five, three front yard layout rule, coarse positioning is carried out to left and right two;

(a2) central point of eyes is accurately oriented;

(a3) according to the calculating angle of inclination, center of two;

(a4) Sloped rotating correction is carried out to I.D. direct picture;

Describedly according to five-pointed star position in national emblem, slant correction is carried out to I.D. back side image, specifically comprises:

(b1) central point of five stars in national emblem is accurately oriented;

(b2) based on maximum star, the relative position of four starlets is oriented respectively;

(b3) according to the calculating angle of inclination, center of two stars in outside;

(b4) Sloped rotating correction is carried out to I.D. back side image.

6. I.D. Chinese characters recognition method according to claim 3, is characterized in that, in step (36), describedly carries out morphology operations to two-value vertical edge characteristic pattern, obtains connected region, specifically comprises:

A () Height value data to vertical connection edges all in two-value vertical edge characteristic pattern is added up;

B the Height value data counted sorts by () from big to small, and obtain the mean value of the altitude information come within 1/3rd positions above, as the average height at vertical connection edge in two-value vertical edge characteristic pattern

C () utilizes structural element template, carry out two-value vertical edge characteristic pattern secondary morphological dilations computing, wherein, expression is not more than maximum integer;

D () utilizes structural element template, to process the two-value vertical edge characteristic pattern of secondary morphological dilations computing carries out 2 closing operation of mathematical morphology;

E () utilizes structural element template, carry out the two-value vertical edge characteristic pattern through 2 closing operation of mathematical morphology secondary morphological erosion computing.

7. I.D. Chinese characters recognition method according to claim 4, is characterized in that,

In step (41), described Accurate Segmentation Chinese character, specifically comprises:

A () adopts following formula, carry out local binarization to gray scale Chinese character region:

$g(x,y)=\left\{\begin{array}{cc}255& f(x,y)\&GreaterEqual;T\\ 0& f(x,y)<T\end{array}\right.$

$T=\frac{1}{M*N}\underset{m=1}{\overset{M}{\mathrm{\&Sigma;}}}\underset{n=1}{\overset{N}{\mathrm{\&Sigma;}}}f(x_m,y_n)+\mathrm{offset}$

Wherein, the gray-scale value at pixel (x, the y) place that g (x, y) is corresponding after representing binaryzation, f (x, y) represents the gray-scale value at binaryzation preceding pixel (x, y) place, f (x _m, y _n) represent pixel (x in the front M*N neighborhood centered by pixel (x, y) of binaryzation _m, y _n) gray-scale value at place, M, N represent width and the height of neighborhood respectively, and offset represents side-play amount;

B (), according to size, filters interference connected region;

C () carries out vertical projection, obtain the foreground target height of each row;

D (), by the position relationship of crest and trough, carries out Character segmentation;

E () carries out secondary splitting to adhesion character;

F () carries out merging treatment to fracture character.

8. I.D. Chinese characters recognition method according to claim 4, is characterized in that, in step (43), described enhancing Chinese character pattern structure, specifically comprises:

A () adds up the connected region number of single character, judge whether to be greater than 1, if so, then perform step (b), if not, then directly performs step (44);

B () connects various piece connected region, specifically comprise:

(b1) select two connected regions arbitrarily, calculate minor increment therebetween;

(b2) in minimum distance, draw straight line, connect two connected regions, form new total connected region;

(b3) connect remaining connected region and total connected region successively, form final connected region.

9. I.D. Chinese characters recognition method according to claim 4, is characterized in that, in step (45), described based on nearest neighbor algorithm carry out numeral identify, specifically comprise:

A () adopts following formula, calculate the characteristic distance between character to be identified and all template characters:

dis _i＝ΣΣs(x，y)

$s(x,y)=\left\{\begin{array}{cc}1& f(x,y)\&NotEqual;m(x,y)\\ 0& f(x,y)=m(x,y)\end{array}\right.$

Wherein, dis _irepresent the characteristic distance between character to be identified and i-th template character, f (x, y) represents the pixel (x of character to be identified, y) gray-scale value at place, m (x, y) represents the gray-scale value at pixel (x, the y) place of template character;

B template character that () selects minimal characteristic distance corresponding is as recognition result.