CN105095830B - Handwriting tracks recognition methods, handwriting tracks identification equipment and handwriting input device - Google Patents
Handwriting tracks recognition methods, handwriting tracks identification equipment and handwriting input device Download PDFInfo
- Publication number
- CN105095830B CN105095830B CN201410182172.5A CN201410182172A CN105095830B CN 105095830 B CN105095830 B CN 105095830B CN 201410182172 A CN201410182172 A CN 201410182172A CN 105095830 B CN105095830 B CN 105095830B
- Authority
- CN
- China
- Prior art keywords
- rotation angle
- handwriting tracks
- handwriting
- tracks
- identification
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Landscapes
- Character Discrimination (AREA)
Abstract
A kind of handwriting tracks recognition methods, handwriting tracks identification equipment and handwriting input device, the handwriting tracks recognition methods include：First obtaining step obtains the tracing point of the preceding strokes for the handwriting tracks to be entered；Rotation angle determines step, and the rotation angle of handwriting tracks is determined using the rotation angle detection model for the rotation angle that can detect the preceding strokes based on the tracing point；Second obtaining step obtains the remaining stroke of handwriting tracks；And identification step, handwriting tracks are identified based on identified rotation angle.
Description
Technical field
The application generally relates to a kind of handwriting tracks recognition methods and handwriting tracks identification equipment, especially and handwriting tracks
The unrelated handwriting tracks recognition methods and handwriting tracks identification equipment in direction of rotation.The application further relates to include the handwriting tracks
The handwriting input device of identification equipment.
Background technology
Handwriting recognizer, which can be used for identifying, to be written directly to touch in sensitizing input equipment (such as screen) and/or be used for from book
The handwriting tracks of the character scanned in written document.It come when identifying character, such as is identified with non-chronological and with whole word pattern
When the character of OCR scannings, frequently referred to identified off-line.Alternatively, when character is write and to be identified with time sequencing on such as screen,
I.e. with character is written into and it is identified when, frequently referred to online recognition.In the case of online recognition, when user is by with finger
It is write on the touching sensitive screen curtain of handheld device come when hand-written character is input on the screen of handheld device, due to inputting word
The orientation of symbol may change with equipment and user, it is possible that can not be accurately identified as it is expected user.
Therefore need a kind of effective handwriting recognizer, so as to input character it is this it is rotationally-varying obtain it is accurate and
Steady recognition result.However it is difficult to develop to the rotationally-varying steady and at the same time the small speed of size is at low cost of handwriting tracks
Effective handwriting recognizer.In order to handle the rotationally-varying of handwriting input, it has been proposed that two kinds of possible methods.First method is
One group of new samples is obtained by artificially rotating handwriting samples with possible angle, and uses the mould by specification sample training
Type identifies that the group may sample.Second method is the one group of corresponding model established for one group of rotation angle, and is then used
Such model identifies unknown handwriting samples.
For example, U.S. Patent Publication US20110268351 (hereinafter referred to as document 1) proposes one kind using rough sort first
Device come obtain recognition result candidate and the rotation angle candidate of handwriting samples so as to be applied to estimation affine transformation, then lead to
The method that meticulous grader is identified.Fig. 1 shows the flow chart of this method.Its key step is as follows：
1) handwriting tracks input step S101：The handwriting tracks for representing character to be identified or word are received from user；
2) rough sort device identification step S102：Use identification engine (such as the base of " rotating freely " and opposite " low cost "
In the grader of simple template) it is used as rough sort device, it is candidate to obtain recognition result candidate and rotation angle for handwriting tracks；
3) disaggregated classification device identification step S103：Use the identification engine of " high cost " but " accurate " as disaggregated classification device from time
Choose acquisition recognition result.
Although above method can improve the identification accuracy under rotational case, there is following defect.First, rough segmentation
Class device uses entire character as input, this will influence the advantages of prediction based on part stroke of handwriting recognition.Secondly, draw
It is still high-cost to enter rough sort device.Rough sort device itself is the balance between cost and accuracy.Too simple grader
The candidate that cannot have been obtained filters out, therefore will bring burden that is too many candidate and leading to disaggregated classification device for disaggregated classification device.If made
With the rough sort device filtered out with acceptable candidate, even if then the cost of rough sort device is in contrast lower than disaggregated classification device, repeatedly
Identification will also decrease recognition speed.
It can be seen that the method for the prior art is rotationally-varying in order to detect, exists and the handwriting tracks of input are located as a whole
The problem of managing and entire handwriting tracks must repeatedly being identified.
So far, it is not also a kind of selection to be detected using a part for handwriting tracks rotationally-varying, this is because one
The end that aspect detects the part is difficult or cost is very high, on the other hand not expectability device is easily recognized can produce for several strokes
It is raw to estimate rotationally-varying high accuracy.
Invention content
The first purpose of the application is to improve the rotationally-varying robustness that identifier is directed to handwriting tracks with low cost
And accuracy, without repeatedly being identified to entire handwriting tracks, to improve the efficiency of handwriting recognition.
The application's relates in one aspect to a kind of handwriting tracks recognition methods, including：First obtaining step, acquisition will be entered
Handwriting tracks preceding strokes tracing point；Rotation angle determines step, based on the tracing point use can detect it is described before
The rotation angle detection model of the rotation angle of strokes determines the rotation angle of the handwriting tracks；Second obtaining step, is obtained
Take the remaining stroke of the handwriting tracks；And identification step, the handwriting tracks are identified based on identified rotation angle.
The application yet another aspect relates to a kind of handwriting tracks methods of adjustment, including：First obtaining step, acquisition are defeated
The tracing point of the preceding strokes of the handwriting tracks entered；Rotation angle determines step, former for detecting based on the tracing point use
The rotation angle detection model of the rotation angle of pen determines the rotation angle of the handwriting tracks；Second obtaining step obtains
The remaining stroke of handwriting tracks；And set-up procedure, handwriting tracks are adjusted based on identified rotation angle.
The another aspect of the application is related to a kind of handwriting tracks identification equipment, including：First acquisition device is configured as obtaining
Take the tracing point of the preceding strokes for the handwriting tracks to be entered；Rotation angle determining device is configured as being based on the tracing point
The rotation angle of the handwriting tracks is determined using the rotation angle detection model for the rotation angle that can detect the preceding strokes
Degree；Second acquisition device is configured as obtaining the remaining stroke of the handwriting tracks；And identification device, it is configured as being based on
Identified rotation angle identifies the handwriting tracks.
Preferably, the preceding strokes be can by the character of character set share a part of stroke or by the word of set of words
Shared a part of stroke.
Preferably, rotation angle determining device may include：Tracing point acquisition device is configured as obtaining current track
Point；Decoding apparatus is configured with the rotation angle detection model to including the partial traces until current trace points
It is decoded；Judgment means are configured as judging the end tracing point whether current trace points reach the preceding strokes；And really
Determine device, is configured as determining in the case where judging result is affirmative described in rotation angle conduct corresponding with decoding result
The rotation angle of handwriting tracks.
Preferably, decoding apparatus include with based on hidden Markov model come the rotation angle detection model trained to packet
The partial traces included until current trace points are decoded to obtain the device of optimum state transfer path；Judgment means include
Judge whether optimum state transfer path reaches the device of end state；And determining device is the case where judging result is certainly
The device of rotation angle of the lower determination rotation angle corresponding with optimum state transfer path as the handwriting tracks.
Preferably, decoding apparatus includes with the rotation angle detection model trained based on dynamic time programming DTW models
To including that the partial traces until current trace points are decoded to obtain the device of best matching path；Judgment means include sentencing
Whether disconnected best matching path reaches the device of end node；And it is in the case of affirming that determining device, which is included in judging result,
Determine the device of rotation angle of the rotation angle corresponding with best matching path as the handwriting tracks.
Preferably, rotation angle detection model can be integrated into the whole word model for the whole word of online recognition or be used for
In the whole word model of the entire word of online recognition.
Preferably, the confidence level for the preceding strokes that rotation angle determining device obtains can be integrated into the remaining stroke of identification
In the process.
Preferably, identification device includes：Correcting device is configured as correcting hand-written rail based on identified rotation angle
At least part of mark, identification device after correction are configured as at least part of the handwriting tracks after identification is corrected.
Preferably, in the case where handwriting recognizer is online recognition device, at least part that be repaired and identify
It is the remaining stroke obtained by the second acquisition device.
Preferably, it in the case where handwriting recognizer is offline recognizer, to be repaired and the part identified is entire hand
Write track.
Preferably, identification device includes：Model selection device is configured as the rotation of selection and identified handwriting tracks
The corresponding rotation identification model of angle, identification device after selection are configured as identifying hand with selected rotation identification model
Write at least part of track.
Preferably, in the case where handwriting recognizer is offline recognizer, the part to be identified is entire handwriting tracks.
Preferably, in the case where handwriting recognizer is online recognition device, the part to be identified is to obtain to fill by second
Set the remaining stroke of acquisition or the stroke of entire handwriting tracks.
The another further aspect of the application is related to a kind of handwriting input device, including：Input unit is configured as receiving hand-written rail
The input of mark；And handwriting tracks identification equipment as elucidated before.
Therefore, according to the various aspects of the application, need not identified off-line repeatedly can be carried out to entire handwriting tracks, so that it may
With detection rotation angle is worked as to carry out more acurrate and steady identification to handwriting tracks to improve online at low cost
The accuracy of preceding identifier and robustness, reduce cost, improve efficiency.
In addition, rotation angle is detected before inputting all writing strokes.Therefore, for being based on part stroke
The handwriting recognition advantage of prediction do not influence.As a result, natural can be with the recognition methods of prediction of the support based on part stroke
It is combined.
Description of the drawings
With reference to specific embodiment, and with reference to attached drawing, above and other objects and advantages of the application are done into one
The description of step.In the accompanying drawings, identical or corresponding technical characteristic or component will be using identical or corresponding reference numerals come table
Show.
Fig. 1 shows the flow chart of handwriting tracks recognition methods in the prior art；
Fig. 2 shows the schematic diagrames according to the application environment of the handwriting tracks recognition methods of the application；
Fig. 3 describes the flow chart of the handwriting tracks recognition methods of one embodiment according to the application；
Fig. 4 shows the flow chart of the rotation angle determination process according to the application；
Topological diagram and the rotation that model is determined according to the rotation angle of the application one embodiment is shown respectively in Fig. 5 A and 5B
The flow chart of angle determination process；Fig. 5 C and 5D are shown respectively determines model according to the rotation angle of the application another embodiment
Topological diagram and rotation angle determination process flow chart；
Fig. 6 A show the flow chart of one embodiment of identification step；Fig. 6 B show that in handwriting recognizer be online recognition device
In the case of identification model used topological diagram；Fig. 6 C show to identify step in the case where handwriting recognizer is online recognition device
Rapid schematic diagram；Fig. 6 D show the schematic diagram of the identification step in the case where handwriting recognizer is offline recognizer；
Fig. 7 shows the flow chart of another embodiment of identification step；
Fig. 8 shows the block diagram of the exemplary configuration of the handwriting tracks identification equipment according to one embodiment of the application；
Fig. 9 shows the block diagram of the exemplary configuration of the handwriting input device of one embodiment according to the application；And
Figure 10 shows that the block diagram of the hardware configuration of the computer system of embodiments herein can be implemented.
Specific implementation mode
The exemplary embodiment of the application is described hereinafter in connection with attached drawing.For clarity and conciseness,
All features of embodiment are not described in the description.It should be understood, however, that must during implementing to embodiment
The setting much specific to embodiment must be made, to realize the objectives of developer, for example, meeting and equipment and industry
Those of correlation of being engaged in restrictive condition, and these restrictive conditions may be changed with the difference of embodiment.In addition,
It will also be appreciated that although development is likely to be extremely complex and time-consuming, to having benefited from this field of present disclosure
For technical staff, this development is only routine task.
Herein, it should be noted that in order to avoid having obscured the application because of unnecessary details, only show in the accompanying drawings
The processing step and/or device structure closely related with the scheme according at least to the application, and be omitted and the application relationship
Little other details.
First, Fig. 2 is the signal for the application environment for showing handwriting tracks recognition methods according to an embodiment of the invention
Figure.User 201 inputs handwriting tracks to be identified by using writing pencil 203 on the screen of handwriting input device 202
A part 204.Alternatively, user 201 can directly be inputted using finger or mouse in equipment 202.As an example, hand
It can be any equipment for having touch screen or similar hand-write input function to write input equipment 202, such as tablet computer and intelligent hand
Machine etc..
Referring to Fig. 3 descriptions according to the flow chart of the handwriting tracks recognition methods of one embodiment of the application.At this
In handwriting tracks recognition methods, rotation angle is detected by identifying the preceding strokes of inputted handwriting tracks.Then, it is based on inspection
The rotation angle measured is come the remaining stroke for correcting or adjusting handwriting tracks or whole strokes.
Step S301 is the first obtaining step, wherein obtaining the handwriting tracks to be entered in the environment similar to Fig. 2
Preceding strokes tracing point.Handwriting tracks can for example represent East Asia character and western language word.
Preferably for East Asia character, " preceding strokes " is radical, such as " yarn, speech, stone, standing grain, Fu ", referred to as " starting radical ".
For western language word, " preceding strokes " is at least one starting letter, such as a, ab etc..The starting radical or starting letter can divide
It is not shared by the character in character set or the word in set of words.
Preceding strokes is not limited to the above situation.Alternatively, it can also be a part for the radical in above-mentioned example.Even root
Can also be any preceding several strokes being entered at first according to writing style difference.
In addition, in writing process, such as the screen of equipment 202 adopts the movement locus of the pen tip of writing pencil 203
Sample, and the result sampled is expressed as a series of coordinate points for such as lifting or writing state containing tip position information.Therefore, it moves
Track can at least be expressed as coordinate points and the set of tip position information.
Step S302 is that rotation angle determines step, wherein can detect the preceding strokes based on tracing point use
The rotation angle detection model of rotation angle determines the rotation angle of the handwriting tracks.
After obtaining the tracing point of preceding strokes of handwriting tracks, directly detected using rotation angle detection model former
The rotation angle of pen, and the rotation angle is determined as the rotation angle of entire handwriting tracks.Rotation angle detection model
Foundation will be described later.
Several strokes (as originated letter or radical) detect rotation angle before being focused only on due to step S302, so
The size of the identification model of strokes is smaller before being applied not only to, and can be in the case where avoiding repeatedly identifying entire handwriting tracks
Improve detection speed.Rotation angle determination process will be described in greater detail below.
Step S303 is the second obtaining step, wherein obtaining the remaining stroke of the handwriting tracks.It is corresponding with step S301,
Remaining stroke can be other strokes in addition to preceding strokes.
Preferably for East Asia character, if " preceding strokes " is radical, remaining stroke is to form the remainder of the character
It is first.For western language word, if " preceding strokes " is at least one starting letter, remaining stroke is to form the remaining word of the word
It is female.
Step S304 is identification step, wherein identifying handwriting tracks based on identified rotation angle.That is,
After rotation angle is determined, can it is more acurrate and steadily to handwriting tracks represent input be identified.
Alternatively, rotation angle determined by being based only upon as needed without identification adjusts handwriting tracks,
Such as rotation angle is adjusted to bigger.Or handwriting tracks can also be adjusted and be identified again later.
Therefore, handwriting tracks recognition methods according to an embodiment of the present application need not repeatedly carry out entire handwriting tracks
Identified off-line, so that it may more acurrate and steady knowledge be carried out to handwriting tracks to detect rotation angle online at low cost
Not, to improve accuracy and the robustness of current identifier, cost is reduced, efficiency is improved.According to below detailed
The advantage may be better understood in description.
Fig. 4 describes the flow chart of the rotation angle determination process according to the application.In Fig. 4, step S401 is tracing point
Obtaining step, wherein obtaining current trace points.Step S402 is decoding step, wherein using rotation angle detection model to including
Partial traces until current trace points are decoded, this will be illustrated later.Step S403 is judgment step, wherein
Judge the end tracing point of strokes before whether current trace points reach.If it is judged that for affirmative, then step is entered step
Otherwise S404 is repeated from step S401.Step S404 is to determine step, wherein really in the case where judging result is affirmative
Fixed rotation angle of the rotation angle corresponding with decoding result as the handwriting tracks.
Rotation angle according to an embodiment of the invention is described with reference to Fig. 5 A and 5B separately below and determines opening up for model
Flutter the flow chart of figure and rotation angle determination process.Rotation angle determines that model is to be based on hidden markov in this embodiment
Model HMM is trained.
Rotation angle determines that model only needs for example to originate radical or starting letter for preceding strokes and be established.
It is obtained from 50 Japanese 3991 Hanzi specimens first as statistical sample, by 255 chinese characters
Composition, each character include 5 starting radicals " yarn, speech, stone, standing grain, one of Fu ".
Then, it is built for the rotation angle of each radical predetermined quantity (such as 5 °, 10 °, 15 °, 20 °, 25 °, 30 °)
Found corresponding model.Therefore it needs to establish 7 rotating models for each radical, as shown in Figure 5A.
In fig. 5,7 rotation angles for each radical that the HMM based on tiled configuration is trained are given to examine
Survey the topological diagram of model.Each circle indicates a state of model, such as indicates but be not limited to a stroke.Per a line
The state transition path of circle sequence table representation model.Arrow around circle indicates state transfer.Although topology in fig. 5
HMM based on tiled configuration, but this is not limited, the application can also be opened up to use based on other according to the feature extracted
Flutter the rotation angle detection model that the HMM of structure is trained.
The flow chart of rotation angle determination process according to the embodiment of the application is described referring to Fig. 5 B.
Step S501 corresponds to the tracing point obtaining step of step S401, wherein current trace points are obtained, such as writing pencil
203 coordinate points being currently moved to and relevant tip position information.
Step S502 corresponds to the decoding step of step S402, wherein with the rotation trained based on hidden Markov model
Gyration detection model is to including that the partial traces until current trace points are decoded to obtain optimum state transfer path.
Partial traces until acquired current trace points are first converted to the rotation angle detection model can be with
The observation value sequence of reception.For example, can extract the sequence of inputted pen section according to partial traces, each pen section can be seen
Make an observed value.Although listing example of the pen section as observed value, observed value is not limited to pen section, can also be extraction
Stroke or other feature vectors etc..
Next, determining that model is decoded the partial traces using each rotation angle that training obtains.Decoding problem
It is the known problem that HMM is solved, its purpose is to find one group of optimum state metastasis sequence so that handy model obtains representative and is somebody's turn to do
The maximum probability of the observation value sequence of partial traces.Any algorithm that can realize the purpose can be used.Preferably, can make
Probability and the maximum model of select probability that the observation value sequence is obtained by each model are calculated with viterbi algorithms, from
And obtain optimum state transfer path corresponding with the model.
Step S503 corresponds to the judgment step of step S403, wherein judging whether optimum state transfer path reaches end
State.As shown in Figure 5A, each model there are one terminate state, correspond to preceding strokes recently enter and therefore can be with base
Judge rotation angle in the optimum state transfer path.If a determination be made that certainly, then enter step S504；Otherwise from
Step S501 is repeated.
Step S504 corresponds to the determination step of step S404, wherein if the judging result of step S503 is affirmative,
Then determine rotation angle of the rotation angle corresponding with optimum state transfer path as the handwriting tracks.
It, need not be to entire hand-written rail by above step as it can be seen that in rotation angle determination process according to this embodiment
Mark handle and therefore avoids offline Multiple recognition, and detection speed is improved in the case where moulded dimension is smaller
And accuracy.
Fig. 5 C and 5D describe topological diagram and the rotation that rotation angle in accordance with another embodiment of the present invention determines model respectively
The flow chart of gyration determination process.Rotation angle determines that model is instructed based on dynamic time warping DTW in this embodiment
Experienced.
Rotation angle is detected in order to use DTW, it is similar with previous embodiment, for the rotation of each radical predetermined quantity
Gyration (such as 5 °, 10 °, 15 °, 20 °, 25 °, 30 °) establishes corresponding model.Therefore it needs to establish for this each radical
7 rotating models, as shown in Figure 5 C.
The rotation angle detection model for each radical trained based on DTW is given in Fig. 5 C, wherein storing
DTW templates.Each circle indicates a node of model, such as indicates but be not limited to each sample for same character
Correspondence tracing point average value.Circle sequence per a line indicates the matching of the model of the rotation angle of strokes before capable of detecting
Path.
The flow chart of rotation angle determination process according to the embodiment of the application is described referring to Fig. 5 D.
Step S501 ' corresponds to the tracing point obtaining step of step S401, wherein current trace points are obtained, such as writing pencil
203 coordinate points being currently moved to and relevant tip position information.
Step S502 ' corresponds to the decoding step of step S402, wherein with the rotation angle detection model trained based on DTW
To including that the partial traces until current trace points are decoded to obtain best matching path.
Best matching path for example determines as follows：The tracing point on the partial traces is extracted first, is then calculated at this
A little tracing points and the distance between the corresponding tracing point in the DTW templates in model, finally select the path of total distance minimum to make
For best matching path.Although listing the example for calculating distance as the standard of determination, the standard of determination is not limited to distance, may be used also
To be other geometric standards.
Preferably, the partial traces can be decoded using dynamic time warping algorithm to obtain best match road
Diameter.
Step S503 ' corresponds to the judgment step of step S403, wherein judging whether best matching path reaches end section
Point.As shown in Figure 5 C, there are one end nodes for each model, correspond to recently entering and therefore being based on for preceding strokes
The best matching path judges rotation angle.If a determination be made that certainly, then enter step S504 '；Otherwise from step
S501 ' is repeated.
Step S504 ' corresponds to the determination step of step S404, wherein if the judging result of step S503 ' is affirmative
, it is determined that rotation angle of the rotation angle corresponding with best matching path as the handwriting tracks.
Similarly, in rotation angle determination process according to this embodiment, need not be to entire handwriting tracks at
It manages and therefore avoids offline Multiple recognition, and improve in the case where moulded dimension is smaller detection speed and accurate
Property.
The description of Fig. 3 is turned again to below.For identification step S304, Fig. 6 A show the flow chart of one embodiment.
Step S601 is to correct step, wherein correcting at least one of handwriting tracks based on identified rotation angle
Point.
Here " correct " is that instigate handwriting tracks relative to the rotation angle of the vertical axis of the screen of equipment 202 be zero
Operation.
Step S602 is identification step after correcting, wherein at least part of the handwriting tracks after identification correction.
This can be at least partially identified using general handwriting recognizer.According to one embodiment, hand-written
In the case that identifier is online recognition device, because preceding strokes is come out by online recognition, to be repaired and identify
This is the remaining stroke obtained by the second obtaining step S303 at least partially.
In this case, rotation angle detection model can be integrated into the whole word model for the whole word of online recognition
Or in the whole word model for the entire word of online recognition.Fig. 6 B are shown in the case where handwriting recognizer is online recognition device
The topological diagram of identification model used.That is, the state transition path of rotation angle determination process is (such as dotted line left side institute
Showing) the state metastasis sequence that can be used directly to detect corresponding remaining stroke follows even shared, it is determined from rotation angle
The process that the confidence level of the preceding strokes obtained in step also can directly be identified remaining stroke uses, so as to avoid to entire
The multiple identification of handwriting tracks.In the embodiment shown in Fig. 6 B, also use HMM as the knowledge for the handwriting tracks after correcting
Other model.It is understood that this is also applicable for the identifier based on DTW principles.
Fig. 6 C show the schematic diagram of the identification step in the case where handwriting recognizer is online recognition device.Label 601 indicates
The similar handwriting input device indicated with the label 202 in Fig. 2, strokes input before label 602 indicates, and label 603 indicates
The remaining stroke of handwriting tracks.As shown in Figure 6 C, in the case where handwriting recognizer is online recognition device, remaining pen is only corrected for
It draws part and continues identification without again identifying that preceding strokes for being identified device.
Fig. 6 D show the schematic diagram of the identification step in the case where handwriting recognizer is offline recognizer.According to the reality of Fig. 6 D
Example is applied, in the case where handwriting recognizer is offline recognizer such as OCR, because identified off-line is not based on time sequencing,
So to be repaired and the part identified is entire handwriting tracks.In this case it is necessary to correct the complete of entire handwriting tracks
Continue to identify for being identified device in portion stroke part.
Fig. 7 shows the flow chart of another embodiment of identification step S304.
Step S701 is model selection step, wherein selection rotation corresponding with the rotation angle of identified handwriting tracks
Turn identification model.
Here " rotation identification model " refers to the model of the handwriting tracks with specific rotation angle for identification, is established
Mode is similar to the mode with reference to Fig. 5 A or Fig. 5 C descriptions.
Step S702 is identification step after selection, wherein identifying handwriting tracks at least with selected rotation identification model
A part.
Similarly, in the case where handwriting recognizer is online recognition device, because preceding strokes is come out by online recognition,
So the part to be identified is the remaining stroke obtained by the second obtaining step S303, naturally it is also possible to be whole strokes.
In the case where handwriting recognizer is offline recognizer, part to be identified is entire handwriting tracks.
Although be above using the handwriting tracks recognition methods illustrated for East Asia character according to the application, should
Understand that this is also fully applicable for European word, for example, if at least one starting letter of word is regarded as " preceding strokes ", then remains
Remaining stroke is to form the residue letter of the word.
It may be better understood by above detailed description, it need not be more according to the handwriting tracks recognition methods of the application
It is secondary to entire handwriting tracks carry out identified off-line, so that it may at low cost online detection rotation angle so as to handwriting tracks into
The more acurrate and steady identification of row reduces cost, improves effect to improve accuracy and the robustness of current identifier
Rate.
The exemplary of the handwriting tracks identification equipment 800 of one embodiment according to the application is described referring next to Fig. 8
The block diagram of configuration.The handwriting tracks identification equipment 800 includes：First acquisition device 801 is configured as obtaining the hand to be entered
Write the tracing point of the preceding strokes of track；Rotation angle determining device 802 is configured as to detect based on the tracing point use
The rotation angle detection model of the rotation angle of the preceding strokes determines the rotation angle of handwriting tracks；Second acquisition device
803, it is configured as obtaining the remaining stroke of handwriting tracks；And identification device 804, it is configured as based on identified rotation angle
It spends to identify handwriting tracks.
Device 801-804 can be configured as executes step S301-S304 respectively.
Accoding to exemplary embodiment, rotation angle determining device 802 may include：Tracing point acquisition device 805, is configured
To obtain current trace points；Decoding apparatus 806 is configured with rotation angle detection model to including being to current trace points
Partial traces only are decoded；Judgment means 807 are configured as judging the knot whether current trace points reach the preceding strokes
Beam trajectory point；And determining device 808, it is corresponding with decoding result to be configured as the determination in the case where judging result is affirmative
Rotation angle of the rotation angle as handwriting tracks.
Preferably, decoding apparatus 806 includes the rotation angle inspection being configured to being trained based on hidden Markov model
Model is surveyed to including that the partial traces until current trace points are decoded to obtain the device of optimum state transfer path；Sentence
Disconnected device 807 includes being configured to judge the device whether optimum state transfer path reaches end state；And determining device 808
Including being configured to determine rotation angle conduct corresponding with optimum state transfer path in the case where judging result is affirmative
The device of the rotation angle of the handwriting tracks.
Preferably, decoding apparatus 806 include be configured to based on DTW models come the rotation angle detection model trained to packet
The partial traces included until current trace points are decoded to obtain the device of best matching path；Judgment means 807 include matching
It is set to the device for judging whether best matching path reaches end node；And determining device 808 includes being configured to judging to tie
Fruit is the dress that rotation angle of the rotation angle corresponding with best matching path as handwriting tracks is determined in the case of affirming
It sets.
Accoding to exemplary embodiment, identification device 804 may include：Correcting device 809 is configured as based on determined by
Rotation angle corrects at least part of handwriting tracks；And identification device 810 after correcting, after being configured as identification correction
At least part of handwriting tracks.
According to alternative embodiment, identification device 804 may include：Model selection device 811, selection and identified hand
Write the corresponding rotation identification model of rotation angle of track；And identification device 812 after selection, identify mould with selected rotation
Type identifies at least part of handwriting tracks.
Preferably, rotation angle detection model can be integrated into the whole word model for the whole word of online recognition or be used for
In the whole word model of the entire word of online recognition.
Preferably, the confidence level of the preceding strokes obtained with rotation angle determining device can be integrated into the remaining stroke of identification
During.
Arrangement described above is for implementing the exemplary and/or excellent of the handwriting tracks recognition methods described in the disclosure
The device of choosing.These devices can be hardware cell (such as field programmable gate array, digital signal processor, special integrated electricity
Road or computer etc.) and/or software service (such as computer-readable program).It does not describe at large above each for implementing
The device of step.As long as however, there is the step of some processing of execution, so that it may to be useful for implementing the corresponding device of same processing
(by hardware and/or software implementation).It is limited by all combinations of described step and device corresponding with these steps
Technical solution be all included in present disclosure, as long as these technical solutions that they are constituted are complete and can
Application.
In addition, the above equipment being made of various devices can be incorporated into such as computer etc as function module
In hardware device.Other than these function modules, computer is it is of course possible to having other hardware or software component.
Inventor is tested coming from 50 Japanese 3991 Hanzi specimens.These samples are by 255 Chinese characters
Character forms, and each character includes 5 starting radicals " yarn, speech, stone, standing grain, one of Fu ".In the case that no rotary, it is labeled as
"T0".Then other 6 rotary test samples are generated by the way that each sample is rotated 5 °, 10 °, 15 °, 20 °, 25 ° and 30 ° from T0
This, and it is marked as T5、T10、T15、T20、T25And T30。
Basic engine in following table is the disaggregated classification device based on HMM by master sample training.In addition, inventor uses
Rough sort device of the grader being of relatively low cost as document 1.The performance of the rough sort device is as follows：Speed 30ms/ characters (
On the circuit board of 400MHz ARM), dictionary size 200KB and accuracy rate 88.6%.
Table 1：Accuracy rate variation before and after rotation correction
Velocity variations before and after 2 rotation correction of table
Change in size before and after 3 rotation correction of table
| Moulded dimension (KB) | |
| Document 1 | 697 |
| The present invention | 513 |
From table 1 to table 3 as it can be seen that the application reduces moulded dimension compared with prior art.In addition, although in document 1
Rough sort device is comparatively faster identifier, but due to repeatedly identifying so not being avoided that additional take.Therefore, it is not revolving
In the case of the input turned, the recognition speed of the normal present invention is more faster than document 1.Even if last rotating sizable feelings
The present invention can also realize higher accuracy rate under condition.
There can be many applications according to the handwriting tracks identification equipment of the application.Such as it can be applied to but be not limited to intelligence
The handwriting input device such as mobile phone and tablet computer.Fig. 9 is shown according to the exemplary of the handwriting input device 900 of one embodiment
The block diagram of configuration.Handwriting input device 900 may include：Input unit 901 is configured as receiving the input of handwriting tracks, example
Such as touching sensitive screen；And above-mentioned handwriting tracks identification equipment 800, it is configured as the handwriting tracks of identification input.
Figure 10 is the block diagram for showing to implement the hardware configuration of the computer system of embodiments herein.
As shown in Figure 10, computer system includes the processing unit 1001 connected via system bus 1004, read-only deposits
Reservoir 1002, random access memory 1003, input/output interface 1005, input unit 1006, output unit 1007, storage
Unit 1008, communication unit 1007 and driver 1010.Program can be previously recorded in be situated between as record built-in in computer
In the ROM (read-only memory) 1002 or storage unit 1008 of matter.Alternatively, program can store (record) in removable media
In 1011.Herein, removable media 1011 includes such as floppy disk, CD-ROM (compact disk read-only memory), MO (magnetic
Light) disk, DVD (digital versatile disc), disk, semiconductor memory etc..
Input unit 1006 is configured with keyboard, mouse, microphone etc..In addition, output unit 1007 is configured with LCD (liquid crystal
Display), loud speaker etc..
In addition, program is installed to computer from above-mentioned removable media 1011 except through driver 1010
Except configuration, program can be downloaded to computer to be mounted on built-in storage unit 1008 by communication network or broadcasting network
In.In other words, can for example wirelessly by the satellite for digital satellite broadcasting from download point to computer or with
Wired mode transmits program by the network of LAN (LAN) or internet etc. to computer.
If the inputs order such as manipulating by the user of input unit 1006 via input/output interface 1005,
CPU1001 executes the program stored in ROM1002 according to order.Alternatively, CPU1001 is the journey stored in storage unit 1008
Sequence load is on RAM1003 to execute program.
Therefore, CPU1001 can perform certain processing according to above-mentioned flow chart or pass through above-mentioned frame
The processing that the configuration of figure executes.Next, if it is necessary, then CPU1001 allows the result of processing for example to pass through input/output
Interface 1005 is exported from output unit 1007, is transmitted from communication unit 1009, record etc. in storage unit 1008.
In addition, program can be executed by a computer (processor).In addition, program can be by multiple computers to be distributed
The mode of formula is handled.Furthermore it is possible to which program transportation is executed to remote computer.
Computer system shown in Fig. 10 be merely illustrative and be never intended to the application, its application or purposes into
Row any restrictions.
Computer system shown in Fig. 10 can be incorporated in any embodiment, can be used as stand-alone computer, or also may be used
As the processing system in equipment, one or more unnecessary components can be removed, one or more can also be added to
Multiple additional components.
Can the present processes and equipment be implemented in many ways.For example, can by software, hardware, firmware,
Or any combination thereof implement the present processes and equipment.The order of above-mentioned method and step is merely illustrative, the application
Method and step be not limited to order described in detail above, unless otherwise clearly stating.In addition, in some embodiments
In, the application can also be implemented as recording program in the recording medium comprising for realizing according to the present processes
Machine readable instructions.Thus, the application also covers storage for realizing according to the recording medium of the program of the present processes.
Although some specific implementation modes of the application are described in detail by example, those skilled in the art answer
Work as understanding, above-mentioned example is merely illustrative without limiting the scope of the present application.It should be appreciated by those skilled in the art that above-mentioned reality
Example is applied to be changed without departing from scope of the present application and essence.Scope of the present application is limited by the attached claims
's.
Claims (20)
1. a kind of handwriting tracks recognition methods, including：
First obtaining step obtains the tracing point of the preceding strokes for the handwriting tracks to be entered；
Rotation angle determines step, and the rotation angle for the rotation angle that can detect the preceding strokes is used based on the tracing point
Detection model determines the rotation angle of the handwriting tracks；
Second obtaining step obtains the remaining stroke of the handwriting tracks；And
Identification step identifies the handwriting tracks based on identified rotation angle.
2. according to the method described in claim 1, wherein, the preceding strokes is the part that can be shared by the character of character set
Stroke or a part of stroke shared by the word of set of words.
3. according to the method described in claim 1, wherein, rotation angle determines that step includes：
Tracing point obtaining step obtains current trace points；
Decoding step, using the rotation angle detection model to including that the partial traces until current trace points solve
Code；
Judgment step, judges whether current trace points reach the end tracing point of the preceding strokes；And
It determines step, determines rotation angle corresponding with decoding result as the hand in the case where judging result is affirmative
Write the rotation angle of track.
4. according to the method described in claim 3, wherein,
Decoding step include with based on hidden Markov model come the rotation angle detection model trained to including to working as front rail
Partial traces until mark point are decoded to obtain optimum state transfer path；
Judgment step includes judging whether optimum state transfer path reaches end state；And
Determine that step determines that rotation angle corresponding with optimum state transfer path is made in the case where judging result is affirmative
For the rotation angle of the handwriting tracks.
5. according to the method described in claim 3, wherein,
Decoding step include with based on DTW models come the rotation angle detection model trained to including until current trace points
Partial traces are decoded to obtain best matching path；
Judgment step includes judging whether best matching path reaches end node；And
It is determining rotation angle work corresponding with best matching path in the case of affirmative to determine that step is included in judging result
For the rotation angle of the handwriting tracks.
6. method according to claim 1 or 2, wherein rotation angle detection model can be integrated into for knowing online
In whole word model in the whole word model of not whole word or for the entire word of online recognition.
7. according to the method described in claim 6, wherein, the confidence level energy of the preceding strokes obtained in step is determined from rotation angle
During being enough integrated into the remaining stroke of identification.
8. according to the method described in claim 1, wherein, identification step includes：
Step is corrected, at least part of handwriting tracks is corrected based on identified rotation angle,
Identification step after correction, at least part of the handwriting tracks after identification correction.
9. according to the method described in claim 8, wherein, in the case where handwriting recognizer is online recognition device, to be repaired
And identify this be the remaining stroke obtained by the second obtaining step at least partially.
10. according to the method described in claim 8, wherein, in the case where handwriting recognizer is offline recognizer, to be repaired
And the part identified is entire handwriting tracks.
11. according to the method described in claim 1, wherein, identification step includes：
Model selects step, selects rotation identification model corresponding with the rotation angle of identified handwriting tracks,
Identification step after selection identifies at least part of handwriting tracks with selected rotation identification model.
12. according to the method for claim 11, wherein in the case where handwriting recognizer is offline recognizer, to identify
Part be entire handwriting tracks.
13. according to the method for claim 11, wherein in the case where handwriting recognizer is online recognition device, to identify
Part be the remaining stroke or entire handwriting tracks that are obtained by the second obtaining step stroke.
14. a kind of handwriting tracks method of adjustment, including：
First obtaining step obtains the tracing point of the preceding strokes for the handwriting tracks to be entered；
Rotation angle determines step, and the rotation angle of the rotation angle for detecting the preceding strokes is used based on the tracing point
Detection model determines the rotation angle of the handwriting tracks；
Second obtaining step obtains the remaining stroke of the handwriting tracks；And
Set-up procedure adjusts the handwriting tracks based on identified rotation angle.
15. according to the method for claim 14, wherein the preceding strokes is can be shared by the character of character set one
Divide stroke or by the shared a part of stroke of the word of set of words.
16. according to the method for claim 14, wherein rotation angle detection model can be integrated into for online recognition
In whole word model in the whole word model of whole word or for the entire word of online recognition.
17. according to the method for claim 16, wherein in the set-up procedure, can adjust the remaining stroke or
All strokes of the entire handwriting tracks of person.
18. according to the method described in claim 2, in the case where the handwriting tracks are East Asia characters, the handwriting tracks
Preceding strokes be radical.
19. a kind of handwriting tracks identification equipment, including：
First acquisition device is configured as obtaining the tracing point of the preceding strokes for the handwriting tracks to be entered；
Rotation angle determining device is configured as using the rotation angle that can detect the preceding strokes based on the tracing point
Rotation angle detection model determines the rotation angle of the handwriting tracks；
Second acquisition device is configured as obtaining the remaining stroke of the handwriting tracks；And
Identification device is configured as identifying the handwriting tracks based on identified rotation angle.
20. a kind of handwriting input device, including：
Input unit is configured as receiving the input of handwriting tracks；And
Handwriting tracks identification equipment as claimed in claim 19.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410182172.5A CN105095830B (en) | 2014-04-28 | 2014-04-28 | Handwriting tracks recognition methods, handwriting tracks identification equipment and handwriting input device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410182172.5A CN105095830B (en) | 2014-04-28 | 2014-04-28 | Handwriting tracks recognition methods, handwriting tracks identification equipment and handwriting input device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105095830A CN105095830A (en) | 2015-11-25 |
| CN105095830B true CN105095830B (en) | 2018-11-02 |
Family
ID=54576225
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410182172.5A Active CN105095830B (en) | 2014-04-28 | 2014-04-28 | Handwriting tracks recognition methods, handwriting tracks identification equipment and handwriting input device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105095830B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110647245A (en) * | 2018-06-26 | 2020-01-03 | 深圳市美好创亿医疗科技有限公司 | Handwriting input method based on DTW algorithm |
| CN109635772B (en) * | 2018-12-20 | 2020-12-01 | 广东小天才科技有限公司 | Dictation content correcting method and electronic equipment |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101441549A (en) * | 2008-11-25 | 2009-05-27 | 宇龙计算机通信科技(深圳)有限公司 | Character recognition method, system and electric terminal |
| CN101627398A (en) * | 2007-03-06 | 2010-01-13 | 微软公司 | Radical-based hmm modeling for handwriten east asian characters |
| CN102103693A (en) * | 2011-03-23 | 2011-06-22 | 安徽科大讯飞信息科技股份有限公司 | Method for identifying handwriting |
| CN102937837A (en) * | 2012-08-10 | 2013-02-20 | 上海驿创信息技术有限公司 | Method for inputting words on the basis of incomplete recognition quickly |
-
2014
- 2014-04-28 CN CN201410182172.5A patent/CN105095830B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101627398A (en) * | 2007-03-06 | 2010-01-13 | 微软公司 | Radical-based hmm modeling for handwriten east asian characters |
| CN101441549A (en) * | 2008-11-25 | 2009-05-27 | 宇龙计算机通信科技(深圳)有限公司 | Character recognition method, system and electric terminal |
| CN102103693A (en) * | 2011-03-23 | 2011-06-22 | 安徽科大讯飞信息科技股份有限公司 | Method for identifying handwriting |
| CN102937837A (en) * | 2012-08-10 | 2013-02-20 | 上海驿创信息技术有限公司 | Method for inputting words on the basis of incomplete recognition quickly |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105095830A (en) | 2015-11-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2020221298A1 (en) | Text detection model training method and apparatus, text region determination method and apparatus, and text content determination method and apparatus | |
| CN102855082B (en) | Character recognition for overlay text user input | |
| WO2017020723A1 (en) | Character segmentation method and device and electronic device | |
| CN104008363B (en) | The collection of detection, standardization and the ONLINE RECOGNITION of handwriting tracks and abnormal radical | |
| KR102326395B1 (en) | System and method and product for recognizing multiple object inputs | |
| US20080130996A1 (en) | Method For Character Recognition | |
| CN111103317B (en) | System and method for scanning wafer | |
| US9711117B2 (en) | Method and apparatus for recognising music symbols | |
| CN107729865A (en) | A kind of handwritten form mathematical formulae identified off-line method and system | |
| Julca-Aguilar et al. | Symbol detection in online handwritten graphics using faster R-CNN | |
| JPH0792815B2 (en) | Pattern recognition method | |
| US11449706B2 (en) | Information processing method and information processing system | |
| KR20210017090A (en) | Method and electronic device for converting handwriting input to text | |
| CN103793926A (en) | Target tracking method based on sample reselecting | |
| US20190205589A1 (en) | Latent fingerprint ridge flow map improvement | |
| KR20080100180A (en) | A method for additive character recognition and an apparatus thereof | |
| CN105095830B (en) | Handwriting tracks recognition methods, handwriting tracks identification equipment and handwriting input device | |
| US10755074B2 (en) | Latent fingerprint pattern estimation | |
| Panda et al. | Odia offline typewritten character recognition using template matching with unicode mapping | |
| CN110192250B (en) | Method and system for estimating symbol sequence in speech | |
| US10296788B1 (en) | Method and system for processing candidate strings detected in an image to identify a match of a model string in the image | |
| CN115114887A (en) | Online handwritten text line identification method | |
| EP3785145B1 (en) | System and method for automatic language detection for handwritten text | |
| CN111291820A (en) | Target detection method combining positioning information and classification information | |
| SE528063C2 (en) | Method and apparatus for segmentation-based recognition |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |