CN102103693B - Method for identifying handwriting - Google Patents

Abstract

The invention relates to a method for identifying handwriting, comprising the following steps of: step 1, continuously rotating original characters for many times according to the setting degree so as to obtain a plurality of sample characters of rotating angles; and respectively calculating the deflection degree of each sample character; step 2, calculating the deflection oblique degree of the original character in each rotating angle according to the deflection degree of the sample character; step 3, calculating the initial value and a micro adjusting valve of a rotating correction angle of the original character according to the deflection oblique degree; and step 4, performing coordinate transformation to the original characters according to the initial value and the micro adjusting valve of the rotating correction angle; verifying the original characters so as to identify. The method provided by the invention has the characteristics of accurate error correction effect, better stability and wide application range and capable of supporting random angle conversion in a 360 range.

Description

A kind of hand-written character recognizing method

Technical field

The present invention relates to Data Enter field, particularly a kind of hand-written character recognizing method of supporting rotated type identification.

Background technology

As a kind of simple and convenient and efficient input mode, handwriting recognition has changed traditional keyboard mode based on complexity coding or Pinyin Input, is the man-machine interaction of the Natural humanity condition of providing convenience.In recent years, along with the development of mode identification technology, handwriting input efficiently highlights all the more its importance on day by day universal embedded device naturally.Isolated word recognition, namely processes by individual character recognition technology the monocase that user inputs one by one, because the maturation of core technology and recognition performance superior becomes on market the product of main flow the most.The online Chinese character hand-written recognition system of most is all supported writing frame input, requires user to write in default writing in region.Although the discrimination of higher level has been guaranteed in the setting of writing frame, it is provided with extra restriction to user writing.Particularly the finger in current increased popularity touches on touch-screen, and during due to finger input, scope of activities is larger, and the write mode of inputting in less writing frame is convenient not.

Full frame input pattern, is to support user in the mode of the optional position of touch-screen Free Writing, because of its more the characteristic of Natural humanity attracted increasing concern.Under full frame input pattern, user's input is freer, thereby the character of input also has larger randomness and variation.Wherein topmost distortion is that character tilts, that is to say character write angle and screen orientation has obvious deviation, its corresponding misalignment angle is referred to as the anglec of rotation of this word.

Fig. 1 has shown the deflection sample of Chinese character " greatly ", and the longitudinal axis principal direction of this character and the direction of touch-screen exist certain deviation.If this character with certain angle of inclination is directly inputted to the decline that must cause discrimination in the conventional recognizer that training obtains on normal sample, particularly when Chinese character pitch angle is greater than 30 None-identifieds almost while spending.

In Chinese character, stroke accounts for 3/4ths of all strokes anyhow, is the chief component composition of character.In this structural very strong character of Chinese character, write be generally all particular about the smooth vertical and horizontal while anyhow the angle of direction also have certain rule.And the different directions of stroke anyhow also have consistance in same character, after cluster, be conventionally dispersed in class central peripheral.At present, handwriting recognition field to the angle correction method of isolated character just based on to this two dot characteristics analysis of stroke anyhow.

Now, to support that the inclination word angle in (25,25) scope is example explanation, concrete analysis flow process as shown in Figure 2.

Step S201, collection written handwriting; In on-line handwritten recognition system, user handwriting represents with two dimension or the three-dimensional point coordinate of a sequence conventionally, in order to describe the room and time information of character writing.

Step S202, utilize various noise reduction means, as open country point, remove, the modes such as smoothing processing reduce the possible distortion of character.

Flex point on step S203, extraction written handwriting, and original continuous sampled point sequence is quantified as to segment sequence.

Step S204, calculate the presentation direction of each stroke: suppose that the starting point of certain unicursal and terminal are respectively P ₁=(X ₁, Y ₁), P ₂=(X ₂, Y ₂), the presentation direction of this stroke is $\mathrm{\θ}=\stackrel{\→}{P_1{P}_2}=\arctan \frac{Y_2-Y_1}{X_2-X_1}.$

Step S205, to stroke direction cluster, the pen section that direction is close is got together.Consider that Hanzi structure is comparatively complicated, between different Chinese character, stroke alters a great deal, and the cluster centre number of pen section direction is difficult to determine.Algorithm has adopted the hierarchical clustering method based on mean distance, and wherein distance metric is defined as the angle of pen section direction.

Step S206, according to cluster result determine bar and perpendicular pen direction (suppose here right horizontal direction be 0 degree and counterclockwise for just).

Judgement bar may direction: at the possible presentation direction of restriction bar, under the hypothesis of (30 °, 45 °), in conjunction with cluster result, obtain some direction cluster centres as possible bar direction; According to the angle between direction anyhow, limit to find the direction of perpendicular pen: system postulation anyhow stroke angle in (85,115) scope, by some bar directions, select the perpendicular direction satisfying condition; Further confirm again best bar direction.

Step S207, by the resulting value of optimum orientation anyhow, judge the inclined degree of character, character is corrected.Simply say, after having determined bar and a perpendicular direction, take 0 and-90 degree as reference direction anyhow, respectively character is proofreaied and correct in horizontal and vertical directions.

This algorithm by character anyhow the analysis of principal direction ask for the possible angle of inclination of character, carefully and neatly done word and anyhow the more character of stroke obtain good rectification effect.Yet because the direction anyhow of single character is vulnerable to various noise jamming, thereby lacking the character of stroke anyhow or existing on the character that more Nun stroke disturbs effect still unsatisfactory.Convenience and the validity in order to realize simultaneously, this algorithm can only be to the inclination character correction in limited range, as tilted in (25,25) angle, other is tilted to serious character and correct the levelling thickness range that exceeds this algorithm.

Summary of the invention

Object of the present invention provides a kind of hand-written character recognizing method, and the method error correction effect is accurate, and stability is better, and the scope of application is wider, can support the arbitrarily angled conversion in 360 scopes.

A kind of hand-written character recognizing method of the present invention, the method comprises:

Step 1, by former character by setting the rotation of number of degrees continuous several times, obtain the sample character of a plurality of anglecs of rotation, calculate respectively the deflection of each sample character;

Step 2, according to the deflection of sample character, calculate former character at the deflection tendency degree of each anglec of rotation;

Step 3, according to deflection tendency degree, calculate initial value and the trim values of former character rotation correction angle;

Step 4, according to initial value and the trim values of rotation correction angle, former character is carried out to coordinate transform, proofread and correct former character, identify.

Preferably, before step 1, also comprise: input character is extracted to segment sequence, calculate pen section total length in segment sequence, confirm that pen section total length is greater than 0.

Preferably, calculate pen section total length in segment sequence and be specially: the subtended angle of calculating character sampled point, if sampled point subtended angle is less than the threshold value of setting, confirm that this sampled point is flex point, between two flex points, form a section; Calculate respectively the length of each section, each segment length is added and obtains pen section total length.

Preferably, former character is specially by setting the rotation of number of degrees continuous several times: by former character, by each rotation 45 degree, order is rotated 8 times.

Preferably, the deflection that calculates each sample character is specially: calculate the decomposition value of each section of sample character in default 8 reference directions; According to this decomposition value, calculate respectively direction contribution margin, length contribution margin and the position contribution margin of each section; Direction contribution margin, length contribution margin and the position contribution margin of comprehensive each section, the deflection of calculating sample character.

Preferably, before calculating the decomposition value of each section of sample character, also comprise: determine the word frame of sample character, this word frame is the minimum rectangle horizontal pane that comprises sample character.

Preferably, calculating the decomposition value of each section of former character in 8 default reference directions is specially: during pen section is mapped to 8 reference directions with immediate two reference directions of this section on, calculate respectively this section at the decomposition value of above-mentioned two coordinate axis, the decomposition value of other 6 coordinate axis is made as to 0.

Preferably, during calculating location contribution margin, when pen section is the non-initial pen section of stroke, position contribution margin reduces by half.

Preferably, step 3 is specially: as the character deflection tendency degree of the adjacent anglec of rotation has different positive and negative values, and conversion from positive to negative in the direction of the clock, select corresponding on the occasion of deflection tendency degree for proofreading and correct the initial value of angle, according to the deflection tendency degree calculating of corresponding negative value, finely tune step-length.

Preferably, former character is carried out to coordinate transform and be specially: according to current iteration number of times, calculate when time correction angle, carry out coordinate transform; Judge whether current iteration number of times is less than the iterations setting in advance, if, and current correction angle is not equal to 0, and iterations adds 1, returns to step 1; Otherwise, finish former character correction.

Compared with prior art, the present invention has the following advantages:

The present invention utilizes character at the rotational symmetry along in counter clockwise direction, proposes to correct the method for deflection character.The error correction effect of the method is accurate, and stability is better, and the scope of application is wider, can support the arbitrarily angled conversion in 360 scopes.

Accompanying drawing explanation

Fig. 1 is a deflection sample figure;

Fig. 2 is existing inclination word analysis process flow diagram;

Fig. 3 is hand-written character recognizing method process flow diagram of the present invention;

Fig. 4 is sample character calculation flow chart;

Fig. 5 is word frame schematic diagram;

Fig. 6 is that pen section is at the exploded view of 8 reference directions;

Fig. 7 is character deflection tendency degree distribution plan.

Embodiment

For above-mentioned purpose of the present invention, feature and advantage can be become apparent more, below in conjunction with the drawings and specific embodiments, the present invention is further detailed explanation.

The present invention proposes a kind of rotation correction recognition methods based on character spin balancing.Suppose character to rotate same angle with clockwise direction in the counterclockwise direction respectively, obtain two rotated samples.These two rotated samples in pen section direction, length and pen section starting point, end point, show certain difference on the relative position in whole word frame with respect to the former character of initially just writing, and this difference is called deflection.Deflection has reflected that current character person's handwriting and the character anglec of rotation are 0 o'clock deviation between literary style.

According to normal angled, write, the anglec of rotation of former character is 0, and it is consistent substantially with the deflection that is rotated counterclockwise rear sample character clockwise, and deflection symmetry is best; Otherwise as former character has certain anglec of rotation, it is clockwise and after being rotated counterclockwise, the balance of sample character deflection can be broken, and is partial to the vergence direction of former character.

Visible, character is after suitable, contrary two clockwise rotations, and the balance that obtains sample character deflection has reflected the anglec of rotation of former character to a certain extent.Former character writing more approaches normal condition, and the deflection balance of former character is better, more symmetrical at the postrotational deflection degree of suitable, counterclockwise both direction.

The anglec of rotation of supposing former character is-α that its desirable offset angle that is corrected to normal writing position is α.So, the offset angle providing for former character more approaches α, and after compensation, the deflection symmetry of former character rotation in clockwise and counter clockwise direction is better.

Based on above-mentioned principle, the invention provides a kind of hand-written character recognizing method.Referring to Fig. 3, hand-written character recognizing method of the present invention is shown, specifically comprise the following steps.

Step S301, preset character initial inclination, arrange iterations N.Generally initial inclination 0, and the scope of iterations N is 3-7, and preferably 5.

Step S302, the continuous person's handwriting of original input is done to pretreatment operation, remove noise jamming, and extract segment sequence S.

Extract the serial S of pen section.This section is not that the stroke on conventional meaning is whole, and refers to the more basic substantially linear member with single presentation direction.It can be the stroke of traditional sense, can be also in a certain stroke, to have the part of single presentation direction.As " " word, by 4 pen sections, formed.The extraction of pen section is to locate starting point and the end point that can divide pen section, namely contacts the flex point in person's handwriting.

Inflexion extracting is the method that the subtended angle based on sample point is analyzed.So-called sample point subtended angle refers to the sample point of present analysis and the angle that before and after it, adjacent sample point forms.The subtended angle of flex point is less than common sampled point, calculates respectively the subtended angle of each sampled point, if certain sampled point subtended angle is less than the threshold value (as 120 degree) setting in advance, thinks that it is flex point.Obtain the flex point P={P of character ₁, P ₂..., P _i..., P _n+1, its corresponding segment sequence is S={S ₁, S ₂..., S _i..., S _n, wherein, pen section S _iby flex point P _iand P _i+1determine.

Length and the person's handwriting total length L of each section in step S303, calculating segment sequence S.

Calculate successively the length (l of each section ₁, l ₂..., l _i..., l _n), direction (a ₁, a ₂..., a _i..., a _n) (a _i∈ 0,360), the length of each section is added and obtains pen section total length $L=\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{l}_i.$

Step S304, judge whether the person's handwriting total length L of current character equals 0, if be to represent not have character input, termination character input; Otherwise proceed to step S305.

Step S305, by former character by setting the rotation of number of degrees continuous several times, obtain the sample character of a plurality of anglecs of rotation, calculate respectively the deflection of each sample character.For example, by 45 degree, sequentially rotate 8 times, be equivalent to former character and rotate successively 0 degree, 45 degree, 90 degree, 135 degree, 180 degree, 225 degree, 270 degree, 315 degree, obtain 8 sample characters after conversion, calculate respectively the deflection of 8 sample characters, be designated as respectively rb1, rb2 ..., rb8.

The present invention is initial in connection with pen section direction, length and pen section, the deflection of the feature calculation sample characters such as relative position of end point in whole " word frame ".Calculation process as shown in Figure 4.

41) determine the word frame of character, delimit character writing scope.

This word frame refers to comprise the minimum rectangle horizontal pane of sample character, and transverse and longitudinal axle and the locus of word frame are consistent.Be normally defined hexa-atomic parameter Box (L, R, T, B, W, H).Wherein, L refers to the left margin of word frame, and R is right margin, and T is coboundary, and B is lower boundary, and W, H is respectively the wide and high of word frame.As shown in Figure 5, character " he " is rotation inclination word, and the word frame of this character is and horizontal rectangular frame corresponding to locus.

42) calculate the decomposition value (see Fig. 5) of each section in 8 default reference directions, meet

this direction decomposition value has reflected that this section is subordinate to the probability of different coordinate axis.During actual computation, pen section is mapped in 8 reference directions in immediate two coordinate axis, and the decomposition value in other 6 coordinate axis is made as to 0.

In Fig. 6, due to pen section S _ibetween coordinate axis 1 and coordinate axis 8, and angle is respectively θ ₁and θ ₈, the component w2=w3=w4=w5=w6=w7=0 in other 6 coordinate axis.And the mapping calculation of the decomposition value w8 of the decomposition value w1 of the 1st coordinate axis and the 8th coordinate axis is as follows:

If a) θ ₁≤ 15 or θ ₈>=30:

and ω ₁=1-ω ₈; Wherein, 8 and 26 is preferred coefficient value.

B) if θ ₁> 15 or θ ₈< 30:

and ω ₈=1-ω ₁; Wherein, 32 and 7 is preferred coefficient value.

43) calculate the direction contribution margin of each section.Direction contribution margin is for describing the contribution of the presentation direction of each section to overall deflection.Consider the probability of occurrence of section on 8 reference coordinate axles of normal written character, a definition pen section S _idirection contribution margin W _afor:

If a). pen section s _istroke the initial segment, or the starting point of stroke and starting to write a little,

$W_a=1-\mathrm{OR}\left(\mathrm{OR}(\mathrm{OR}(\frac{{\mathrm{<figure-callout\; id="5"\; label="w"\; filenames="HDA0000051668430000051.png"\; state="\{\{state\}\}">w</figure-callout>}}_5}{2},w_6),w_7),\frac{w_8}{4})\right);$

B) otherwise.:

$W_a=1-\frac{\mathrm{OR}\left(\mathrm{OR}(\mathrm{OR}(\frac{{\mathrm{<figure-callout\; id="5"\; label="w"\; filenames="HDA0000051668430000051.png"\; state="\{\{state\}\}">w</figure-callout>}}_5}{2},w_6),w_7)\right)}{2};$

Visible, direction contribution margin is larger, and this field more may come from the character of normally writing.

44) calculate the length contribution margin of each section.Length contribution margin represents level and the contribution margin of vertical component to overall deflection of a segment length.Note pen section S _icomponent on horizontal ordinate 1 axle and ordinate 3 axles is Lxi and L _yi:

If a). pen section S _ihorizontal component W1=0 or W5=0, the offset of horizontal length component: ${\mathrm{\&Delta;}}_x=l_{x_i}\frac{w_1}{4}$ Or ${\mathrm{\&Delta;}}_x=-l_{{\mathrm{<figure-callout\; id="5"\; label="x\; i\; w"\; filenames="HDA0000051668430000051.png"\; state="\{\{state\}\}">x</figure-callout>}}_i}\frac{w_{}}{}\frac{{}_5}{8}$

B) if. pen section S _ivertical component W3=0 or W7=0, the offset of vertical length component: ${\mathrm{\&Delta;}}_y=l_{y_i}\frac{{\mathrm{<figure-callout\; id="3"\; label="w"\; filenames="HDA0000051668430000051.png"\; state="\{\{state\}\}">w</figure-callout>}}_3}{4}$ Or ${\mathrm{\&Delta;}}_y=-l_{y_i}\frac{w_7}{8}$

And pen section S _ilength overall contribution value be:

$W_{\mathrm{li}}=\mathrm{OR}(\frac{1_{\mathrm{xi}}+\mathrm{\&Delta;x}}{L},\frac{1_{\mathrm{yi}}+\mathrm{\&Delta;y}}{L}).$

45) calculate the position contribution margin of each section.Position contribution margin is mainly initial by portraying pen section, and the relative position of end point in whole word frame reflects its contribution margin to overall deflection.Suppose pen section S _istarting point and end point be respectively P _1i=(x _1i, y _1i), P _2i=(x _2i, y _2i), the position contribution margin W of starting point _psiposition contribution margin W with end point _peibe respectively:

$W_{\mathrm{psi}}=\frac{\max (\frac{\mathrm{Box}.R-x_{1i}}{\mathrm{Box}.W},\frac{\mathrm{Box}.B-y_{1i}}{\mathrm{Box}.H})}{2^i}$ And

$W_{\mathrm{pei}}=\frac{\max (\frac{x_{2i}-\mathrm{Box}.L}{\mathrm{Box}.W},\frac{y_{2i}-\mathrm{Box}.T}{\mathrm{Box}.H})}{2^N-i+1}$

Pen section S _iduring for the initial pen section of stroke, pen section S _iposition contribution margin be calculated as W _pi=OR (W _psi, W _pei); As pen section S _iduring for the non-initial pen section of stroke, position contribution margin reduces by half and is

46) calculate the contribution margin of this section, the deflection using the overall contribution value of all pen sections as character.

Comprehensive pen section S _idirection contribution margin, length contribution margin and position contribution margin, calculating pen section S _icontribution margin be:

W _i＝OR(W _ai*W _li，W _pi)。

All pen section contribution margins are done to fuzzy logic exclusive disjunction, obtain the overall contribution value of sample character, as the deflection of sample character, be:

rb＝1-OR(OR(...OR(OR(W ₁，W ₂)，W ₃)，...，W _i，...)W _N)

Visible, overall contribution value is larger, and the deflection of sample character is less.Repetition above-mentioned steps 1)-6), calculate successively the deflection of each sample character.It should be noted that above-mentioned steps 1)-6) be comparatively reasonable in various calculating character deflections, result of calculation is a kind of comparatively accurately.

Step S306, calculate former character at the deflection tendency degree of each anglec of rotation.

Because character deflection tendency degree has reflected that it is at suitable anticlockwise skewed popularity, thereby can along counter clockwise direction, according to the deflection difference of the postrotational rotated type of fixed angle, embody by former character, calculate:

rt(α)＝|rb(α+θ)-rb(α-θ)|

Wherein, rt (α) is the deflection tendency degree to character after former character clockwise direction rotation alpha angle, and this value obtains by calculating the difference of rb (α ± θ).After rb (α ± θ) represents current rotation alpha angle, sample character continues along the character deflection that continues counterclockwise rotation θ angle.

In the present invention, the average propensity degree of utilization based on multipair anglec of rotation deflection calculates the cyclovergence degree of former character:

$\mathrm{rt}\left(\mathrm{\&alpha;}\right)=\frac{1}{3}\underset{i=1}{\overset{3}{\mathrm{\&Sigma;}}}(\mathrm{rb}(\mathrm{\&alpha;}+i*\mathrm{\&theta;})-\mathrm{rb}(\mathrm{\&alpha;}-i*\mathrm{\&theta;}))$

Wherein, θ=45 °.

Character deflection tendency degree means that person's handwriting is in " balance " of the deflection along counterclockwise rotating.Should " balance " refer to by person's handwriting on current expection anglec of rotation basis in the counterclockwise direction and clockwise direction rotate the symmetry between two rotated samples that same angle obtains, for general Chinese character, when the expection anglec of rotation while being 0 this " balance " best, therefore can be using " balance " that find according to character deflection tendency degree best anglec of rotation as rotation correction angle.

Initial value and the trim values of step S307, calculating input person's handwriting rotation correction angle.

First according to character, the deflection tendency degree in each sense of rotation is determined initial value and the trim values of RA.In general, if adjacent two rotation angle cause the character deflection tendency degree with distinct symbols, and conversion from positive to negative in the direction of the clock, selecting the corresponding deflection tendency degree on the occasion of character is initial correction value a ₀, according to the deflection tendency degree of corresponding negative value character, calculate fine setting step-length a1.

Fig. 7 has provided an embodiment.Input person's handwriting deflection tendency degree of postrotational rotated type on 8 different directions is followed successively by rb1 (> 0), rb2 (< 0), rb3 (< 0), rb4 (< 0), rb5 (< 0), rb6 (> 0), r7 (> 0) and rb8 (> 0).It is correction angle initial value that the sense of rotation that rb1 is corresponding is set, and rb2 value is correction angle trim values, i.e. a0=0.And a1=A/ (rb2).Here A is the systematic parameter arranging based on experience, for cyclovergence degree normalized.

Step S308, calculate when time proofreading and correct the angle row-coordinate conversion of going forward side by side.

Input character is made to Rotating Transition of Coordinate, and the anglec of rotation is a=(a ₀+ a ₁) * ∈ _t, wherein

t is current iteration number of times, and the iterations that N is system to be set in advance.

Step S309, judge whether current iteration number of times is less than the iterations setting in advance, and if rotation correction angle a is not equal to 0, iterations n adds 1 and return to step S305; Otherwise, finish former character correction.

The present invention utilizes character at the rotational symmetry along in counter clockwise direction, proposes to correct the method for deflection character.The error correction effect of the method is accurate, and stability is better, and the scope of application is wider, can support the arbitrarily angled conversion in 360 scopes.

The foregoing is only the preferred embodiment of the present invention, do not form limiting the scope of the present invention.Any any modification of doing within the spirit and principles in the present invention, be equal to and replace and improvement etc., within all should being included in claim protection domain of the present invention.

Claims (9)

1. a hand-written character recognizing method, is characterized in that, the method comprises

Step 1, by former character by setting the rotation of number of degrees continuous several times, obtain the sample character of a plurality of anglecs of rotation, calculate respectively the deflection of each sample character;

Step 2, according to the deflection of sample character, calculate former character at the deflection tendency degree of each anglec of rotation; Wherein, the deflection tendency degree of former character when the anglec of rotation is α be | rb (α+θ)-rb (α-θ) | and, rb (α ± θ) continues along being rotated counterclockwise the deflection after θ angle for the sample character after former character rotation alpha angle;

Step 3, according to deflection tendency degree, calculate initial value and the trim values of former character rotation correction angle;

Step 4, according to initial value and the trim values of rotation correction angle, former character is carried out to coordinate transform, proofread and correct former character, identify;

Step 3 is specially:

As the character deflection tendency degree of the adjacent anglec of rotation has different positive and negative values, and conversion from positive to negative in the direction of the clock, select corresponding on the occasion of deflection tendency degree for proofreading and correct the initial value of angle, according to the deflection tendency degree calculating of corresponding negative value, finely tune step-length;

Described fine setting step-length a1 is a1=A/(rb2), and wherein A is for cyclovergence degree normalized, the deflection tendency degree that rb2 is described corresponding negative value.　

2. the method for claim 1, is characterized in that, before step 1, also comprises:

Input character is extracted to segment sequence, calculate pen section total length in segment sequence, confirm that pen section total length is greater than 0.　

3. method as claimed in claim 2, is characterized in that, calculates pen section total length in segment sequence and is specially:

The subtended angle of calculating character sampled point, if sampled point subtended angle is less than the threshold value of setting, confirms that this sampled point is flex point, forms a section between two flex points;

Calculate respectively the length of each section, each segment length is added and obtains pen section total length.　

4. the method for claim 1, is characterized in that, former character is specially by setting the rotation of number of degrees continuous several times:

By former character, by each rotation 45 degree, order is rotated 8 times.　

5. the method for claim 1, is characterized in that, the deflection that calculates each sample character is specially:

Calculate the decomposition value of each section of sample character in default 8 reference directions;

According to this decomposition value, calculate respectively direction contribution margin, length contribution margin and the position contribution margin of each section;

Direction contribution margin, length contribution margin and the position contribution margin of comprehensive each section, the deflection of calculating sample character;

Described sample character has n pen section, is respectively { S ₁, S ₂..., S _i..., S _n;

Wherein, comprehensively the direction contribution margin of each section, length contribution margin and position contribution margin, the deflection that calculates sample character is specially:

Comprehensive pen section S _idirection contribution margin W _ai, length contribution margin W _liwith position contribution margin W _pi, calculating pen section S _icontribution margin be: W _i=OR(W _ai* W _li, W _pi), all pen section contribution margins are made to fuzzy logic exclusive disjunction, the deflection rb that obtains described sample character is:

ｒｂ　＝1－　ＯＲ（ＯＲ（...ＯＲ（ＯＲ（W ₁，W ₂），W ₃），...，W _i，...）W _N）　。　

6. method as claimed in claim 5, is characterized in that, before calculating the decomposition value of each section of sample character, also comprises:

Determine the word frame of sample character, this word frame is the minimum rectangle horizontal pane that comprises sample character.　

7. method as claimed in claim 5, is characterized in that, calculates the decomposition value of each section of former character in 8 default reference directions and is specially:

By pen section be mapped in 8 reference directions with immediate two reference directions of this section on, calculate respectively this section at the decomposition value of above-mentioned two reference directions, the decomposition value of other 6 reference directions is made as to 0.　

8. method as claimed in claim 5, is characterized in that, during calculating location contribution margin, when pen section is the non-initial pen section of stroke, position contribution margin reduces by half.　

9. the method for claim 1, is characterized in that, former character is carried out to coordinate transform and be specially:

According to current iteration number of times, calculate when time correction angle, carry out coordinate transform;

Judge whether current iteration number of times is less than the iterations setting in advance, if, and current correction angle is not equal to 0, and iterations adds 1, returns to step 1; Otherwise, finish former character correction.　

Images