CN103268490B

CN103268490B - A kind of digit recognition method adopting both sides three quant's sign

Info

Publication number: CN103268490B
Application number: CN201310209889.XA
Authority: CN
Inventors: 徐进; 陈银; 方道恒; 傅志中; 周宁
Original assignee: University of Electronic Science and Technology of China
Current assignee: University of Electronic Science and Technology of China
Priority date: 2013-05-30
Filing date: 2013-05-30
Publication date: 2016-01-13
Anticipated expiration: 2033-05-30
Also published as: CN103268490A

Abstract

The invention discloses a kind of digit recognition method adopting both sides three quant's sign, belong to field of image recognition.The present invention is first by the monocase image normalization that the collects suitable dimension to process, then the identification of 0 ~ 90 numerical character is carried out, the basic thought identified is by the resemblance discriminating digit of numeral, is specially: the left envelope EL asking for character, right envelope ER; Calculate head breadth WH, wide WM and the WM0 of waist and the wide WF of pin, namely both sides three are wide; Based on three quant's sign identification character " 7 " and " 1 "; Completed by the detection of 3 subitems and judge character " 4 "; Ask for left envelope opening parameter L0, L1, right envelope opening parameter R0, R1, based on opening and the closing characteristics of left and right envelope, Classification and Identification character " 0 ", " 2 ", " 3 ", " 5 ", " 6 ", " 8 ", " 9 ".The present invention is used for the identification of printing digital, can realize the accurately quick identification of the printing digital to different fonts, different size, different sharpness on bill, its recognition correct rate and real-time high.

Description

A kind of digit recognition method adopting both sides three quant's sign

Technical field

The invention belongs to the technical field of image recognition, particularly relate to the recognition methods to press figure.

Background technology

Along with popularizing of computer, the most work of people all will complete on computers, and the information of numeral useful on a lot of bill needs manually to be input in computer, as the date on bill, the amount of money, voucher code etc.For the numeral on these bills, typing is manually power consuming time not only, and easily make mistakes, and therefore, on bill, the automatic identification of printing digital has very large researching value and application prospect.

Find through summing up prior art, printing digit recognizing method mainly contains two large classes: the method for structure based feature and the method for Corpus--based Method feature.The classification of Using statistics feature is easy to training, and can obtain higher discrimination on given training set, but cannot ensure recognition effect outside training set.Use the method for architectural feature to utilize the fixed sturcture of character to extract feature, integrated structure information effectively in identifying, has stronger adaptability compared with the method for Corpus--based Method feature.Configuration mode identification can be divided into for skeleton, profile, topography's three kinds of methods.

Digital identification techniques is widely used in every field in recent years, and digital identification techniques is ripe day by day, and the numeral identification of block letter has also reached very high recognition correct rate.But in actual applications, the method robustness of a lot of character recognition is poor, can only be applicable to the identification of the printed character of single type, and can decline much for its recognition accuracy of multiple font, meanwhile, algorithm complex is higher.

Summary of the invention

Goal of the invention of the present invention is: provide a kind of digit recognition method adopting both sides three quant's sign, to realize, method for quickly identifying simple to the printing digital of different fonts, different size, different sharpness.

The present invention performs the following step to the monocase image collected:

Step S1: pre-service is carried out to character picture;

Step S2: the left envelope EL extracting character, right envelope ER, described left envelope EL are the distance value of character leftmost side each point apart from the left front scene vegetarian refreshments of character, and right envelope ER is the distance value of character rightmost side each point apart from the right front scene vegetarian refreshments of character;

Step S3: based on the height H of character picture, by the head breadth WH of character, the wide WM of maximum waist, the wide WMO of minimum waist, the wide WF identification character " 7 " of pin or " 1 ",

Described WH be character picture (0, aH] in region, the breadth extreme of foreground pixel, wherein a ∈ [0.15,0.3];

Described WM and WMO is respectively in [bH, cH] region of character picture, the breadth extreme of foreground pixel, minimum widith, wherein b ∈ [0.4,0.55], c ∈ [0.65,0.75];

Described WF is in [dH, H] region of character picture, the breadth extreme of foreground pixel, wherein d ∈ [0.75,0.85];

Step S4: based on the belt-like zone detecting the font style characteristic that whether there are upper left corner triangle white space and two formations character " 4 " in character picture, identification character " 4 ";

Step S5: based on opening and the closing characteristics of the left and right envelope of character, Classification and Identification character " 0 ", " 2 ", " 3 ", " 5 ", " 6 ", " 8 ", " 9 ":

By upper under, left envelope first minimum point getting character picture is L0, and the last minimum point of left envelope is L1, and right envelope first minimum point is R0, and the last minimum point of right envelope is R1; As L1≤L0, for left envelope closes; During R1≤R0, for right envelope closes; L1>L0 is left envelope opening; During R1>R0, it is right envelope opening;

S501: if left and right envelope is all closed, then based on the pixel distribution of character, in conjunction with corresponding font style characteristic identification character based on corresponding font style characteristic identification character " 0 " or " 8 ";

S502: if left envelope opening, then based on corresponding font style characteristic identification character " 3 ";

S503: if left envelope opening and right envelope close, then based on corresponding font style characteristic identification character " 0 ", " 3 ", " 8 " or " 9 ";

S504: if right envelope opening and left envelope close, then based on corresponding font style characteristic identification character " 8 ", " 0 " or " 6 ";

S505: if the equal opening of left and right envelope, then based on corresponding font style characteristic identification character " 8 ", " 2 " or " 5 ".

In step S5 of the present invention, for different openings and the closed situation of the left and right envelope of character, during based on corresponding font style characteristic identification kinds of characters, the pixel distribution based on character can be adopted, identification that the mode in conjunction with corresponding font style characteristic carries out kinds of characters, also can be other ways customary.

The invention has the beneficial effects as follows: compared with prior art, the present invention is based on the research to general character in most of printing digital form, extract common trait---both sides three quant's sign of the same character of different printing body, set up the model of character recognition, recognition accuracy is higher, strong robustness.

Accompanying drawing explanation

Examples of the present invention will be described by way of reference to the accompanying drawings, wherein:

Fig. 1 is in the specific embodiment of the invention, the process flow diagram that numeral identifies;

Fig. 2 is in the specific embodiment of the invention, image left and right envelope and three wide mark schematic diagram;

Fig. 3 is in the specific embodiment of the invention, the upper left A test zone schematic diagram of character " 4 ";

Fig. 4 is in the specific embodiment of the invention, opening parameter mark schematic diagram.

Embodiment

All features disclosed in this instructions, or the step in disclosed all methods or process, except mutually exclusive feature and/or step, all can combine by any way.

Arbitrary feature disclosed in this instructions (comprising any accessory claim, summary and accompanying drawing), unless specifically stated otherwise, all can be replaced by other equivalences or the alternative features with similar object.That is, unless specifically stated otherwise, each feature is an example in a series of equivalence or similar characteristics.

Main thought of the present invention is: the resemblance discriminating digit using numeral, the width of several key components of resemblance cardinal index word, left and right sides envelope forms etc., except 1,4,7, clearly inerrably can differentiate other numerals according to the left and right envelope morphological feature of numeral.Method is as follows: the left envelope EL asking for character, right envelope ER; Calculate head breadth WH, wide WM and the WM0 of waist and the wide WF of pin, namely both sides three are wide; Based on three quant's sign discriminating digits 7 and 1; 4 word features are comparatively special, complete judgement by the detection of 3 subitems; Ask for left envelope opening parameter L0, L1, right envelope opening parameter R0, R1, based on opening and the closing characteristics of left and right envelope, Classification and Identification numeral 2,3,5,6,8,9,0.The method achieve the accurately identification fast of the printing digital to different fonts, different size, different sharpness on bill, there is very high recognition correct rate and real-time in actual applications.

See Fig. 1, specific implementation process of the present invention is, first pre-service is carried out to the monocase image collected (monocase image normally obtains by carrying out Character segmentation to the printing digital string image collected), namely suitable dimension is normalized to, the height of character picture is H, wide is W, and extracts character left and right envelope, then carries out 0 ~ 90 digital identification.

First character left and right envelope is extracted:

See Fig. 2, in the present invention, left envelope EL is defined as the distance value of character leftmost side each point apart from the left front scene vegetarian refreshments of character, and right envelope ER is defined as the distance value of character rightmost side each point apart from the right front scene vegetarian refreshments of character.

Identification step (1), the three quant's sign discriminating digits 7 and 1 based on character:

1. head breadth WH, wide WM and the WM0 of waist, the wide WF of pin is calculated, wherein,

WH be character picture (0, aH] in region, the breadth extreme of foreground pixel, wherein a ∈ [0.15,0.3];

WM and WMO is respectively in [bH, cH] region of character picture, the breadth extreme of foreground pixel, minimum widith, wherein b ∈ [0.4,0.55], c ∈ [0.65,0.75];

WF is in [dH, H] region of character picture, the breadth extreme of foreground pixel, wherein d ∈ [0.75,0.85];

Such as, head breadth WH of the present invention, wide WM and the WM0 of waist, the wide WF of pin can ask for according to following formula:

WH＝max(W-EL[i]-ER[i])；i＝0,1,...,(0.3H-1)

WM＝max(W-EL[i]-ER[i])；i＝0.4H,...,0.7H

WM0＝min(W-EL[i]-ER[i])；i＝0.4H,...,0.7H

WF＝max(W-EL[i]-ER[i])；i＝0.8H,...,H-1

2. based on WH, WM, WM0, WF value calculated, current character image is identified:

If meet: WH ﹥ [0.55,0.65] W, WM >=[0.95,1.05] WF, (WH-WF) >=[0.2,0.3] WH, (WH-WM) >=[0.2,0.3] WM, WH ﹥ [1.95,2.05] WF; Then judge that current character is as " 7 ", and terminate identifying operation.

If meet: WH≤[0.55,0.65] W, (WM-WM0)≤[0.15,0.25] W, WM≤[0.4,0.5] W, WF >=[0.25,0.35] W, then judge that current character is as " 1 ", and terminate identifying operation.

If current character can not be judged based on identification step (1), then enter identification step (2).

Identification step (2), by A, B and C three grades detection, namely detects the font style characteristic that whether there are upper left corner triangle white space and two formations character " 4 " in character picture, realizes the identifying processing to character " 4 ":

A detects: see Fig. 3, the upper left corner of statistics character picture, and the length of side is in the region of isosceles right triangle Rt of Tran_w, and pixel value is not the pixel summation SV of 0, if SV≤T1, then upper left corner triangle white space detected, continues to perform B and detects; Otherwise proceed to next identification step, i.e. step (3);

Wherein the value of length of side Tran_w is [0.35,0.45] W; T1 is predetermined threshold value, and its value is the area of k1 times of isosceles right triangle Rt; The value of k1 is [0.0042,0.0156].

B detects: to the r row pixel of character picture, by last column up scanning probe of character picture, get the behavior C1 that first is not the pixel place of 0, the summation SHC of foreground pixel in accumulative (C1-1), (C1-2) and (C1-3) row, if SHC >=T2, Article 1 belt-like zone then detected, continue to perform C and detect; Otherwise proceed to identification step (3);

Wherein, (C1-1) represents the lastrow of C1, and roughly the same, the value of the r of r row is [2,6] for (C1-2), (C1-3), and the value of predetermined threshold value T2 is for being [0.55,0.65] × 3W.

C detects: to average R1 to right envelope ER from the c1 of character picture is capable to c2 is capable, the summation SVC of foreground pixel in statistics (W-R1-4), (W-R1-3) and (W-R1-2) row, if SVC >=T3, Article 2 belt-like zone then detected, judge that current character is as " 1 ", and terminate identifying operation; Otherwise proceed to identification step (3);

Wherein (W-R1-4) represents that the width of character picture deducts the value of R1 and 4, and roughly the same, the value of predetermined threshold value T3 is [0.55,0.65] × 3H for (W-R1-3), (W-R1-2).

Identification step (3): based on opening and the closing characteristics of the left and right envelope of character, Classification and Identification character " 0 ", " 2 ", " 3 ", " 5 ", " 6 ", " 8 ", " 9 ":

First, the opening parameter of calculating character image, judges left and right envelope opening and closes and characteristic, as shown in Figure 4, is split shed parameter definition figure of the present invention;

As L1>L0:

Calculate the left envelope size LM between opening L0 ~ L1, if described LM≤0.1 (H × W), then left envelope aperture area grade LV=0; [if 0.08.1.02] (H × W) be <LM≤[0.22,0.28] (H × W), then described LV=1; Otherwise, LV=2;

Calculate left envelope peak position LI, if described LI≤0.4H, then left peak grade LS=0; [if 0.38,0.42] H<LI≤[0.68,0.72] H, then described LS=1; Otherwise LS=2;

As R1>R0:

Calculate the right envelope size RM between opening R0 ~ R1, if RM≤0.1 (H × W), left envelope aperture area grade RV=0; [if 0.08.1.02] (H × W) be <RM≤[0.22,0.28] (H × W), then described RV=1; Otherwise, RV=2;

Calculate right envelope peak position RI, if RI≤0.4H, then right peak grade RS=0; [if 0.38,0.42] H<RI≤[0.68,0.72] H, then described RS=1; Otherwise RS=2;

Based on above-mentioned parameter L0, L1, R0, R1, LV, LS, RV, RS, the detailed process realizing character recognition is:

1. as L1≤L0 and R1≤R0 time:

Extract the character parameter of character " 0 " or " 8 ", to identify that current character is " 0 " or " 8 ":

Detect and determine hole heart height H 0, the H1 in the upper and lower hole on character picture perpendicular bisector, and diameter D0, the D1 in upper and lower hole, judge whether the hole heart is less than or equal to [0.45 apart from ︱ H0-H1 ︱, 0.55] × (D0+D1), if so, then current character possesses the character property of the character property " 8 " of " 8 "; Otherwise be the character property of " 0 ".

2. as L1>L0:

First from (L0-3) OK, in the scope of (L1+3) row, calculate left envelope difference, find out first point being greater than [0.16,0.24] W with last difference absolute value and be designated as i and j respectively; If the distance of i point distance character picture top margin is less than [0.46,0.54] H, the distance of some j distance character picture top margin is less than [0.56,0.64] H, then judge that current character is as " 3 ", and terminate identifying operation; Otherwise carry out next again to identify and judge (i.e. L1>L0 and the situation of R1≤R0);

3. as L1>L0 and R1≤R0 time:

If LV<1, then first based on the character property of " 8 ", identify whether current character is " 8 ", if so, then judges that current character is as " 8 ", terminate identifying operation; Otherwise by the value of LV and LS, based on bivariate table I, obtain recognition result, terminate identifying processing, described bivariate table I is:

4. as L1≤L0 and R1>R0 time:

If RV<1, first based on the character property of " 8 ", identify whether current character is " 8 ", if so, then terminates identifying processing; Otherwise by the value of RV, RS, based on two-dimentional Table II, obtain recognition result, terminate identifying processing, described two-dimentional Table II is:

5. as L1>L0 and R1>R0 time:

First extract the character parameter of character " 8 ", identify whether current character is " 8 ", if so, then terminates identifying processing;

If can not identify, current character is " 8 ", then proceed following two condition judgment:

I () judges whether to meet: LI<RI and RI>0.6H, then by the value of RV, LV, based on two-dimentional Table III, obtain recognition result, terminate identifying processing, described two-dimentional Table III is:

In two dimension Table III, "-1 " represents that current character is nonnumeric character;

(ii) judge whether to meet LI>RI, and RI<0.5H, then judge that current character is as " 5 ", terminate identifying operation.

The present invention is not limited to aforesaid embodiment.The present invention expands to any new feature of disclosing in this manual or any combination newly, and the step of the arbitrary new method disclosed or process or any combination newly.

Claims

1. adopt a digit recognition method for both sides three quant's sign, it is characterized in that, the following step is performed to the monocase image collected:

Step S1: pre-service is carried out to character picture;

Step S3: based on the height H of character picture, by the wide WF identification character " 7 " of the head breadth WH of character, the wide WM of maximum waist, the wide WMO of minimum waist and pin or " 1 ",

S501: if left and right envelope is all closed, then based on corresponding font style characteristic identification character " 0 " or " 8 ";

2. the method for claim 1, is characterized in that, in described step S3,

If meet: WH ﹥ [0.55,0.65] W, WM >=[0.95,1.05] WF, (WH-WF) >=[0.2,0.3] WH, (WH-WM) >=[0.2,0.3] WM, WH ﹥ [1.95,2.05] WF, then current character is " 7 ";

If meet: WH≤[0.55,0.65] W, (WM-WM0)≤[0.15,0.25] W, WM≤[0.4,0.5] W, WF >=[0.25,0.35] W, then current character is " 1 ";

Wherein W is the width of character picture.

3. the method for claim 1, is characterized in that, in described step S4, detects the belt-like zone of the font style characteristic that whether there are upper left corner triangle white space and two formations character " 4 " in character picture, is specially:

S301: the upper left corner of statistics character picture, the length of side is in the region of isosceles right triangle Rt of Tran_w, and pixel value is not the pixel summation SV of 0, if described SV is less than or equal to threshold value T1, then upper left corner triangle white space detected;

The area of the value of wherein said length of side Tran_w to be [0.35,0.45] W, threshold value T1 be k1 times of isosceles right triangle Rt, the value of described k1 is [0.0042,0.0156];

S302: to the r row pixel of character picture, by last column up scanning probe of character picture, get the behavior C1 that first is not the pixel place of 0, the summation SHC of foreground pixel in accumulative (C1-1), (C1-2) and (C1-3) row, if described SHC is more than or equal to threshold value T2, then Article 1 belt-like zone detected;

The value of described r is [2,6], and the value of threshold value T2 is [0.55,0.65] × 3W;

S303: to c2 is capable, right envelope ER is averaged R1 from the c1 of character picture is capable, the summation SVC of foreground pixel in statistics (W-R1-4), (W-R1-3) and (W-R1-2) row, if SVC is more than or equal to threshold value T3, then Article 2 belt-like zone detected;

Wherein said W is the width of character picture, and the value of threshold value T3 is [0.55,0.65] × 3H.

4. the method for claim 1, is characterized in that, in described step S501, to the concrete identification step of character " 0 " and " 8 " is:

Detect and determine hole heart height H 0, the H1 in the upper and lower hole on character picture perpendicular bisector, and diameter D0, the D1 in upper and lower hole, judge whether the hole heart is less than or equal to [0.45,0.55] × (D0+D1) apart from ︱ H0-H1 ︱, if so, then current character possesses the character property of " 8 "; Otherwise be the character property of " 0 ".

5. method as claimed in claim 4, is characterized in that, in described step S502, be specially based on corresponding font style characteristic identification character " 3 ":

From (L0-3) OK, in the scope of (L1+3) row, calculate left envelope difference, find out first point being greater than [0.16,0.24] W with last difference absolute value and be designated as i and j respectively;

If the distance of i point distance character picture top margin is less than [0.46,0.54] H, the distance of some j distance character picture top margin is less than [0.56,0.64] H, then current character is " 3 ".

6. method as claimed in claim 5, is characterized in that, in described step S5:

As L1>L0:

As R1>R0:

Then described step S503-S505 is specially:

S503: if LV<1, then first based on the character property of " 8 ", identify whether current character is " 8 ", if so, then judges that current character is as " 8 ", terminate identifying operation; Otherwise by the value of LV and LS, based on bivariate table I, obtain recognition result, terminate identifying processing, described bivariate table I is:

S504: if RV<1, first based on the character property of " 8 ", identifies whether current character is " 8 ", if so, then terminates identifying processing; Otherwise by described RV, RS value, based on two-dimentional Table II, obtain recognition result, terminate identifying processing, described two-dimentional Table II is:

S505: first based on the character property of " 8 ", identifies whether current character is " 8 ", if so, then terminates identifying processing;

Otherwise continue to judge whether to meet: LI<RI and RI>0.6H, then by the value of described RV, LV, based on two-dimentional Table III, obtain recognition result, terminate identifying processing, described two-dimentional Table III is:

If LI>RI, and RI<0.5H, then judge that current character is as " 5 ", terminate identifying operation.