CN1093965C

CN1093965C - On-line character recognition method and apparatus thereof

Info

Publication number: CN1093965C
Application number: CN96121613A
Authority: CN
Inventors: 川又武典; 小川勇; 冈野祐一; 龟代泰三; 河村彻
Original assignee: Mitsubishi Electric Corp
Current assignee: Mitsubishi Electric Corp
Priority date: 1996-01-16
Filing date: 1996-11-13
Publication date: 2002-11-06
Anticipated expiration: 2016-11-13
Also published as: KR100236247B1; CN1155129A; KR970059977A; TW315446B; JPH09198466A; JP3360513B2

Abstract

To provide a high-accuracy recognition and also to provide an on-line character recognizing device capable of a high-speed recognition by providing a step, etc., for outputting a standard character selected in a character form recognition step corresponding to the degree of similarity calculated in a stroke order mutual correspondence recognition step. Based on a stroke sequence sent from a control means 59, a character form recognizing means 52 outputs candidate characters from the distance between a standard pattern in a character form dictionary and an input pattern. The control means 59 sends the candidate characters to a stroke order mutual correspondence recognizing means 54 and at this means 54, the features of respective strokes are extracted from the input pattern. Next, the strokes are collated with the standard pattern in a stroke feature dictionary 53 for candidate characters. At a character form recognizing means 52, the number of strokes are collated and when they are coincident, they are made to correspond to each other using the standard pattern, etc., in the stroke feature dictionary 53. Next, the distance between the stroke features of both the input and standard patterns is calculated, the distance between the respective features is weighted, and the final distance to the standard pattern is found.

Description

Online character recognition method and online character recognition device

The present invention relates to handwriting is carried out the online character recognition method and the device of high precision identification.

In online character recognition device in the past, have by drawing an online character recognition device of preface change identification literal and passing through to draw the online character recognition device that the identification literal is drawn in a preface change.Below describe with regard to various online character recognition devices respectively.

Prior art example 1:

Drawing in the existing online character recognition device of identification literal by drawing a preface change, be between all strokes of all strokes of the standard font of character library and input font, ask for the distance of combination separately of stroke, do corresponding name by the minimum input font stroke of in each stroke of standard font, adjusting the distance, just import the stroke change of font and carry out respective handling (the clear 54-061146 of Japanese patent application), or information such as a preface variation are pre-deposited in the standard font, according to depositing content in, the corresponding name of restriction input font and standard font stroke is to alleviate the treatment capacity (Jap.P. discloses clear 57-178579) that correspondence is named.

For example, Figure 26 is illustrated in the block scheme that Jap.P. discloses the formation of the existing online character recognition device of a disclosed absorption preface change among the clear 57-178579.

Among Figure 26,1 for calculating the font coalignment of the distance of importing font stroke and standard font stroke, 2 standard font memory storages for the order of strokes of the standard font of storing each literal in advance, 3 are the begin to write preface combination memory storage of preface combined method etc. of storage, 4 for the standard font that is stored in the order of strokes of the standard font in the standard font memory storage 2 according to an information-change of preface combination memory storage 3 forms device again, and 5 for determining the recognition device of recognition result according to the distance of font coalignment 1 output.

Below, the action of the online character recognition device of the existing example that uses Figure 26 is described.

At first, in font coalignment 1, import font information X input.Input font information X is made of n bar stroke when the literal that n draws, and is X1 if set a preface of each stroke, X2 ... Xn then is expressed as in polynary vector space

X＝(X1，X2，…Xn)。

In addition, be SK (K represents the systematic name of standard font) if establish the standard font of drawing formation that is stored in the standard font memory storage 2 with n, the order of strokes of the various strokes of SK is SK1, SK2 ... SKn, then standard font SK is expressed as in polynary vector space

SK＝(SK1，SK2，……SKn)

Then, font coalignment 1 basis of calculation font SK and the distance of importing font X, but because input font X not necessarily charges to a correct preface, so just will import each key element and the corresponding name of standard font key element order of font X simply, can not correctly discern the literal of using the preface different to charge to a preface of standard font.

In addition, the standard font forms device 4 bases again and is stored in a preface combined information of preface combination memory storage 3, a conversion preface formation standard font SKm (m=1,2 ..., p), batch transportation is in font coalignment 1.Here, the combination sum of a preface change of this standard font of p.

Then, 1 pair of font coalignment forms the standard font that device 4 is sent here successively again by the standard font, utilizes following formula to obtain itself and the distance P xkm of input font X.

Then, identify minor increment in P the distance that 5 pairs of recognition devices are sent here by font coalignment 1, with the distance P xk of this minor increment as input font X and standard font Sk.Have the standard font that n draws in all classifications in the standard font storage device and carry out above processing being stored in, at last with minor increment classification export as recognition result.

Prior art example 2

Drawing in the existing online character recognition device of counting change identification literal by drawing, there be a plurality of unique points and the stroke feature point of standard font with the stroke of input font to carry out the DP coupling, carry out corresponding name that feature narrows so that the formation minor increment is named (the clear 62-62394 of patent disclosure), perhaps, in order to improve the accuracy of identification of stroke number change font, in the standard font (the clear 60-79485 of patent disclosure, the clear 60-136783 of patent disclosure) or the like recorded and narrated in the stroke method of attachment.

For example, Figure 28 is the formation block diagram of original online character recognition device of drawing of the absorption stroke number change that illustrates in the clear 62-62394 of patent disclosure.Figure 27 is the recognition sequence that is used for illustrating original online character recognition device of drawing the change of stroke number.

Figure 27 (a) expression standard font, Figure 27 (b) expression input font.Figure 27 (c) is the corresponding name of each stroke with each stroke of input font of description standard font (D).

Among Figure 27, the stroke of 30～39 expression input fonts, the stroke of 10～21 expression standard fonts.

And in Figure 28,40 is that stroke is counted test section, the 41st, and the font address generating unit, 42 is character library, 43 is the feature extraction unit, the 44th, the DP matching part, 46 is detection unit, the 47th, the scope specifying part.

Below, describe with Figure 27,28 pairs of actions of drawing the original online character recognition device of drawing the number change.

At first, will import the font input word and draw several test sections 40 and feature extraction unit 43, extract the unique point of drawing number and each stroke out.Count test section 40 with stroke, the input word type-word is drawn several n relatively, the stroke of reply photograph is counted scope (n+ α)～(n-β) be defeated by character library 42.Under the situation of " being " word shown in Figure 27 (b), because n=5, so establish α=2, the stroke scope of counting of β=1 is 7 strokes～4 strokes.

The such input font of Figure 27 (b) is 5 strokes, form 10 unique points, and the standard font is 6 strokes, forms 12 unique points.But, if from the beginning carry out the correspondence name simply, then left over the characteristic number of as Figure 27 (c), not carrying out corresponding name, owing to do not carry out the corresponding name of stroke correctly, distance values becomes big as a result.

Therefore, in this prior art example, adopted DP (dynamic programming) coupling, to drawing different the carrying out correspondence by the distance minimum and name of number.Here, the following concrete DP coupling of carrying out.

Each unique point of input font is expressed as follows:

A＝a1，a2，……，ai，…aI

And the unique point of the standard font in marking-up storehouse is expressed as follows:

B＝b1，b2，…，bj，…，bJ

At this moment, ai, bj (i j) can followingly obtain apart from d.If d (i, j)=| ai-bj| is (1) i=1 then, during j=1,

G (1,1)=d (1,1) (2) i=1, j ≠ 1 o'clock

G (1, j)=g (1, j-1)+(1, j) (3) i ≠ 1 is during j=1 for d

G (i, 1)=g (i-1,1)+other situation of d (i, 1) (4)

By above-mentioned recursion formula obtain g (I, J) (starting condition), obtain with following formula the feature point range of A and B the feature point range apart from S (A, B).Then, with its apart from reckling as recognition result.

S(A，B)＝1/(I+J)*g(I，J)

The standard font that scope is counted with above-mentioned stroke in DP matching part 44 is read from character library 42 successively, obtain with above-mentioned DP matching process input unique point of font and standard font unique point distance and be transported to detection unit 46.Detection unit 46 is exported the literal apart from reckling in the above-mentioned distance as recognition result.

In this example, even the such input font of Figure 27 (b) is correctly carried out the correspondence name of unique point as Figure 27 (d), the distance of naming than the correspondence of Figure 27 (c) is little.

As above explanation with regard to standard font all combination calculation distances with the input font, is drawn the such online character recognition device of a prior art example of preface change 1, be to all combination calculation distance, and calculated amount is big, carries out Real time identification and has any problem.

In addition, an ordered pair is answered the such online character recognition device of prior art example 1 of naming method in the record standard font, can only handle a preface change of preparing in advance in the standard font, will set many prefaces accordingly in order to improve corresponding name precision, and have the character library of making to become big problem.

In addition, in the device of the correspondence of original technology example 1 a preface change, can corresponding preface change and can not correspondingly draw the number change, there be asking that the accuracy of identification that makes the word that connects together etc. descends to be.

Equally, in the such online character recognition device of prior art example that the correspondence of drawing the different fonts of number with DP coupling is named 2, draw and import the distance that font m draws in order to obtain 1 standard font n, must carry out the distance calculation of n * m, calculate like this all standard fonts of drawing the number scope of (m+d)～(n-β).Here, if consider the occasion of the literal of 11 strokes of inputs as input characters, then m=11 for example sets α=1, and distance calculation then must be carried out to the 13 standard fonts of drawing with regard to 10 strokes in β=1.Even be limited to the occasion in JIS the 1st level literal, 10 strokes of distance calculation that have 289 words will do 289 * 10 * 11=31790 time, 11 strokes of distance calculation that have 300 words will do 300 * 11 * 11=36300 time, 12 strokes of distance calculation that have 297 words will do 297 * * 12 * 11=39204 time, 13 strokes of distance calculation that have 242 words will do 242 * 13 * 11=34606 time.For this reason, total will be made 141900 times distance calculation, and calculated amount is very many, and Real time identification is had any problem.

And the such online character recognition device of prior art example 2 that has information such as associated methods in character library will have the method for attachment information of each each stroke of literal, forms character library and has any problem.And character library capacity ground must be very big.And, because the continuous word that can not corresponding be not ready in advance just has the problem that accuracy of identification reduces.

In addition, the correspondence of prior art example 2 is drawn the character recognition device of counting change, can only change and can not a corresponding preface change by a correspondence stroke number, and the problem of the literal accuracy of identification reduction that a preface is morphed is arranged,

And, utilizing the stroke information of literal to discern energetically with the character recognition device of above-mentioned prior art example 1,2, the literal that changes for some gointed hands or a preface is though literal global shape beauty reads but be difficult to high precision.

No matter the quality of literal is in order to discern processing, even the occasion of writing on smaller ground, also need and pollute the identical processing time of literal write, for accuracy of identification is improved,, also has the long problem of recognition time for all literal of wanting intricately to handle.

The object of the invention is to solving relevant issues, obtains a kind of high precision identification, the online character recognition method that may discern at a high speed again and online character recognition device both realized.

Specifically, should name the calculated amount of identification, carry out high speed, high-precision the 1st purpose that is identified as by reducing an ordered pair.

In addition, by reducing the calculated amount of the corresponding name identification of stroke number, carry out high speed, high-precision the 2nd purpose that is identified as.

Also have, outside decapacitation is discerned at a high speed, can also and draw the number change and carry out respective handling a preface change simultaneously, carry out high-precision identification, this is the 3rd purpose.

The 4th purpose is, even a preface of the writing words occasion different with a preface of grapholect also can be carried out the good identification of precision.

As the 5th purpose, be to change corresponding to soft preface about the high literal of correct interpretation possibility, a literal preface of carrying out within reason that may be lower about correct interpretation changes, and can both discern with high-speed, high precision ground.

The 6th is in the good occasion of recognition result of font identification recognition result to be exported at a high speed.

In addition, owing to can not realize that actual writing this is the 7th purpose to dealing with name, will preventing that discrimination from becoming bad.

For realizing above purpose, the online character recognition method of first invention includes:

Follow the action of writing of literal to import the written information input step of this written information;

From the feature that above-mentioned written information is extracted the global shape of the above-mentioned institute of expression writing words out, select a plurality of font identification steps that are similar to the grapholect of above-mentioned institute writing words from character library according to the feature of the above-mentioned global shape of expression;

Extract the first stroke feature that expression constitutes the stroke feature of above-mentioned institute writing words out from above-mentioned written information, the corresponding name of the stroke of selected grapholect in a preface of considering above-mentioned institute writing words is carried out above-mentioned institute writing words according to above-mentioned first stroke feature stroke and the above-mentioned font identification step, calculate an ordered pair of the similar degree of the above-mentioned relatively writing words of selecting in the above-mentioned font identification step of a plurality of grapholects and should name identification step, and

Should name the said similar degree of calculating in the identification step according to above-mentioned ordered pair, export the first output step of selected grapholect in the above-mentioned font identification step.

Here, the written information input step is equivalent to the step (Fig. 2 has S1) that literal is write label etc. of aftermentioned embodiment.The first stroke feature then is equivalent to the direction from the origin-to-destination of word stroke in the embodiment of back, the boundary rectangle of stroke wide, height etc.The similar degree to above-mentioned writing words of a plurality of grapholects of selecting in the above-mentioned font identification step is equivalent to the grapholect of back embodiment and the distance of writing words etc.

The online character recognition method of second invention includes:

Follow the written information input step of this written information of character calligraph action input;

From the feature that above-mentioned written information is extracted the above-mentioned institute of expression writing words global shape out, select a plurality of font identification steps that are similar to the grapholect of above-mentioned writing words from character library according to the feature of the above-mentioned global shape of expression;

Extract second stroke feature that expression constitutes the stroke feature of above-mentioned writing words out from above-mentioned written information, according to above-mentioned second stroke feature, consider that above-mentioned writing words draws the corresponding name that number carries out the stroke of selected grapholect in above-mentioned writing words stroke and the above-mentioned font identification step, calculate the number correspondence of drawing of the similar degree of the above-mentioned relatively writing words of selecting in the above-mentioned font identification step of a plurality of grapholects and name identification step, and

According to the above-mentioned similar degree of calculating in the corresponding name of the above-mentioned stroke of number identification step, export the 2nd output step of the grapholect of selecting in the above-mentioned font identification step.

Wherein, second stroke feature is equivalent to the stroke width in the aftermentioned embodiment, height etc.

Fa Ming online character recognition method thirdly includes:

Follow the written information input step of its written information of character calligraph action input;

From the feature that above-mentioned written information is extracted the global shape of the above-mentioned writing words of expression out, select a plurality of font identification steps that are similar to the grapholect of above-mentioned writing words from character library according to the feature of the above-mentioned integral body of expression;

Extract the feature that expression constitutes the first stroke of above-mentioned writing words stroke feature out from above-mentioned written information, according to above-mentioned first stroke feature, consider the corresponding name of the stroke of the grapholect selected in stroke that an above-mentioned writing words preface carries out above-mentioned writing words and the above-mentioned font identification step, calculate an ordered pair of the similar degree of the above-mentioned relatively writing words of a plurality of grapholects that above-mentioned font identification step selects and to name identification step;

By answering similar degree and the predetermined value calculated in the identification step to compare above-mentioned ordered pair, differentiate first discriminating step of whether exporting above-mentioned grapholect;

When in above-mentioned first discriminating step, differentiating for output, the first output step of the grapholect that should name the above-mentioned similar degree of calculating in the identification step to export to select in the above-mentioned font identification step according to above-mentioned ordered pair;

In above-mentioned first discriminating step, differentiate when not exporting, extract second stroke feature that expression constitutes the stroke feature of above-mentioned writing words out from above-mentioned written information, consider that according to above-mentioned second stroke feature above-mentioned writing words draws the corresponding name that number carries out the relative stroke of grapholect selected in above-mentioned writing words stroke and the above-mentioned font identification step, calculate the number correspondence of drawing of similar degree of the above-mentioned writing words of a plurality of grapholects of selecting in the above-mentioned font identification step and name identification step; And

According to the above-mentioned similar degree of calculating in the corresponding name of the above-mentioned stroke of number identification step, the second output step of the grapholect that output is selected in above-mentioned font identification step.

Wherein, first discriminating step is equivalent to the judgement that no-output is arranged by the preferential output mechanism 55 of a preface (S4 of Fig. 2) of the embodiment of back.

The online character recognition method of the 4th invention, be first or above-mentioned the ordered pair of online character recognition method of thirdly invention should name in the identification step, carry out the corresponding name of the grapholect stroke selected in the stroke of above-mentioned writing words and above-mentioned font identification step by a preface of above-mentioned writing words according to above-mentioned second stroke feature, the corresponding name of change order of strokes carries out once more corresponding name in the time of can not carrying out above-mentioned corresponding name.

The online character recognition method of the 5th invention is in the 4th the online character recognition method of invention institute, calculate the above-mentioned similar degree of selecting grapholect of above-mentioned relatively writing words with above-mentioned font identification step, should name identification step with above-mentioned ordered pair, repeatedly change the correspondence name of above-mentioned order of strokes according to the similar degree of the above-mentioned grapholect of calculating in the above-mentioned font identification step.

The 6th the invention online character recognition method be thirdly the invention online character recognition method in, also have after above-mentioned font identification step, similar degree and predetermined value by the grapholect of calculating in will the above-mentioned font identification step to above-mentioned writing words compare, and judge whether to export the 2nd discriminating step of above-mentioned grapholect; And, when in above-mentioned second declares suddenly, differentiating,, export the 3rd output step of the grapholect of selecting in the above-mentioned font identification step according to the above-mentioned similar degree of calculating in the above-mentioned font identification step for output,

In above-mentioned second discriminating step, differentiate when not exporting, carry out above-mentioned ordered pair and should name identification step.

Wherein, second discriminating step is equivalent to the judgement (S3 of Figure 18) that the output with the preferential output mechanism 110 of font of back embodiment has or not.

In addition, the 7th the online character recognition method of invention is in the online character recognition method of second invention,

The above-mentioned stroke of corresponding name of number identification step calculated step by DP coupling step and similar degree and constituted,

Above-mentioned DP coupling step is extracted second stroke feature of the feature of the stroke of representing the above-mentioned writing words of formation out from above-mentioned written information, utilize DP coupling to carry out the corresponding name of above-mentioned writing words stroke and the grapholect stroke of selecting by above-mentioned font identification step based on above-mentioned second stroke feature

Above-mentioned similar degree is calculated step and extract the stroke of above-mentioned writing words out and the stroke of above-mentioned grapholect is stablized stroke one to one through the stroke of correspondence name from above-mentioned DP coupling step, the aforementioned stable stroke is calculated similar degree with respect to the above-mentioned writing words of the grapholect of selecting in the above-mentioned font identification step

The grapholect that the above-mentioned second output step is selected in above-mentioned font identification step according to the above-mentioned similar degree output of calculating in the above-mentioned similar degree calculation procedure.

Wherein, similar degree is calculated the detection (S51 of Figure 22) that usefulness that step is equivalent to the back embodiment is stablized stroke testing agency 132.

The online character recognition device of the 8th invention comprises:

Follow the written information input block of this written information of character calligraph action input;

By the feature that above-mentioned written information is extracted the above-mentioned writing words global shape of expression out, select a plurality of font recognition units that are similar to the grapholect of above-mentioned writing words from character library according to the feature of above-mentioned expression global shape;

Extract the first stroke feature of feature that expression constitutes the stroke of above-mentioned writing words out by above-mentioned written information, consider that according to the first stroke feature an above-mentioned writing words preface carries out the corresponding name of above-mentioned writing words stroke and the stroke of the grapholect of being selected by above-mentioned font recognition unit, calculate an ordered pair of the similar degree of the above-mentioned relatively writing words of selecting by above-mentioned font recognition unit of a plurality of grapholects and should name recognition unit; And

Should make the above-mentioned similar degree of calculating in the recognition unit export first output unit of the grapholect of in above-mentioned font recognition unit, selecting according to above-mentioned ordered pair.

Wherein, the written information input block is equivalent to the input block 50 of back embodiment.And first output unit is equivalent to the display unit 58 among the embodiment of back.

The online character recognition device of the 9th invention comprises:

Follow the written information input block of its written information of character calligraph action input;

By the feature that above-mentioned written information is extracted the above-mentioned writing words global shape of expression out, select a plurality of font recognition units that are similar to the grapholect of above-mentioned writing words from character library according to the feature of the above-mentioned global shape of expression;

Extract the feature that expression constitutes second stroke of above-mentioned writing words stroke feature out by above-mentioned written information, consider that according to above-mentioned second stroke feature above-mentioned writing words draws the corresponding name of the stroke of the grapholect of selecting in stroke that number carries out above-mentioned writing words and the above-mentioned font recognition unit, calculate the number correspondence of drawing of the similar degree of the above-mentioned relatively writing words of selecting in the above-mentioned font recognition unit of a plurality of grapholects and name recognition unit; And

Export second output unit of the grapholect of in above-mentioned font recognition unit, selecting according to the above-mentioned similar degree of calculating in the corresponding name of the above-mentioned stroke of number recognition unit.

Wherein, second output unit is equivalent to the display unit 58 among the embodiment of back.

The online character recognition device of the 10th invention comprises:

From the feature that above-mentioned written information is extracted the above-mentioned writing words global shape of expression out, select a plurality of font recognition units that are similar to above-mentioned writing words from character library according to the feature of above-mentioned expression global shape;

Extract the first stroke feature of the feature of the stroke of representing the above-mentioned writing words of formation out from above-mentioned written information, consider an above-mentioned writing words preface according to above-mentioned first stroke feature, carry out the corresponding name of the stroke of the grapholect selected in above-mentioned writing words stroke and the above-mentioned font recognition unit, calculate an ordered pair of the similar degree of the above-mentioned relatively writing words of selecting in the above-mentioned font recognition unit of a plurality of grapholects and should name recognition unit;

By naming similar degree and the predetermined value calculated in the recognition unit to compare above-mentioned ordered pair, differentiate first judgement unit of whether exporting above-mentioned grapholect;

When in above-mentioned first judgement unit, differentiating, should name the above-mentioned similar degree of calculating in the recognition unit to export first output unit of the grapholect of in above-mentioned font recognition unit, selecting according to above-mentioned ordered pair for output; With

In above-mentioned first judgement unit, differentiate when not exporting, extract second stroke feature that expression constitutes the stroke feature of above-mentioned writing words out from above-mentioned written information, consider that according to above-mentioned second stroke feature above-mentioned writing words draws number carries out above-mentioned writing words stroke and the grapholect stroke of selecting in above-mentioned font recognition unit corresponding name, calculate the number correspondence of drawing of the similar degree of the above-mentioned relatively writing words of selecting in the above-mentioned font recognition unit of a plurality of grapholects and name recognition unit; And

According to the above-mentioned similar degree that the corresponding name of above-mentioned stroke of number recognition unit is calculated, export second output unit of the grapholect of selecting in the above-mentioned font recognition unit.

Wherein, first judgement unit is equivalent to the preferential output unit 55 of a preface among the embodiment of back.

In addition, the online literal identification of the 11st invention is in the online character recognition device of the 9th invention,

The above-mentioned corresponding name of the number unit of drawing is made of DP matching unit and similar degree computing unit;

Above-mentioned DP matching unit is extracted second stroke feature that expression constitutes the stroke feature of above-mentioned writing words out from above-mentioned written information, by the DP coupling grapholect stroke of selecting in above-mentioned writing words stroke and the above-mentioned font recognition unit is carried out corresponding name according to above-mentioned second stroke feature;

Above-mentioned similar degree computing unit is the stroke of process through naming from above-mentioned DP matching unit, extract the stroke of above-mentioned writing words and the stroke of above-mentioned grapholect out and stablize stroke one to one, calculate the similar degree of the above-mentioned relatively writing words of selecting in the above-mentioned font recognition unit of grapholect with regard to the aforementioned stable stroke;

Above-mentioned second output unit is calculated the above-mentioned similar degree of calculating in the unit according to above-mentioned similar degree, exports the grapholect of selecting in the above-mentioned font recognition unit.

Wherein, similar degree is calculated the unit and is equivalent to stable stroke detecting unit 132 among the embodiment of back.

The online character recognition device of the 12nd invention is in the online character recognition device of the 10th invention, includes:

Similar degree and predetermined value by the grapholect of calculating in the above-mentioned font recognition unit to above-mentioned writing words compare, and differentiate second judgement unit of whether exporting above-mentioned grapholect;

When in above-mentioned second judgement unit, differentiating, export the 3rd output unit of the grapholect of in above-mentioned font recognition unit, selecting according to the above-mentioned similar degree of calculating in the above-mentioned font recognition unit for output;

Above-mentioned ordered pair should be named recognition unit, differentiates in second judging unit and do not handle when not exporting.

Below, with reference to description of drawings working of an invention mode.In the accompanying drawing:

Fig. 1 is the pie graph of the online character recognition device of expression embodiment 1;

Fig. 2 is the process flow diagram of control module 59 actions of expression embodiment 1;

Fig. 3 is the figure that is illustrated in the input font of writing in the input block 50 of embodiment 1;

Fig. 4 represents the process flow diagram of font recognition unit 52 actions of embodiment 1;

Fig. 5 is the figure of kind of assign direction code of the font recognition unit 52 of expression embodiment 1;

Fig. 6 is the input font figure after the travel direction assignment of code of expression embodiment 1 is handled;

Fig. 7 is the direction code Region Segmentation of embodiment 1, the figure of the code of each direction counting in each zone;

Fig. 8 is the process flow diagram that an ordered pair of expression embodiment 1 should be named recognition unit 54 actions;

Fig. 9 is the figure of the direction of an ordered pair of expression embodiment 1 should name recognition unit 54 to distribute stroke direction, imaginary stroke direction the time;

Figure 10 should name recognition unit 54 to represent to import the figure of stroke feature of the first stroke of font with an ordered pair of embodiment 1;

Figure 11 represents to carry out the figure of the corresponding rule of naming of stroke for should name recognition unit 54 with an ordered pair of embodiment 1;

The figure that Figure 12 arranges for the stroke of the standard font of the literal " " in the stroke feature character library of expression embodiment 1;

Figure 13 is the literal " " of expression embodiment 1 figure with the input font charged to of a difference preface;

Figure 14 is the figure of process of the preface change of an ordered pair of the expression embodiment 1 input font that should name recognition unit 54;

Figure 15 is the figure for input font that the gointed hand of drawing 57 actions of the corresponding name of number recognition unit is described of expression embodiment 1;

Figure 16 is the standard font of drawing the corresponding name of number recognition unit 57 of expression embodiment 1 and the corresponding name result's of input font figure;

Figure 17 is the pie graph of embodiment 2 online character recognition devices;

Figure 18 is the process flow diagram of the motion flow of the control module 59 of expression embodiment 2;

Figure 19 represents the pie graph of the online character recognition device of embodiment 3;

Figure 20 is the process flow diagram of the motion flow of the control module 59 of explanation embodiment 3;

Figure 21 is the pie graph of the online character recognition device of embodiment 4;

Figure 22 is the process flow diagram of control module 59 motion flows of explanation embodiment 4;

Figure 23 becomes the figure in the path of minimum with the distance of the DP matching result of standard font for the input font of the DP matching unit 130 of explanation embodiment 4;

Figure 24 writes the figure in the path can not carry out corresponding the name time for input font and the standard font of the DP matching unit 130 of explanation embodiment 4 actual;

Figure 25 is the figure of expression with the correspondence name of the stable stroke detecting unit 132 detected stable strokes of embodiment 4;

Figure 26 is the pie graph of the online character recognition device of the original technology example 1 of expression;

Figure 27 is input font, standard font and its corresponding name result's of the original technology example 2 of expression figure;

Figure 28 is the pie graph of the online character recognition device of the original technology example 2 of expression;

Embodiment 1

The online character recognition device of present embodiment is to extract standby literal out by font identification, should standby literal carries out an ordered pair and should name identification, draws the corresponding name of number again and discerns, and below describes with reference to Fig. 1～16.

Fig. 1 is the pie graph of the online character recognition device of present embodiment.

Among Fig. 1,50 input blocks for the written information of input label etc., 51 do not rely on the font character library of overall shape feature of the literal of drawing number, a preface of each literal for storage in advance.

The 52nd, by with extract out from the written information of aforementioned input block 50 gained do not rely on draw the overall shape feature that number, a preface do not have aforementioned font character library 51 relatively, the font recognition unit of discerning.What is called does not rely on the overall shape feature of the literal of drawing number, a preface and holds the time aftermentioned.

The 53rd, storing drawing of each literal in advance and counting, a preface, and the stroke feature character library of the stroke feature of each stroke of formation literal.

In this manual, so-called stroke is exactly the various strokes that constitute a certain literal.

The 54th, extract the stroke feature of input out from the written information of aforementioned input block 50 gained, the corresponding ordered pair of discerning of naming of feature that is stored in each literal in the aforementioned stroke feature character library 53 by a consideration ordered pair should be named recognition unit.

The 55th, should name recognition unit 54 gained recognition result information according to aforementioned ordered pair, determine whether the preferential output unit of a preface as final recognition result output with the aforementioned identification result.

The 57th, consider that stroke several stroke features to the input font carry out drawing of corresponding name identification with the stroke of the standard font of aforementioned stroke feature character library 53 and count corresponding name recognition unit.

The 58th, the show pen ordered pair should be named the recognition result of recognition unit 54 or draw the display unit of the recognition result of the corresponding name of number recognition unit 57.

The 59th, the control module of the processing action of each unit 50～57 in the control character recognition device, it is connected respectively to each unit 50～57.

Below, with Fig. 2 the action of the online character recognition device of present embodiment is described.

Fig. 2 is the process flow diagram of processing procedure of control module 59 of the online character recognition device of expression present embodiment.

As description of the present embodiment, to input characters The time the literal recognizer describe.

At first, at step S1, control module 59 instruction input blocks 50 obtain the input font.The input font is the input process information that includes along with the action that is written to input block 50.This information is written information.

Fig. 3 is the explanation input block 50 input font of writing

Figure, 60～66th, the order of strokes of in chronological sequence importing.The input sequence that 60～66 order represents to write stroke is an i.e. preface also.Write with seven strokes in this example, so be to constitute by 7 strokes.

Therefore, in S1, obtain 7 stroke 6s 0～66 as written information.

Then enter step S2, control module 59 sends instruction, and the order of strokes of input font is sent into font recognition unit 52, carries out font identification.

In this example, the stroke 6 0～66 of the input font of Fig. 3 is sent into font recognition unit 52, font recognition unit 52 is extracted out and is drawn number, the irrelevant literal overall shape feature of a preface from the stroke 6 0～66 of input font, discerns.

With the elaborate process of the step S2 that the identification of font recognition unit 52 handles of Fig. 4.

Fig. 4 is the process flow diagram of the processing procedure of expression font recognition unit 52.

At first, at the step S11 of Fig. 4, each stroke that font recognition unit 52 is just imported font with 4 kinds of direction assignment of code on the each point of multiple sampling.

4 kinds of direction codes are 4 kinds of direction codes that are shown in Fig. 5, vertical (V) 70 are arranged particularly, level (H) 71, tiltedly upper right (R) 72, the direction code of tiltedly upper left (L) 73.

To be that the center is regarded same direction as up to the scope of dotted line with the direction of arrow in all directions.

With regard to a stroke multidraw, each point is assigned to any of aforementioned 4 kinds of direction codes.

But common sampled point is not because writing speed deviation and sampling rate have certain interval, and point is not necessarily compact arranged.

Thereby, if the direction assignment of code is arrived its original each point, even the literal of identical font also can produce deviation because of writing speed makes direction code number.In addition, because of the literal size also produces the difference of counting, therefore, make the suitable normalization of size and between point, carry out interpolation to make the normalization of counting.

Just obtain from present to next total direction, whether be equivalent to the either direction of Fig. 5 by judging this direction through the each point of normalization, with the direction assignment of code on each point.Here, because the terminal point of stroke does not exist down a bit, distribute and the same direction of preceding point.Import the normalization of each stroke 6 0～66 size of font, interpolation is handled, with direction assignment of code Fig. 6 that the results are shown in to each point.

Fig. 6 (a) is the result who has distributed horizontal direction code (H), and Fig. 6 (b) is the result who has distributed vertical direction code (H).The font that this is routine, because of there not being tiltedly upper right (R), the direction of tiltedly upper left (L) has been omitted these directions.

Then, enter the step S12 of Fig. 4, whole word is divided into 4 * 4 zones, each every kind of direction code in zone is counted.

Fig. 7 calculates the diagram of the direction code number in each zone for whole literal is divided into 4 * 4 zones.

Fig. 7 (c) is that to carry out result calculated, Fig. 7 (d) be that vertical direction code to Fig. 6 (b) carries out result calculated to the horizontal direction code to Fig. 6 (a).

Therefore, the direction code characteristic F that has extracted 4 directions out from the input font is expressed as 4 * 4 16 n dimensional vector ns along all directions.

F＝(FH，FV，FR，FL)

FH＝FH1，FH2，…，FH16

FV＝FV1，FV2，…，FV16

FR＝FR1，FR2，…，FR16

FL＝FL1，FL2，…，FL16

FH1, FH2 ..., FH16 represents the calculating number of direction code in 4 * 4 the zone.To FV, FR, FL also have same implication.

Then enter the S13 of Fig. 4, obtain the distance D of feature of direction code distribution characteristics and input font of the standard font of each word that pre-deposits the font character library.

The distribution characteristics of the direction code of standard font can be done following expression by S:

S＝(SH，SV，SR，SL)

SH＝SH1，SH2，……，SH16

SV＝SV1，SV2，……，SV16

SR＝SR1，SR2，……，SR16

SL＝SL1，SL2，……，SL16

Here, the distance of input font and standard font can be expressed as follows.

D＝(DH，DV，DR，DL)

DH＝|FH1-SH1|+|FH2-SH2|+……+|FH16-SH16|

DV＝|FV1-SV1|+|FV2-SV2|+……+|FV16-SV16|

DR＝|FR1-SR1|+|FR2-SR2|+……+|FR16-SR16|

DL＝|FL1-SL1|+|FL2-SL2|+……+|FL16-SL16|

By following formula, obtain the distance of standard font and input font in the font character library, export 50 candidate characters successively from distance is little as recognition result.

More than, the processing of the step S11 of Fig. 4～S13 is not rely on the literal global shape Feature Recognition of drawing number, a preface to handle.

Select after 50 candidate characters, move to the step S3 of Fig. 2.

Then enter the S2 of Fig. 2, control module 59 sends instruction and will deliver to an ordered pair by 50 candidate characters of font recognition unit 52 gained and should name recognition unit 54, should name identification so that carry out an ordered pair.

Should name the action of recognition unit 54 to be elaborated with Fig. 8 to an ordered pair.

Fig. 8 is the action flow chart that an explanation ordered pair should be named recognition unit 54.

At first, an ordered pair should name recognition unit 54 to enter the step S20 of Fig. 8, extracts the stroke feature of each stroke out from the input font.

Extract each stroke shape facility of each stroke out as stroke feature, stroke from initial point to the terminal point direction, the width of stroke boundary rectangle, stroke boundary rectangle height, the direction of imagination stroke, imaginary stroke boundary rectangle width, totally 7 kinds of imaginary stroke boundary rectangle height.

The boundary rectangle of stroke is exactly to carry out external rectangle with respect to stroke.

The imagination stroke is exactly to connect the terminal point of existing stroke and the straight line of next stroke initial point.When final stroke, next record divides initial stroke into.The boundary rectangle of imagination stroke is exactly as cornerwise rectangle with imaginary stroke.

And the direction of each stroke is with 8 directional quantizations shown in Figure 9.In addition, for width, highly be with the boundary rectangle of literal integral body wide, be made as 100 normalization value highly separately.Figure 10 represents just to import first stroke of 66 example of extracting 7 kinds of stroke features out of font.

Then, standard font stroke in the stroke feature character library of the stroke of input font and 50 candidate characters that obtained by font recognition unit 52 is compared, because should name recognition unit only to carry out the correspondence name of a preface at an ordered pair, so stroke number of drawing number and standard font as if the input font is inconsistent, just can not corresponding name.

At this,, so contain the literal of various stroke of number in the candidate character because in font recognition unit 52, do not have to utilize the information of drawing number.Thereby, at first import drawing of font and standard font and count relatively at S21.Draw number when inconsistent, S21 be " No ", does not then carry out and show the corresponding name of standard font of candidate character, and carries out the corresponding name with next literal.In the occasion of S21 unanimity, " YES " enter S22.

At S22, carry out corresponding name with the stroke feature of input font with the stroke feature of the standard alphabet of stroke feature character library 53.Here, the standard font be as shown in Figure 12 a correct preface or the stroke feature of majority's lettering pen preface.In addition, the stroke judgement that whether correspondence is named is each other carried out correspondence name calibrating with reference to corresponding naming rule described later.

Figure 11 illustrates the example of corresponding naming rule.Record and narrate to have in the example various stroke shapes are detected fixed feature, import the calibrating of font and standard font with regard to its feature.For example, draw in the calibrating of correspondence name each other at mould, only the stroke width of the key character of drawing with regard to the mould metal marker is examined and determine.

Here, with regard to the calibrating of direction, the difference of direction for+-2 be judged to be not corresponding name when above.

Wide height to the boundary rectangle of literal integral body respectively is 100, and wide, high difference just is judged to be not corresponding name 70 when above to wide, high calibrating.But benchmark is different, and the stroke of the weak point occasion of corresponding name is each other compared the direction instability with the occasion of the mutual corresponding name of stroke of length, therefore will enlarge the tolerance band of corresponding name.

In addition, for the direction of imaginary stroke, width, highly, closing because of position between the expression stroke is yardstick, so do not examine and determine with regard to the stroke shape.Here, above-mentioned correspondence name is carried out when correspondence is named successfully, during the correspondence name of the corresponding kaput occasion of name or all strokes, enters S23.

Follow at S23, within the stroke of input font, whether successful with regard to all strokes inspection correspondence names, all to entering S26 at once, middle correspondence then enters S24 when naming failure.In this example, because of the input font is to write with a correct preface, all corresponding name of all strokes just enters S26.

Then at S26, calculate the distance of stroke feature with the stroke feature of standard font of input fonts with regard to all strokes, after each feature normalization, to finally obtaining behind each feature distance weighted and the existing distance of standard font.Weighting can preestablish with the raising discrimination, or by change weights such as stroke numbers.

But, when distance calculation, do not use the stroke shape.

Here, the distance D that the distance D vd after distance D vh, the false stroke direction normalization tried to achieve and calibrated font after the distance D d after the stroke direction normalization, the distance D w after the stroke width normalization, the distance D n after the normalization of stroke height, the distance D vw after the imaginary stroke width normalization, the normalization of imaginary stroke height.

D=(Wd * Dd+Ww * Dw+Wn * Dh+Wvw * Dvw+Wvh * Dvh+Wvd * Dvd) here, Wd represents that stroke direction weight, Ww represent the stroke width weight, Wh is a stroke height weight, Wvw is imaginary stroke width weight, Wvh is imaginary stroke height weight, and Wvd is imaginary stroke direction weight.These weightings have following relation.

Wd+Ww+Wh+Wvw+Wvh+Wvd＝1

This weight is predefined for improving discrimination.In addition, this weighting also can be drawn number and comes change setting according to discerning literal.

Below, the action under the corresponding name failure scenarios in S23 is described.If the literal that Figure 13 is such Write by a preface different with the standard font.At this moment, import second stroke 91 of font and second stroke 81 of standard font and be all perpendicular drawing, judgement not corresponding name in the calibrating of the height of stroke.Therefore, name the stage that is through with, enter S24 in correspondence to second stroke.

Then at S24, judge a preface difference of input font, the stroke of the kaput input font of just corresponding name, by with the stroke of the input font that in corresponding name, does not use successively (if will carry out the corresponding name of drawing with the K that shows the standard font, then in the name of correspondence that the K-1 to the standard font draws in the used input stroke of the institute stroke in addition, carry out the correspondence name that K draws) transposing, carry out a preface and arrange replacing.

But, only when the final not corresponding name of stroke, or when not having untapped stroke with regard to existing stroke of just carrying out the standard font of corresponding name, could judge that correspondence name so far has inerrancy, equally, have only the correspondence of last stroke is named as not corresponding name, could carry out the correspondence name untapped input stroke.

In this example, as Figure 14 (a), with the stroke of correspondence not being named 91 and next stroke 92 transposing.

Then enter S25, the double calculation number of times (a preface transposing number of times) of so far input stroke and the predetermined double calculation number of times upper limit are compared, become " No " when not running off, get back to S22, never the stroke of the standard font of corresponding name begins corresponding name once more.

The input stroke 92 of lining up at S22, with second stroke stroke, the 81 corresponding names of standard font, with last time same, the corresponding name failure of second stroke stroke of standard font does not enter S24.

At S24, next untapped input stroke 93 is just in time at second stroke, and the

input stroke

91,92 after second stroke correspondence name failure of standard font turns back to a preface of original input font.

Input stroke after Figure 14 (b) expression is moved is arranged.

Above-mentioned repeatedly processing is shown in Figure 14 (c)～(f) until the transposing process of a preface of last all corresponding name of stroke.In this example, through 6 corresponding names successfully of retrying, with usefulness correctly the preface occasion of writing together enter S26, carry out calculating with standard font distance.In addition, along with the transposing of above-mentioned preface, in the input stroke that has changed a preface, calculate the direction, wide, high of the imaginary stroke of position relation between the expression stroke again.

After just having finished above-mentioned corresponding name, enter the S4 of Fig. 2 by whole identification candidate characters of font recognition unit 52 gained.

At S4, control module 59 will should name the distance of the literal of minor increment in the corresponding literal of having named successfully of recognition unit 54 to send into the preferential output unit 55 of a preface at an ordered pair.

At the preferential output unit 55 of a preface, the distance of the literal that correspondence has been named successfully compares with predetermined threshold value, if judge distance than threshold value hour then with result's output, will judge that when bigger then the result does not export than threshold value.

Control module 59, if when judging the preferential output unit 55 of a preface and export, in 50 literal in word recognition unit 52, selecting, begin successively a certain number of display unit 58 of delivering to is demonstrated literal on display unit 58 from the little literal of distance that should name recognition unit 54 to calculate by an ordered pair.

Under the situation of the input font of the input font of Fig. 3 and Figure 13, the correspondence of a common preface is named successfully, the distance of gained is more much smaller than threshold value, and the preferential output unit 55 of a judgement preface is exported, and should name the recognition result of recognition unit 54 to export as net result an ordered pair.

If judge the occasion of not exporting for the preferential output unit 55 of a preface, then enter the S5 of Fig. 2.

Below, for the action of the S5 of key diagram 2, just Figure 15 such with literal

Discuss with six strokes of occasions of writing into font.

With the input font of six strokes of Figure 15 that write, because of in stroke feature character library 53, having only 7 strokes

The standard font of word can not should name recognition unit 54 to do corresponding name with the input font of Figure 15 with an ordered pair.

Thereby, even supposition is named successfully with other the corresponding of six strokes of literal, also be the corresponding name of different literal because of becoming, very big with the distance of input font, the preferential output unit 55 of preface is judged because of not being the value littler than threshold value, does not export at the S4 of Fig. 2, enters S5.

At S5, draw 50 literal of number corresponding name recognition unit 57 with regard to selecting in the word recognition unit 52, utilize the DP coupling same with original technology example 2, the correspondence name of drawing the different persons of number makes apart from becoming minimum.

The stroke of input font is expressed as follows,

A=a1, a2 ..., ai ..., the stroke of the standard font of aI (I draws number for the input font) character library is expressed as follows,

B=b1, b2 ..., bj ..., bJ (J draws number for the standard font) is ai then, bj (i j) can calculate by following apart from d.

d(i，j)＝|wai-wbj|+|Hai-Hbj|

Here, W is the wide of stroke, and H is the height of stroke.

Setting d (i, j)=| ai-bj|

(1) at i=1, during j=1

g(1，1)＝d(1，1)

(2) i=1, j ≠ 1 o'clock

g(1，j)＝g(1，j-1)+d(1，j)

(3) i ≠ 1 is during j=1

g(i，1)＝g(i-1，1)+d(i，1)

(4) other situation

By the stepping type of representing above obtain g (i, j) (starting condition), the stroke row of obtaining A with following formula and the stroke row of B apart from S (A, B).

S (A, B)=I/ (I+J) * g (I, J) candidate character with regard to 50 recognition results of font recognition unit 52 gained carries out above-mentioned DP coupling, apart from S (A, B) from the beginning of minimum successively as recognition result to display unit 58 outputs.

Display unit 58 will be from the minimum literal of distance of being calculated by the DP coupling, and order shows literal.

In this example,, the input font of Figure 15 is done the such corresponding name of Figure 16 with the stroke arrangement of the standard font of Figure 12, can correctly discern by the DP coupling.

In addition, in the present embodiment, though the feature as font recognition unit 52 is to use the direction code characteristic, but also can be to touch the circumferential distribution feature of the area of character stroke until the front from the literal boundary rectangle, or the font style characteristic of reflection such as the line density of stroke literal integral body as feature.

Also have, in the present embodiment, be divided into 4 * 4 zones to code characteristic as generation, but also can be according to the character library capacity, recognition capability changes the quantity of cutting apart in zone.

Also have, in the present embodiment, be fixed with 50 as candidate character number, but also can change this value according to the recognition capability of font recognition unit 52 from 52 outputs of font recognition unit.

In addition, in the present embodiment, from the candidate character of font recognition unit 52, but also can change output candidate character number according to the recognition capability change of each recognition unit to a preface change corresponding name recognition unit 54 and stroke same quantity of the corresponding name of number recognition unit 57 outputs.

Effect with regard to the online character recognition device of present embodiment explains below.

If adopt above-mentioned present embodiment, only just pass through the candidate character of the general characteristics recognition result of literal, carry out to order outer identification by an ordered pair, by drawing the identification of the corresponding name of number, without font not similarly the feature of literal carry out correspondence and name, therefore, can realize discerning at a high speed and accurately.

Particularly, when carrying out handwriting identification, at a part of literal with JIS second level is the occasion of object, need 4000 kinds of literal are carried out high speed identification, an ordered pair should be named or stroke number is corresponding names as if just about 4000 words being carried out from beginning, in fact the processing time that will spend is inconvenient in fact.Here, in the present embodiment, make the first step with the identification of the 52 pairs of literal of font recognition unit and handle, utilize the general characteristics of literal to compress at high speed and carry out the literal that an ordered pair should be named identification or draw the corresponding name identification of number in subordinate phase.

And because output font similar literal is as candidate character, thereby the impression when the user having been improved mistake discerned has also improved the ratio that is included in the correct literal in the identification candidate character.

In addition, owing to should name the correspondence name of carrying out stroke in the recognition unit 54 by a plurality of features closely at an ordered pair, so can high-precision identification observe a stroke number writing words, exactly for the non-existent input font of drawing the number change of standard fonts such as gointed hand, can correctly send into and draw the corresponding name of number recognition unit 57, make gointed hand also can obtain high precision identification.

Show that as the literal of present embodiment a situation that the recognition result with the preferential output unit 55 of a preface shows and a situation about showing with the recognition result of drawing the corresponding name of number recognition unit 57 are arranged.

Thereby, the output of display unit 58 was formed by two stages.Therefore, in the occasion that is defined as preferential output unit 55 outputs of a preface, because can show, so can make recognition time in advance in previous stage.

In addition, in the present embodiment, though be to carry out font identification at first, second is carried out an ordered pair and should name identification, draw the corresponding name identification of number for the 3rd, but,, also can according to circumstances change second later recognition sequence in this preface occasion etc. of the company of attention pen more in the occasion of more paying attention to a preface than gointed hand.

Therefore, as the literal recognition sequence, can be thought of as following mode.

A: font identification → draw the corresponding name of a number identification → ordered pair should name identification

B: a font identification → ordered pair should be named identification

C: font identification → draw the number correspondence to name identification.Embodiment 2

Whether the online character recognition device of present embodiment exports the candidate character of being extracted out by the identification of this shape by the font identification decision, does not export occasion and carries out an ordered pair and should name identification being judged to be, and draws the corresponding name identification of number.Following with reference to Figure 17, Figure 18 describes.

Among Figure 17,50～50th, the equipment same with embodiment 1, explanation is omitted.

The 110th, basis is judged the preferential output unit of font of the identification candidate character of whether exporting font recognition unit 52 by the information of the identification candidate character of font recognition unit 52 gained.

Below, with Figure 18 the action of the online character recognition device of embodiment is described.

Figure 18 represents the process flow diagram of processing procedure of control module 59 of the online character recognition device of present embodiment.

Processing and embodiment 1 mode to S2 illustrate omission.

Finish S1, after S2 handles, control module 59 will be sent into the preferential output unit 110 of font by the literal of minor increment in 50 identification candidate characters of font recognition unit 52 gained.

Then enter the S30 of Figure 18, the preferential output unit 110 of font judges whether the recognition result of font recognition unit 52 is exported.Judge, literal for the minor increment of sending here by font recognition unit 52, obtain the similar degree (for example normalization distance) of output font and standard font, this similar degree than in advance for the threshold value of each literal preparation is judged to be the output result when big, be judged in little occasion and do not export the result.

If input font vector is f, standard font vector is g, and then similar degree is obtained as follows.

S(f，g)＝(f、g)/‖f‖·‖g‖

Here, the threshold value of literal A is obtained similar degree according to the standard font in the font character library 51 of literal A and the literal A direction code characteristic of writing font in addition of writing in the font of all text fonts that comprise identifying object, similar degree (the standard font of the most close literal A, the similar degree of the font beyond the literal A) to maximum is set.Like this, the threshold value of the non-existent literal A of the similar literal of font becomes little value, and the ratio that the result of font recognition unit 52 intactly exports increases.

In addition, because the literal threshold value that the similar literal of font exists becomes big, the part of being exported diminishes, and then can be discerned by detailed stroke feature.

If be judged to be the occasion of the preferential output unit 110 of font, then begin 50 the identification candidate characters of order demonstration by font recognition unit 52 gained from the little person of distance who calculates by S2 with result's output at S30.

If be judged to be the preferential output unit 110 of font not with the occasion of result output, then just discern candidate characters and carry out the S3 later processing same with embodiment 1 by 50 of font recognition unit 52 gained at S30.

In addition, in the present embodiment, the output of the preferential output unit 110 of font judges, though adopt the normalization distance as similar degree, also can be with Euclidean distance or Mahalanobis distance etc.

Below describe with regard to the effect of the online character recognition device of present embodiment.

The online character recognition device of present embodiment, when not having the similar literal of font, the ratio of the recognition result of output font recognition unit 52 increases, and may obtain not to be subjected to a preface to become, and draws the number change, and the local deformation of literal about high precision identification.In addition, the occasion at output setting recognition unit 52 recognition results may obtain the very identification of high speed.

And, if employing present embodiment, for the literal of very carefully writing, obtain high accuracy of identification by the font recognition unit, do not carry out an ordered pair and should name identification and at full speed show, therefore and usually the occasion of people's identification is same, can promptly discern the literal of carefully writing, improve the impression of giving the user.

And in the online character recognition device of present embodiment, in the result's of the font identification of carrying out the phase one the high occasion of reliability, owing to be to show this stage recognition result, the literal of therefore may identification at a high speed more carefully writing.

By way of example 3

The online character recognition device of present embodiment, just the correct high literal of literal possibility of resolving in the candidate character of extracting out by font identification should name the double calculation number of times of identification repeatedly to set to an ordered pair, below describes with reference to Figure 19～Figure 20.

Among Figure 19, the 50～59, the 110th, the equipment identical with the front embodiment omits its explanation.The 120th, should name recognition unit to the input font with the ordered pair that the corresponding name of the stroke of standard font is weighted.

The action of the online character recognition device of present embodiment describes with Figure 20.

Figure 20 is the process flow diagram of processing procedure of control module 59 of the online character recognition device of present embodiment.

Among Figure 20, identical to the processing of S30 with embodiment 2, omission is described.

Carry out S1, S2, the processing of S3 enters S40.At S40, control module 59 will similarly be sent into a weighting ordered pair by font recognition unit 52 resulting 50 candidate characters with embodiment 1 should name recognition unit 120, and a weighting ordered pair should be named recognition unit 120 to import font and discern with the corresponding name of the stroke of standard font.

The corresponding naming operation of detailed stroke is identical with the operation of the embodiment 1 of Fig. 8, but the arrangement of the preface of S25 transposing number of times (double calculation number of times), at embodiment 1, candidate character with regard to font recognition unit 52 gained is identical, otherwise, in the present embodiment, then change according to the order of the candidate character of font recognition unit 52 (order that begins to be arranged in order from the little person of distance).

Particularly, within the candidate character of the recognition result of font recognition unit 52, arrange the high literal of possibility that the high literal of precedence is correct explanation, the threshold value of double calculation number of times is just big more, the literal that precedence is low is the low literal of possibility of correct explanation, and the threshold value of double calculation number of times is just more little.

If employing present embodiment, to the low literal of correct answer possibility, reduced a quantity that preface row changes, needn't carry out above-mentioned correspondence name, so can discern at a high speed, and can reduce an above-mentioned wrong identification that the preface transposing causes of answering necessity of the low literal of possibility because of correct.

Embodiment 4

The online character recognition device of present embodiment is when drawing number correspondence name identification, extracts input font and standard font stable point one to one out, carry out the correspondence of each stroke with regard to stable point and name, below, describe with reference to Figure 21～Figure 25.

Among Figure 21,50～59 are identical equipment with embodiment 1,110 and embodiment 2,120 with embodiment 3, and omission is described.

130 is the DP matching unit, and 131 is the corresponding name calibrating of stroke unit, and 132 for stablizing the corresponding name of stroke unit.

With reference to Figure 22 the action of the online character recognition device of present embodiment is described.

Figure 22 is the process flow diagram of processing procedure of control module 59 of the online character recognition device of expression present embodiment.

Among Figure 22, identical to the action of S4 with front embodiment 1～3, omission is described, only later action describes with regard to S50.

At first, when S4 was judged to be preferential output unit 55 no-outputs of a preface, control module 59 was all delivered to DP matching unit 130 with 50 of the candidate characters of font recognition unit 52 gained recognition results.

Then, at S50,59 pairs of DP matching units 130 of control module are made indication, with regard to 50 candidate characters of the recognition result of font recognition unit 52 gained, import the corresponding name of font and standard font.

At this, the corresponding order path of the corresponding name calibrating of stroke unit 131 calibrating DP matching units 130, right should the name compensated, to prevent to select this path.

With regard to the path of the correspondence of DP matching unit 130 name, describe with the corresponding called after example of the standard font of Figure 12 with the input font of Figure 15.The input font of Figure 15 and the standard font of Figure 12 are done should name as shown in Figure 16, and the path of its corresponding name is presented on Figure 23 with two dimension.

Among Figure 23, transverse axis is that the stroke of standard font is arranged, and the longitudinal axis is arranged for the stroke of input font.Thick line is the corresponding named track of DP coupling among the figure.

The certain no problem of correct corresponding name occasion of Figure 24, but the place that will carry out the such correspondence name of Figure 24 in addition.

Carry out the local path of not corresponding name in the input font that Figure 24 is illustrated in DP matching unit 130 and actual the writing of standard font.

The correspondence name situation of Figure 24 is that the the 3rd, the 4th, the 5th stroke with the 3rd stroke stroke and the standard font of input font is corresponding.Such corresponding states is called the combination of stroke, and expression is original adopts two to mark all strokes of carrying and remember into stroke and connect pen state.

And, the 5th stroke of the standard font the 3rd stroke and the 4th stroke of corresponding name with the input font.Such state is called stroke and separates, and expression now demarcated record with two with the whole stroke of stroke record originally.

That is, the combination of drawing of the 5th metal marker of standard font takes place with separating simultaneously.Because such phenomenon does not take place in general character calligraph, so prevent not corresponding name by the corresponding name calibrating of stroke unit 131.

Promptly, the corresponding name calibrating of stroke unit 131 is in the DP unit 130 when carrying out the DP coupling, just from horizontal to perpendicular path and from perpendicular to cross road footpath etc. common write not corresponding name, the distance value by to its selected path the time is compensated greatly prevents to select this path.

Then enter S51, control module 59 makes on DP matching unit 130 the gained routing information deliver to stablize stroke calibrating unit 132.So-called routing information is exactly to the information that shows the correspondence of each stroke with line shown in Figure 25 as Figure 23.

Stablizing stroke calibrating unit 132 extracts this point out when the each point on the path is oblique path in the path, front and back as stable point.But when being oblique, be stable point when being the wrong path footpath in the terminal point dead ahead in path in the path in starting point dead astern, path.

Figure 25 represents with stablizing the stable point that stroke detecting unit 132 is extracted out.

Among Figure 25, in the input font of Figure 15 and the corresponding name result's of the standard font of Figure 12 path, represent with O stablize the stable point of being extracted out stroke extraction unit 132.Extract 5 stable point in this example out.

Then, stablize stroke calibrating unit 132, to the stable point of extracting out import font and standard font stroke in corresponding name one by one, should name recognition unit 54 used 7 special sheets that stroke is named with an ordered pair and examine and determine.

This corresponding name when among the stroke of judging at stable point being the corresponding name of no one, is then judged the correspondence name failure in the DP coupling, the correspondence name that enters down the candidate character of in-line recognition unit 52.

In the example of Figure 25, just import within the stroke of font, carry out with the calibrating of the corresponding name of the stroke of standard font owing to carry out correct corresponding name, so the judgement correspondence is named successfully in conjunction with the stroke beyond the stroke 102.

All 50 candidate characters with regard to word recognition unit 52 carry out above processing, will name the literal of distance minimum within the successful literal to begin the final recognition result output of a certain number of conduct of order from correspondence.

Now the effect with regard to the online character recognition device of present embodiment describes.

If adopt above present embodiment, the correspondence name when mating about DP is owing to control path in actual the writing, so that not corresponding name does not take place, therefore can realize high-precision identification.

In addition,,, the correspondence name of stroke is examined and determine, name, can realize high-precision identification so can prevent wrong correspondence with a plurality of features with regard to DP matching result stroke one to one if adopt present embodiment.

The present invention is owing to there is formation as described above, and following effect is arranged.

First invention should be named identification because of carrying out an ordered pair with regard to the grapholect of selecting in the font identification step, Can reduce and carry out the grapholect quantity that an ordered pair is answered name identification, and can do the identification of high-speed, high precision.

The second point invention is named identification because draw the number correspondence with regard to the grapholect of selecting in the font identification step, The grapholect quantity of drawing the corresponding name identification of number can be reduced, and the identification of high-speed, high precision can be done.

Thirdly invention is because have by will naming the similar degree of calculating in the identification step with pre-at an ordered pair Definite value compares first discriminating step that judges whether the outputting standard literal, and differentiates the step above-mentioned first When differentiating for output in rapid, should name the above-mentioned similar degree output of calculating in the identification step according to above-mentioned ordered pair The first output step of the grapholect of selecting in the font identification step is so should name the identification step at an ordered pair The occasion that rapid recognition result is good can be exported recognition result at a high speed.

The 4th invention is owing to carry out stroke and the above-mentioned standard literary composition of writing words with an order of above-mentioned writing words The correspondence name of the stroke of word, the order of the corresponding name of change stroke again in the time can not doing above-mentioned corresponding name Carry out the correspondence name, therefore, also can high accuracy even an above-mentioned writing words order is different from a grapholect order Identify.

The 5th some invention is because change with the corresponding number of times of the similar degree of calculating in the font identification step Ordered pair should be named the correspondence name of the order of strokes of identification step, so can make correct parsing possibility high The order order change of submissiveer ground of literal correspondence makes the low literal of correct parsing possibility less carry out an order change, But and the identification of high-speed, high precision.

The 6th invention has similar degree and predetermined value by grapholect selected in the above-mentioned font identification step with relatively above-mentioned writing words to compare after the font identification step because have, judge whether to export second discriminating step of above-mentioned grapholect

And, when in above-mentioned second discriminating step, differentiating for output, according to calculating in the above-mentioned font identification step The above-mentioned similar degree that goes out is exported the 3rd output step of above-mentioned grapholect, so, at the font identification step The occasion that recognition result is good can be with at a high speed output of identification.

The 7th invention be because from the stroke of corresponding name the DP coupling, the stroke of extracting writing words out with The stroke of grapholect is stablized stroke one to one, calculates with respect to above-mentioned with regard to the aforementioned stable stroke and writes The stable stroke of the similar degree of the above-mentioned grapholect of literal detects step, so can prevent in actual writing The discrimination that causes because of not corresponding name worsens.

The 8th invention should be named identification because of carrying out an ordered pair with regard to the grapholect of selecting in the font recognition unit, Can reduce and carry out the grapholect quantity that an ordered pair should be named identification, but the identification of high-speed, high precision.

The 9th invention identified because draw the corresponding name of number with regard to the grapholect of selecting in the font recognition unit, Can reduce the grapholect quantity of drawing the corresponding name identification of number, and can identify with high-speed, high precision ground.

The 10th invention is because having by will naming similar degree and the predetermined value of calculating in the recognition unit at an ordered pair Compare first judgement unit that judges whether the outputting standard literal, and, declare in the unit the above-mentioned the 1st When differentiating for output, should name the above-mentioned similar degree of calculating in the recognition unit at font according to above-mentioned ordered pair First output unit of the grapholect of selecting in the recognition unit is so should name in the recognition unit at an ordered pair The occasion that recognition result is good can be exported recognition result at a high speed.

The 11 invention is because having in the DP matching unit from the stroke of correspondence name, the extraction writing words Stroke is stablized one to one stroke and is calculated relative above-mentioned book with regard to the aforementioned stable stroke with the stroke of grapholect The stable stroke of similar degree of writing the above-mentioned grapholect of literal detects the unit, so can prevent from writing actual The middle discrimination that causes because of not corresponding name worsens.

The ten two invention compares because having by the grapholect that will select in word recognition unit similar degree and the predetermined value with respect to above-mentioned writing words, judge whether to export second judgement unit of above-mentioned grapholect, and

When in above-mentioned second judgement unit, differentiating for output, according to what calculate in the above-mentioned font recognition unit Above-mentioned similar degree is exported the 3rd output unit of above-mentioned grapholect, therefore, at above-mentioned font recognition unit The occasion that recognition result is good can be exported recognition result at high speed.

Claims

1. an online character recognition method has the written information input step that the action of writing of following literal is imported this written information,

It is characterized in that also having following steps:

From the feature that above-mentioned written information is extracted the global shape of the above-mentioned writing words of expression out, select a plurality of font identification steps that are similar to the grapholect of above-mentioned writing words according to the feature of the above-mentioned global shape of expression from character library,

Extract the first stroke feature of the feature that the stroke that constitutes above-mentioned writing words is described out from above-mentioned written information, according to above-mentioned first stroke feature, a preface ground of considering above-mentioned writing words carries out from the corresponding name of above-mentioned writing words stroke with the grapholect stroke of selecting above-mentioned font identification step, calculate an ordered pair of similar degree of the above-mentioned relatively writing words of a plurality of grapholects selected in above-mentioned font identification step and should name identification step, and

The first output step of the above-mentioned similar degree output selected grapholect in above-mentioned font identification step that should name in the identification step to be calculated according to above-mentioned ordered pair.

2. an online character recognition method has the written information input step that the action of writing of following literal is imported its written information,

It is characterized in that also having following steps:

From the global shape feature that above-mentioned written information is extracted the above-mentioned writing words of explanation out, select a plurality of font identification steps that are similar to the grapholect of above-mentioned writing words according to the feature of the above-mentioned global shape of explanation from character library,

Extract second stroke feature of the feature that the stroke that constitutes above-mentioned writing words is described out from above-mentioned written information, according to above-mentioned second stroke feature, consider that above-mentioned writing words draws the corresponding name of carrying out above-mentioned writing words stroke and the stroke of the grapholect of selecting in above-mentioned font identification step severally, calculate a plurality of grapholects of selecting in the above-mentioned font identification step and name identification step with respect to the number correspondence of drawing of the similar degree of above-mentioned writing words, and

The second output step according to the above-mentioned similar degree output of calculating in the corresponding name of the above-mentioned stroke of number identification step selected grapholect in above-mentioned font identification step.

3. an online character recognition method has the written information input step of importing its written information with the action of writing of literal,

It is characterized in that also including following steps:

Extract the above-mentioned writing words global feature of explanation out from above-mentioned written information, select a plurality of font identification steps that are similar to the grapholect of above-mentioned writing words from character library according to the feature of the above-mentioned global shape of explanation,

Extract the feature that explanation constitutes the first stroke of above-mentioned writing words stroke feature out from above-mentioned written information, according to above-mentioned first stroke feature, the corresponding name of the stroke of the grapholect of selecting in the stroke of considering to carry out above-mentioned writing words in above-mentioned writing words preface ground and the above-mentioned font identification step, calculate an ordered pair of the similar degree of the above-mentioned relatively writing words of selecting in the above-mentioned font identification step of a plurality of grapholects and should name identification step

By naming similar degree and the predetermined value calculated in the identification step to compare above-mentioned ordered pair, differentiate first discriminating step of whether exporting above-mentioned grapholect,

When in above-mentioned first discriminating step, differentiating for output, the first output step of the grapholect that should name the similar degree of calculating in the identification step to export to select in the above-mentioned font identification step according to above-mentioned ordered pair,

In above-mentioned first discriminating step, differentiate when not exporting, extract second stroke feature of the feature of the stroke of representing the above-mentioned writing words of formation out from above-mentioned written information, according to above-mentioned second stroke feature, consider that above-mentioned writing words draws the corresponding name of the stroke of carrying out the grapholect selected in above-mentioned writing words stroke and the above-mentioned font identification step severally, calculate a plurality of grapholects of selecting in the above-mentioned font identification step and name identification step with respect to the number correspondence of drawing of the similar degree of above-mentioned writing words, and

Export the second output step of the grapholect of selecting in the above-mentioned font identification step according to the above-mentioned similar degree of calculating in the above-mentioned stroke of corresponding identification step of number.

4. by claim 1 or 3 character recognition methods of putting down in writing, it is characterized in that, above-mentioned ordered pair should name identification step according to above-mentioned second stroke feature, carry out the corresponding name of above-mentioned writing words stroke and the stroke of the grapholect of in above-mentioned font identification step, selecting by an above-mentioned writing words preface, can not carry out in above-mentioned corresponding name, the order of strokes of the corresponding name of change carries out correspondence once more and names.

5. by the online character recognition method of claim 4 record, it is characterized in that above-mentioned font identification step is calculated the similar degree with respect to the above-mentioned grapholect of selecting of above-mentioned writing words,

Above-mentioned ordered pair should be named identification step, according to the number of times of the similar degree of the above-mentioned grapholect of calculating in the above-mentioned font identification step, changes the correspondence name of above-mentioned order of strokes.

6. the online character recognition method of putting down in writing by claim 3; It is characterized in that; Have by behind above-mentioned font identification step; Similar degree and the predetermined value of the grapholect of calculating in will the above-mentioned font identification step with respect to above-mentioned writing words compare; Differentiate second discriminating step of whether exporting above-mentioned grapholect; When in above-mentioned second discriminating step, differentiating for output; Export the 3rd output step of the grapholect of in above-mentioned font identification step, selecting according to the above-mentioned similar degree of calculating in the above-mentioned font identification step

In above-mentioned second determining step, differentiate when not exporting, carry out above-mentioned ordered pair and should name identification step.

7. by the online character recognition method of claim 2 record, it is characterized in that the above-mentioned stroke of corresponding name of number identification step calculated step by DP coupling step and similar degree and constituted,

Above-mentioned DP coupling step is extracted second stroke feature of the feature of the stroke of representing the above-mentioned writing words of formation out from above-mentioned written information, according to second stroke feature, carry out the corresponding name of above-mentioned writing words stroke and the stroke of the grapholect of in above-mentioned font identification step, selecting by DP coupling

Above-mentioned similar degree is calculated step, from the corresponding stroke of having named the above-mentioned DP coupling step, extract the stroke of above-mentioned writing words and the stroke of above-mentioned grapholect out and stablize stroke one to one, calculate the grapholect selected in the above-mentioned font identification step similar degree with respect to above-mentioned writing words with regard to the aforementioned stable stroke

The above-mentioned second output step is exported selected grapholect in the above-mentioned font identification step according to calculating the above-mentioned similar degree of calculating in the step at above-mentioned similar degree.

8. an online character recognition device has the written information input block with its written information of character calligraph action input,

It is characterized in that also including:

From the feature that above-mentioned written information is extracted the global shape of the above-mentioned writing words of expression out, select a plurality of font recognition units that are similar to the grapholect of above-mentioned writing words according to the above-mentioned global shape feature of expression from character library,

Extract the first stroke feature of the feature of the stroke of representing the above-mentioned writing words of formation out from above-mentioned writing words information, according to above-mentioned first stroke feature, consider that above-mentioned writing words preface ground carries out the corresponding name of above-mentioned writing words stroke and the grapholect stroke of being selected by above-mentioned font recognition unit, calculate a plurality of grapholects of selecting by above-mentioned font knowledge unit and should name recognition unit with respect to an ordered pair of the similar degree of above-mentioned writing words, and

Should name the above-mentioned similar degree of calculating in the recognition unit to export first output unit of the grapholect of selecting in the above-mentioned font recognition unit according to above-mentioned ordered pair.

9. an online character recognition device has the written information input block with its written information of character calligraph action input,

It is characterized in that also including:

From the feature that above-mentioned written information is extracted the global shape of the above-mentioned writing words of expression out, from character library, select a plurality of font recognition units that are similar to the grapholect of above-mentioned writing words according to the feature of above-mentioned expression global shape,

Extract second stroke feature that expression constitutes the stroke feature of above-mentioned writing words out from above-mentioned written information, according to above-mentioned second stroke feature, consider that above-mentioned writing words draws the corresponding name of carrying out the grapholect stroke selected in above-mentioned writing words stroke and the above-mentioned font recognition unit severally, calculate a plurality of grapholects selected in the above-mentioned font recognition unit and count corresponding recognition unit with respect to the drawing of similar degree of above-mentioned writing words

Export second output unit of the grapholect of selecting in the above-mentioned font recognition unit according to the above-mentioned similar degree of being calculated in the corresponding name of the above-mentioned stroke of number recognition unit.

10. an online character recognition device has the written information input block of importing its written information with the action of writing of literal,

It is characterized in that also including:

Extract the above-mentioned writing words global shape feature of expression out from above-mentioned written information, select a plurality of font recognition units that are similar to the grapholect of above-mentioned writing words from character library according to the feature of above-mentioned expression global shape,

Extract the first stroke feature of the feature of the stroke of representing the above-mentioned writing words of formation out from above-mentioned written information, consider that according to above-mentioned first stroke feature the preface ground of above-mentioned writing words carries out the corresponding name of the grapholect stroke selected in above-mentioned writing words stroke and the above-mentioned font recognition unit, calculate the ordered pair with respect to the similar degree of above-mentioned writing words of a plurality of grapholects of selecting in the above-mentioned font recognition unit and should name recognition unit

By above-mentioned ordered pair being named similar degree and the predetermined value calculated in the recognition unit compare, differentiate first judgement unit of whether exporting above-mentioned grapholect,

When in above-mentioned first judgement unit, differentiating output, should name the above-mentioned similar degree of calculating in the recognition unit to export first output unit of the grapholect of in above-mentioned font recognition unit, selecting according to above-mentioned ordered pair,

In above-mentioned first judgement unit, differentiate when not exporting; Extract second stroke feature of the feature of the stroke that represents the above-mentioned writing words of formation out from above-mentioned written information; Consider that according to above-mentioned second stroke feature drawing of above-mentioned writing words carry out the corresponding name of the stroke of the grapholect selected in above-mentioned writing words stroke and the above-mentioned font recognition unit severally; Calculate the drawing of similar degree of the relatively above-mentioned writing words of a plurality of grapholects of selecting in the above-mentioned font recognition unit and count corresponding name recognition unit

Export second output unit of the grapholect of selecting in the above-mentioned font recognition unit according to the above-mentioned similar degree of calculating in the corresponding name of the above-mentioned stroke of number recognition unit.

11. the online character recognition device according to claim 9 record is characterized in that:

The above-mentioned stroke of corresponding name of number recognition unit is made of DP matching unit and similar degree computing unit;

Above-mentioned DP matching unit is extracted the 2nd stroke feature of the feature of the stroke of representing the above-mentioned writing words of formation out from above-mentioned written information, according to above-mentioned second stroke feature, utilize DP coupling to carry out the corresponding name of the stroke of the grapholect selected in above-mentioned writing words stroke and the above-mentioned font recognition unit;

Above-mentioned similar degree computing unit is the corresponding stroke of having named from above-mentioned DP matching unit, extract above-mentioned writing words stroke out and above-mentioned grapholect stroke is stablized stroke one to one, calculate the similar degree with respect to above-mentioned writing words of the grapholect of selecting in the above-mentioned font recognition unit with regard to the aforementioned stable stroke;

Above-mentioned second output unit is exported the grapholect of selecting according to the above-mentioned similar degree of calculating in the above-mentioned similar degree calculating in last class font recognition unit.

12. the online character recognition device by claim 10 record is characterized in that having

Similar degree and predetermined value by the grapholect that will calculate in above-mentioned font recognition unit with respect to above-mentioned writing words compare, and differentiate second judgement unit of whether exporting above-mentioned grapholect, and

When in above-mentioned second judgement unit, differentiating, export the 3rd output unit of the grapholect of selecting in the above-mentioned font recognition unit according to the last similar class degree of calculating in the above-mentioned font recognition unit for output,

Above-mentioned ordered pair should be named recognition unit, and differentiation is handled for the occasion of no-output in above-mentioned second judgement unit.