CN103577843B - A kind of aerial hand-written character string recognition methods - Google Patents
A kind of aerial hand-written character string recognition methods Download PDFInfo
- Publication number
- CN103577843B CN103577843B CN201310597539.5A CN201310597539A CN103577843B CN 103577843 B CN103577843 B CN 103577843B CN 201310597539 A CN201310597539 A CN 201310597539A CN 103577843 B CN103577843 B CN 103577843B
- Authority
- CN
- China
- Prior art keywords
- character string
- stroke section
- identification
- finger
- module
- 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
- 238000000034 method Methods 0.000 title claims abstract description 32
- 238000000605 extraction Methods 0.000 claims abstract description 22
- 230000009471 action Effects 0.000 claims abstract description 6
- 238000013210 evaluation model Methods 0.000 claims description 36
- 238000012217 deletion Methods 0.000 claims description 16
- 230000037430 deletion Effects 0.000 claims description 16
- 239000012634 fragment Substances 0.000 claims description 12
- 238000011156 evaluation Methods 0.000 claims description 11
- 238000012549 training Methods 0.000 claims description 9
- 238000004422 calculation algorithm Methods 0.000 claims description 5
- 238000012706 support-vector machine Methods 0.000 claims description 5
- 230000002452 interceptive effect Effects 0.000 abstract description 3
- 230000006870 function Effects 0.000 description 7
- 230000008901 benefit Effects 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 230000005484 gravity Effects 0.000 description 3
- 238000013459 approach Methods 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 238000005286 illumination Methods 0.000 description 2
- 230000010354 integration Effects 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000019771 cognition Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
Landscapes
- Character Discrimination (AREA)
Abstract
The invention discloses a kind of aerial hand-written character string identification system and method。This system includes: gesture recognition module, is connected with finger track extraction module, starts the gesture writing and terminating to write for identification marking；Finger track extraction module, is connected with pretreatment module, for the hand-written action of input carries out extraction and the record of finger trace；Pretreatment module, is connected with aerial hand-written character string identification module, for finger track data is carried out pretreatment；Character string identification module, is connected with pretreatment module, for finger track data is carried out character string identification。The invention also discloses a kind of aerial hand-written character string recognition methods。The ways of writing novelty that the present invention supports is convenient, identify accurately rapidly, outside traditional handwriting mode, provide one hommization more and intelligentized handwriting input mode for people, be widely portable in man-machine interactive system, such as game operation, TV control, teaching system etc.。
Description
Technical field
The invention belongs to computer utility and mode identification technology, especially a kind of aerial hand-written character string recognition methods。
Background technology
Word, as a kind of Information Communication commonly used and media of communication, plays important role in man-machine interactive system。Now widely used character input modes includes: keyboard, touch screen, handwriting pad etc.。These input modes have respective advantage, but there is also respective deficiency, and as keyboard is subject to the restriction of size and number of keys, the equipment such as touch screen and handwriting pad is subject to the restriction of volume size, to write region limited etc.。Therefore, designing character input modes more natural, convenient, efficient is an important research direction。In recent years, people devise the character input modes of a class view-based access control model gesture by handwriting identification and computer vision system being combined, one of which approach is that people normally make paper using pen write to then pass through and write ink marks on camera collection paper, it is identified as word, concrete grammar is referred to document " M.E.Munich, P.Perona, Visualinputforpen-basedcomputers, IEEETrans.PatternAnalysisandMachineIntelligence, 24 (3): 313-328, 2002 ", but this method has nevertheless suffered from the restriction of external condition (size etc. of paper), there is certain limitation。Another kind of approach is then aloft write the movement locus of the Real-time Collection fingers such as virtual word, photographic head or motion sensor such as Kinect either directly through finger, is then word by track identification。This character input modes is owing to not by the restriction writing region and ways of writing, comparing other character input modes more natural and convenient。But, owing to being difficult to judge starting writing/starting to write in writing in such systems, cause that between word, cutting is very difficult。So, current this kind of system can only identify single word, is specifically referred to document " Jin Lianwen, Ding Kai, Yan Hanyu, the virtual character recognition method of a kind of visible detection, Chinese Patent Application No. 200810029420, publication number CN10132029A "。
The present invention proposes a kind of aerial hand-written character string recognition methods, and user uses finger aloft to write, and gathers finger motion track by photographic head or motion sensor, is then character string by track identification。Lack, in aerial hand-written character string, information of starting writing/start to write, cause that character string exists the problem of unnecessary stroke after crossing cutting, method proposes the unitary of stroke section/binary and delete evaluation model for calculating the deletion cost of stroke section。Under the framework of integrated cutting-identification, delete evaluation model by the output of effective integration Character recognizer, character geometric model and stroke section and can effectively remove unnecessary stroke section, thus improving the accuracy rate of character cutting and identification。
Summary of the invention
Based on above-mentioned problems of the prior art, the invention provides a kind of aerial hand-written character string recognition methods, the method is based on Gesture Recognition, finger tracking technology and hand-written character string identification technology, aerial hand-written character string can be identified rapidly and accurately, solve the character string identification problem inputted by gesture。The present invention is directed to aerial hand-written character string and there is the feature connecting pen between word, candidate's cutting-identification path evaluation function has merged stroke section and has deleted cost, it is possible to effectively removed unnecessary stroke；And based on the unrelated path evaluation criterion of character cutting length, it is possible to effectively utilize dynamic programming algorithm fast search optimal path to be identified result；Additionally, under integrated cutting-identification framework, delete evaluation model by effective integration character recognition confidence level, character geometric model and stroke section, further increase the precision of aerial hand-written character string identification。
According to an aspect of the present invention, disclosing a kind of aerial hand-written character string identification system, this system includes gesture recognition module, finger track extraction module, pretreatment module, character string identification module, wherein:
Described gesture recognition module is connected with described finger track extraction module, starts the gesture writing and terminating to write for identification marking；
Described finger track extraction module is connected with described pretreatment module, and for after described gesture recognition module recognizes the gesture starting to write, the hand-written action for inputting carries out extraction and the record of finger trace；
Described pretreatment module is connected with described aerial hand-written character string identification module, carries out pretreatment for the finger trace data that described finger track extraction module is extracted；
Described character string identification module is connected with described pretreatment module, for the finger trace through pretreatment is carried out character string identification。
According to a further aspect in the invention, also disclosing a kind of aerial hand-written character string recognition methods, the method comprises the following steps:
Step 201, detects whether there is the gesture that mark starts to write；
Step 202, if detecting, mark starts the gesture write, then forward step 203 to, otherwise returns step 201；
Step 203, real-time tracking also gathers the track that finger is aloft write, and preserves the finger trace data collected；
Step 204, detects whether there is the gesture that mark terminates to write, if it is detected preserve described finger trace data, forwards step 205 to, otherwise returns step 203；
Step 205, carries out pretreatment to preserving the finger trace data obtained；
Step 206, carries out character string identification output string recognition result based on the finger trace data through pretreatment；
Step 207, if detecting that again mark starts the gesture write, then returns step 201, otherwise terminates。
Aerial hand-written character string recognition methods provided by the invention, it provides the benefit that:
(1) user aloft uses finger Free Writing, and make to use gesture the beginning and end that control to write, gathers finger trace by photographic head or motion sensor, is character string by track identification, conventional book of comparing WriteMode nature more and hommization；
(2) based on character string identification, more convenient than the input mode based on individual character identification；
(3) spending the cutting stage, it is contemplated that the feature of aerial hand-written character string, effectively possible company's pen is being disconnected, do not affect follow-up identification simultaneously；
(4) there is the feature of unnecessary stroke after crossing cutting for aerial hand-written character string, it is proposed that stroke section deletes evaluation model, delete evaluation model including unitary and binary deletes evaluation model, for calculating the deletion cost of stroke section；
(5) effectively merged Character recognizer, character geometric model and stroke section unitary/binary and deleted evaluation model so that character cutting and accuracy of identification have been improved；
(6) Character recognizer of the present invention have employed and has the Nearest prototype grader that discrimination is high and computation complexity is low, so that the present invention disclosure satisfy that in practical application, memory space is little and the requirement of quick identification to taking。
In sum, the aerial hand-written character string identification system and method for the present invention, ways of writing novelty is convenient, identifies accurately rapidly, provides one hommization more and intelligentized handwriting input mode for people outside traditional handwriting mode。Disclosed by the invention is widely suitable in man-machine interactive system, such as game operation, TV control, teaching system etc.。
Accompanying drawing explanation
Fig. 1 is the structural representation of the present invention aerial hand-written character string identification system；
Fig. 2 is the flow chart of the aerial hand-written character string recognition methods of the present invention；
Fig. 3 is the example carrying out hand-written character string identification according to the present invention。
Detailed description of the invention
For making the object, technical solutions and advantages of the present invention clearly understand, below in conjunction with specific embodiment, and with reference to accompanying drawing, the present invention is described in more detail。
Fig. 1 is the structural representation of the present invention aerial hand-written character string identification system, as it is shown in figure 1, described system includes gesture recognition module, finger track extraction module, pretreatment module, character string identification module, wherein:
Described gesture recognition module is connected with described finger track extraction module, starts the gesture writing and terminating to write for identification marking；
In an embodiment of the present invention, respectively define gesture for " starting to write " and " end is write " the two state and trigger event: in the face of photographic head, the action " stretching out palm " represents and starts to write, the action " holding palm " represents that end is once write, when described gesture recognition module recognizes both gestures, corresponding state will be entered into。Both gestures of the present embodiment definition are simple, efficient, it is simple to user grasps and uses。
Described finger track extraction module is connected with described pretreatment module, and for after described gesture recognition module recognizes the gesture starting to write, the hand-written action for inputting carries out extraction and the record of finger trace；
Described gesture recognition module recognizes and starts after the gesture write enters and write state, the finger of described finger track extraction module identification labelling user writing, then real-time tracking record the track data that finger aloft moves in whole writing process, system time including finger (X, Y) coordinate in space and correspondence。When described gesture recognition module detects the gesture of " holding palm ", once writing end, the finger trace data of described identification system keeping records, for follow-up character string identification。
Described pretreatment module is connected with described aerial hand-written character string identification module, carries out pretreatment for the finger trace data extracted for described finger track extraction module；
The purpose of described pretreatment is to reduce the noise in finger trace data, to facilitate follow-up character string identification。Described pretreatment includes: first described finger trace data are carried out smooth operation, to reduce the impact that tracing point position is produced by the factor such as illumination variation in finger tracking, finger shake；Subsequently, the redundant points in described finger trace data is removed so that the same position in described finger trace only retains a tracing point。
Described character string identification module is connected with described pretreatment module, for carrying out character string identification for the finger trace through pretreatment。
Pretreated finger trace data are identified as word by described character string identification module。In an embodiment of the present invention, the identification of described character string is based on candidate's cutting-identification grid, first finger trace data were carried out during identification cutting and obtained a series of basic fragment, it is merged into candidate characters again by adjacent one or more basic fragment, being identified obtaining candidate's cutting-identification grid to candidate characters with Character recognizer, the cutting-identification path finally searching for optimum within a grid obtains final character cutting and recognition result。
Fig. 2 is the flow chart of the aerial hand-written character string recognition methods of the present invention, as in figure 2 it is shown, said method comprising the steps of:
Step 201, detects whether there is the gesture that mark starts to write；
In an embodiment of the present invention, the definition of gesture that mark starts to write is " stretching out palm "。
Step 202, if detecting, mark starts the gesture write, then show that user prepares written character string, forward step 203 to, otherwise returns step 201；
Step 203, real-time tracking also gathers the track that finger is aloft write, and preserves the finger trace data collected；
User by finger can aloft Free Writing, by photographic head or motion sensor real-time tracking and gather the track that finger is aloft write。Described finger trace data include: the current finger point (X, the Y) coordinate figure in space and current system time, described finger trace data are for character string identification afterwards。
Step 204, detects whether there is the gesture that mark terminates to write, if it is detected show that user terminates once to write, preserves described finger trace data, forward step 205 to, otherwise returns step 203；
In an embodiment of the present invention, the definition of gesture that mark terminates to write is " holding palm "。
Step 205, carries out pretreatment to preserving the finger trace data obtained；
Pretreatment can reduce effect of noise in track gatherer process, thus improving the precision of character string identification, described pretreatment specifically includes:
1) smooth operation carries out for described finger trace data, remove noise spot, by each tracing point with on average the replacing of all tracing points in certain neighborhood about, so can reduce impact as produced in illumination variation, finger shake etc.；
2) remove the redundant points in described finger trace data, if there is multiple tracing point at the same position place namely on track, then only retain a tracing point。
Step 206, carries out character string identification output string recognition result based on the finger trace data through pretreatment；
Described step 206 further includes steps of
Finger trace data through pretreatment were carried out cutting based on multiple cut-offs of crossing by step 2061, thus obtaining a series of stroke block, i.e. and basic fragment；
Wherein, described cut-off of crossing is made up of following 3 class points:
1) Y coordinate Local Extremum: comprise Y coordinate Local modulus maxima and minimum point。The Y coordinate of each tracing point with all tracing points in certain distance before and after it is compared, if the y-coordinate of this point is maximum or minima, then selects this point to cross cut-off for candidate。
2) local minimum point of floor projection value: the method first passing through linear interpolation fills the white space between each tracing point, then project in the horizontal direction, the projection value at coordinate Y place is all abscissas is the number of the tracing point of X, and choosing the point that projection value is local minimum is candidate's cut-off。
3) long stroke section joins end to end the maximum point of the distance of straight line to stroke section: obtaining above-mentioned 1), 2) after class candidate's cut-off, choose the stroke section (here threshold value be set to character string estimate 1.2 times of height) more than a certain threshold value of the path length between adjacent cut-off, calculate and long stroke section joins end to end the maximum point of the distance of straight line as candidate's cut-off to this stroke section。
The described here method obtaining cut-off takes full advantage of features of shape and the time feature of aerial hand-written character string, the cut-off of crossing obtained by character in even pen place disconnection, can not affect character recognition (because the fracture of stroke does not affect character global shape feature extraction) below simultaneously。
Step 2062, builds candidate's cutting-identification grid based on a series of basic fragment obtained；
Described step 2062 further includes steps of
One or more adjacent basic fragment combination are built candidate characters pattern by step 20621；
In this step, the rule building candidate characters pattern is as follows: first, constitutes the candidate segment number of candidate characters no more than 5, and the basic fragment number candidate characters more than 5 thinks non-character, does not carry out character recognition；Secondly, the width of candidate characters not can exceed that some predetermined threshold, and the width setting candidate characters here can not estimate 2 times of height more than character string。
Wherein, estimating that the method for character string height is: travel through all of tracing point, find maximum and the minima of Y coordinate, namely both differences estimate height as character string。
Step 20622, uses Character recognizer (grader) to distribute candidate characters classification for each candidate characters pattern, thus obtaining candidate's cutting-identification grid。
Owing to aerial hand-written character string is not started writing/starts to write information, therefore there is unnecessary stroke after crossing cutting, if these unnecessary stroke are not deleted and be may result in wrong identification, so in candidate's cutting-identification grid, need to judge that each candidate segment is belonging to a candidate characters or unnecessary stroke, say, that each basic fragment or belong to a candidate characters or be marked as unnecessary stroke。
In this step, the grader of employing is to have higher recognition correct rate and non-character refusal ability, simultaneously amount of storage and all less Nearest prototype grader of amount of calculation；Nearest prototype grader obtains by training on the hand-written character sample have in a large number category label。In prototype grader, each classification has one or more prototype (representing with characteristic vector), during classification, calculates input pattern (representing with characteristic vector) to the distance of each prototype, provides the classification of input pattern apart from minimum prototype；Apart from the minimum candidate categories that multiple classifications are input pattern。
Wherein, feature extraction (the being expressed as characteristic vector) method of candidate characters pattern adopts using partial stroke direction Histogram drawing method conventional in hand script Chinese input equipment character recognition, due to the method that it is openly known, here do not describe in detail, it is referred to document: Cheng-LinLiu, Xiang-DongZhou, OnlineJapanesecharacterrecognitionusingtrajectory-basedn ormalizationanddirectionfeatureextraction, Proc.10thInternationalWorkshoponFrontiersofHandwritingRe cognition, LaBaule, France, 2006, pp.217-222。
Step 2063, in described candidate's cutting-identification grid enterprising line character string identification: calculate the mark in each cutting-identification path according to path evaluation function, obtains optimum cutting-recognition combination with dynamic programming algorithm search, obtains final character string recognition result。
Wherein, described path evaluation function has merged Character recognizer, character geometric model and stroke section deletes evaluation model, can be adjusted the effect size of each several part by weights, and described weights can be set by experience, it is also possible to obtain by training；Meanwhile, use dynamic programming method can obtain the optimal path in multiple cutting-identification path by fast search, namely obtain final character string recognition result。
Described stroke section is deleted evaluation model and is divided into two classes: unitary deletes evaluation model and binary deletes evaluation model, wherein, unitary is deleted evaluation model and is used the linear SVM of two class (SVM) that the geometric properties deleting stroke section is modeled, calculate the deletion cost of stroke section, by observing, character string need the stroke section deleted compared obvious geometric properties with other stroke section, such as their boundary rectangle frame the ratio of width to height is bigger, the slope of tracing point fitting a straight line is less, less and stroke section the center of gravity of floor projection value average is less to the distance at its boundary rectangle frame center, extract four class unitary accordingly and delete geometric properties: the ratio of width to height of the boundary rectangle frame of stroke section, the slope of fitting a straight line, floor projection value average and center of gravity are to the distance at boundary rectangle frame center, described unitary is deleted the parameter of evaluation model and can be obtained with deletion stroke section labeled in a large number and normal stroke section sample training；Binary is deleted evaluation model and is deleted the effective supplementary of evaluation model as unitary, deletes by utilizing the binary geometrical relationship (such as deletion stroke section is usual and its adjacent character degree of overlapping is less) deleting stroke section and its adjacent character pattern can correct part stroke section by mistake。Based on this, binary deletes 11 category features such as the logarithm of distance that geometric properties includes deleting between stroke section and the width ratio of adjacent character each boundary rectangle frame, coboundary/lower boundary/left margin/right margin, the width of boundary rectangle frame overlapping region, catercorner length, the square root of area and the ratio of width to height, described binary is deleted evaluation model and is used two class Linear SVMs that the binary geometric properties of stroke section and its adjacent character pattern is modeled, and stroke section and the training of adjacent character equal samples that its model parameter serviceable indicia is crossed obtain。
Described character geometric model is used for whether evaluating a candidate characters as a normal character。Character geometric model extracts nine features from candidate characters: the X/Y coordinate of character center of gravity, the width of boundary rectangle frame, catercorner length, the square root of area, the logarithm of the ratio of width to height and boundary rectangle frame up/down border are to the distance of character horizontal center line。Then use quadric discriminant function (QDF) to model, give a mark to each character class。
Described character geometric model mark, character recognition mark are deleted with stroke section and together with evaluation model mark, candidate's cutting-identification path are evaluated, it is possible to play the effect improving character cutting and recognition performance。
In this step, the calculation of path score is as follows: be a pair candidate characters pattern (characteristic vector) sequence (O=o by each candidate's cutting-identification path representation1,L,on) and candidate characters classification sequence (C=c1,L,cn), the wherein length of n delegated path, the namely number of candidate characters, owing to would be likely to occur unnecessary stroke, so candidate characters classification ciIt is likely to belong to significant character category set (i.e. ci∈{ω1,L,ωM, M is the character class number that grader determines) or belong to idle character classification (ci=ω0, i.e. unnecessary stroke)；
Definition D (O, C) is path evaluation function, then optimum cutting-identification path (O*,C*) it is:
Path evaluation function D (O, C) has merged Character recognizer, character geometric model and stroke section deletes evaluation model。It is R that the set of stroke section is deleted in definition on a path candidate, and not deleting stroke section set, to be N, D (O, C) be:
In above formula, D (oi,ci) andRepresent Character recognizer output distance and character geometric model output distance respectively；It it is the output distance of unitary stroke section deletion evaluation model； With Being the output distance of binary deletion evaluation model, both measure deletion stroke section o respectivelyiAdjacent strokes section o with iti-1,oi+1Between binary geometrical relationship；{ λ1,λ2,λ3It is three weights, it is used for balancing difference model roles in D (O, C), can experience set, it is also possible to obtain by training；KiIt is constitute candidate characters ciStroke block number；Unrelated with cutting route length (n) to identifying that mark weighting makes to be added total item in path evaluation function by stroke block number, thus available dynamic programming method fast search obtains optimal path。
Step 207, if detecting that again mark starts the gesture write, then returns step 201, otherwise terminates。
The example of hand-written character string identification is carried out as shown in Figure 3 according to the present invention, Figure 30 1 is the initial trace file reading an aerial hand-written character string the result it shown on computer screen, Figure 30 2 is the pre-processed results to character string shown in Figure 30 1, Figure 30 3 is that the character string to Figure 30 2 crosses cutting result, scheme the position that position was cut-off place of short-and-medium vertical line mark, Figure 30 4 is that character string crosses candidate's cutting-identification grid example after cutting, the top display candidate characters of each square frame, square frame bottom is two candidate characters classification results, symbol "×" represents that stroke section is unnecessary stroke, should be deleted, being of thick line mark searches for, by dynamic programming algorithm, the optimal path obtained。From example it can be seen that the present invention can effectively identify aerial hand-written character string。
Particular embodiments described above; the purpose of the present invention, technical scheme and beneficial effect have been further described; it is it should be understood that; the foregoing is only specific embodiments of the invention; it is not limited to the present invention; all within the spirit and principles in the present invention, any amendment of making, equivalent replacement, improvement etc., should be included within protection scope of the present invention。
Claims (6)
1. an aerial hand-written character string identification system, it is characterised in that this system includes gesture recognition module, finger track extraction module, pretreatment module, character string identification module, wherein:
Described gesture recognition module is connected with described finger track extraction module, starts the gesture writing and terminating to write for identification marking；
Described finger track extraction module is connected with described pretreatment module, and for after described gesture recognition module recognizes the gesture starting to write, the hand-written action for inputting carries out extraction and the record of finger trace；
Described pretreatment module is connected with described character string identification module, carries out pretreatment for the finger trace data that described finger track extraction module is extracted；
Described character string identification module is connected with described pretreatment module, for the finger trace data through pretreatment are carried out character string identification；
Wherein, described character string identification module is carried out as follows character string identification:
Based on multiple cut-offs of crossing, the finger trace data through pretreatment were carried out cutting, thus obtaining a series of stroke block, i.e. basic fragment；
Candidate's cutting-identification grid is built based on a series of basic fragment obtained；
In described candidate's cutting-identification grid enterprising line character string identification: calculate the mark in each cutting-identification path according to path evaluation function, obtain optimum cutting-recognition combination with dynamic programming algorithm search, obtain final character string recognition result；
Wherein, described path evaluation function includes Character recognizer, character geometric model and stroke section and deletes evaluation model, and described stroke section is deleted evaluation model and included unitary deletion evaluation model and binary deletion evaluation model；Described unitary is deleted evaluation model and is used the two linear SVMs of class that the geometric properties deleting stroke section is modeled, and calculates the deletion cost of stroke section；Described unitary is deleted the parameter of evaluation model and is used deletion stroke section labeled in a large number and normal stroke section sample training to obtain；Binary is deleted evaluation model and is deleted by utilizing the binary geometrical relationship deleting stroke section and its adjacent character pattern to correct part stroke section by mistake；Described binary is deleted evaluation model and is used two class linear SVMs that the binary geometric properties of stroke section and its adjacent character pattern is modeled, and its model parameter labeled stroke section and adjacent character sample training obtain。
2. system according to claim 1, it is characterised in that the finger trace data that described finger track extraction module is extracted include the system time of finger coordinate figure in space and correspondence。
3. system according to claim 1, it is characterised in that described pretreatment includes: for the smooth operation of finger trace data, and remove the redundant points in described finger trace data。
4. an aerial hand-written character string recognition methods, it is characterised in that the method comprises the following steps:
Step 201, detects whether there is the gesture that mark starts to write；
Step 202, if detecting, mark starts the gesture write, then forward step 203 to, otherwise returns step 201；
Step 203, real-time tracking also gathers the track that finger is aloft write, and preserves the finger trace data collected；
Step 204, detects whether there is the gesture that mark terminates to write, if it is detected preserve described finger trace data, forwards step 205 to, otherwise returns step 203；
Step 205, carries out pretreatment to preserving the finger trace data obtained；
Step 206, carries out character string identification output string recognition result based on the finger trace data through pretreatment；
Step 207, if detecting that again mark starts the gesture write, then returns step 201, otherwise terminates；
Described step 206 further includes steps of
Finger trace data through pretreatment were carried out cutting based on multiple cut-offs of crossing by step 2061, thus obtaining a series of stroke block, i.e. and basic fragment；
Step 2062, builds candidate's cutting-identification grid based on a series of basic fragment obtained；
Step 2063, in described candidate's cutting-identification grid enterprising line character string identification: calculate the mark in each cutting-identification path according to path evaluation function, obtains optimum cutting-recognition combination with dynamic programming algorithm search, obtains final character string recognition result；
Wherein, described path evaluation function includes Character recognizer, character geometric model and stroke section and deletes evaluation model, and described stroke section is deleted evaluation model and included unitary deletion evaluation model and binary deletion evaluation model；Described unitary is deleted evaluation model and is used the two linear SVMs of class that the geometric properties deleting stroke section is modeled, and calculates the deletion cost of stroke section；Described unitary is deleted the parameter of evaluation model and is used deletion stroke section labeled in a large number and normal stroke section sample training to obtain；Binary is deleted evaluation model and is deleted by utilizing the binary geometrical relationship deleting stroke section and its adjacent character pattern to correct part stroke section by mistake；Described binary is deleted evaluation model and is used two class linear SVMs that the binary geometric properties of stroke section and its adjacent character pattern is modeled, and its model parameter labeled stroke section and adjacent character sample training obtain。
5. method according to claim 4, it is characterised in that described cut-off of crossing includes: the Local Extremum of Y coordinate, joins end to end the maximum point of the distance of straight line to stroke section in the local minimum point of floor projection value and long stroke section。
6. method according to claim 4, it is characterised in that described step 2062 further includes steps of
One or more adjacent basic fragment combination are built candidate characters pattern by step 20621；
Step 20622, uses Character recognizer, i.e. grader, distributes candidate characters classification for each candidate characters pattern, thus obtaining candidate's cutting-identification grid。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310597539.5A CN103577843B (en) | 2013-11-22 | 2013-11-22 | A kind of aerial hand-written character string recognition methods |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310597539.5A CN103577843B (en) | 2013-11-22 | 2013-11-22 | A kind of aerial hand-written character string recognition methods |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103577843A CN103577843A (en) | 2014-02-12 |
| CN103577843B true CN103577843B (en) | 2016-06-22 |
Family
ID=50049591
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310597539.5A Active CN103577843B (en) | 2013-11-22 | 2013-11-22 | A kind of aerial hand-written character string recognition methods |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103577843B (en) |
Families Citing this family (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104978010A (en) * | 2014-04-03 | 2015-10-14 | 冠捷投资有限公司 | Three-dimensional space handwriting trajectory acquisition method |
| CN103984943B (en) * | 2014-05-30 | 2018-06-19 | 厦门大学 | A kind of scene text recognition methods based on Bayesian probability frame |
| CN105320248B (en) * | 2014-06-03 | 2018-12-07 | 深圳Tcl新技术有限公司 | Aerial gesture input method and device |
| CN105094544B (en) * | 2015-07-16 | 2020-03-03 | 百度在线网络技术（北京）有限公司 | Method and device for acquiring characters |
| CN104951083A (en) * | 2015-07-21 | 2015-09-30 | 石狮市智诚通讯器材贸易有限公司 | Remote gesture input method and input system |
| CN107092902B (en) * | 2016-02-18 | 2021-04-06 | 富士通株式会社 | Character string recognition method and system |
| CN107291215A (en) * | 2016-04-01 | 2017-10-24 | 北京搜狗科技发展有限公司 | A kind of body-sensing input message processing method and device |
| CN107305630B (en) * | 2016-04-25 | 2021-03-19 | 腾讯科技（深圳）有限公司 | Text sequence identification method and device |
| CN106446757A (en) * | 2016-05-20 | 2017-02-22 | 北京九艺同兴科技有限公司 | Human body motion data similarity automatic evaluation method |
| CN106723722B (en) * | 2016-12-26 | 2018-08-07 | 东莞理工学院 | Ring, system for keeping a diary and application method |
| CN107330430B (en) * | 2017-06-27 | 2020-12-04 | 司马大大（北京）智能***有限公司 | Tibetan character recognition device and method |
| CN107562203A (en) * | 2017-09-14 | 2018-01-09 | 北京奇艺世纪科技有限公司 | A kind of input method and device |
| CN107390880A (en) * | 2017-09-15 | 2017-11-24 | 西安建筑科技大学 | One kind is based on the contactless multi-angle input equipment of shadow and input method |
| CN107992792A (en) * | 2017-10-16 | 2018-05-04 | 华南理工大学 | A kind of aerial handwritten Chinese character recognition system and method based on acceleration transducer |
| CN109992124B (en) * | 2018-01-02 | 2024-05-31 | 北京搜狗科技发展有限公司 | Input method, apparatus and machine readable medium |
| CN109344793B (en) * | 2018-10-19 | 2021-03-16 | 北京百度网讯科技有限公司 | Method, apparatus, device and computer readable storage medium for recognizing handwriting in the air |
| CN109493652B (en) * | 2018-11-05 | 2021-12-24 | 广州南洋理工职业学院 | Practice teaching system based on VR technique |
| CN110196635B (en) * | 2019-04-28 | 2020-07-31 | 浙江大学 | Gesture input method based on wearable equipment |
| CN111580649B (en) * | 2020-04-24 | 2023-04-25 | 佛山科学技术学院 | Deep learning-based aerial handwriting interaction method and system |
| CN112215178B (en) * | 2020-10-19 | 2024-05-28 | 南京大学 | Chemical experiment recording system based on pen type interaction |
| CN112462935A (en) * | 2020-11-05 | 2021-03-09 | 深圳市易平方网络科技有限公司 | Text input processing method, device, terminal and medium based on remote gestures |
| CN112686134B (en) * | 2020-12-29 | 2023-12-01 | 科大讯飞股份有限公司 | Handwriting recognition method, handwriting recognition device, electronic equipment and storage medium |
| CN113391703A (en) * | 2021-06-16 | 2021-09-14 | 咙咙信息技术(沈阳)有限公司 | System for operating air writing based on media application |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2009040322A1 (en) * | 2007-09-25 | 2009-04-02 | Continental Automotive Gmbh | Method and apparatus for the contactless input of characters |
| CN101853126A (en) * | 2010-05-12 | 2010-10-06 | 中国科学院自动化研究所 | Real-time identification method for on-line handwriting sentences |
| CN102520790A (en) * | 2011-11-23 | 2012-06-27 | 中兴通讯股份有限公司 | Character input method based on image sensing module, device and terminal |
| CN103150019A (en) * | 2013-03-12 | 2013-06-12 | 深圳市国华识别科技开发有限公司 | Handwriting input system and method |
| CN103226388A (en) * | 2013-04-07 | 2013-07-31 | 华南理工大学 | Kinect-based handwriting method |
-
2013
- 2013-11-22 CN CN201310597539.5A patent/CN103577843B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2009040322A1 (en) * | 2007-09-25 | 2009-04-02 | Continental Automotive Gmbh | Method and apparatus for the contactless input of characters |
| CN101853126A (en) * | 2010-05-12 | 2010-10-06 | 中国科学院自动化研究所 | Real-time identification method for on-line handwriting sentences |
| CN102520790A (en) * | 2011-11-23 | 2012-06-27 | 中兴通讯股份有限公司 | Character input method based on image sensing module, device and terminal |
| CN103150019A (en) * | 2013-03-12 | 2013-06-12 | 深圳市国华识别科技开发有限公司 | Handwriting input system and method |
| CN103226388A (en) * | 2013-04-07 | 2013-07-31 | 华南理工大学 | Kinect-based handwriting method |
Non-Patent Citations (1)
| Title |
|---|
| 基于连笔消除的空间手写字符识别方法;赖英超，曾剑铭，沈海斌;《计算机工程》;20121031;第38卷(第19期);第155-156页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103577843A (en) | 2014-02-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103577843B (en) | A kind of aerial hand-written character string recognition methods | |
| US10878195B2 (en) | Automated extraction of unstructured tables and semantic information from arbitrary documents | |
| CN106354252B (en) | A kind of continuation character gesture track recognition method based on STDW | |
| CN101853126B (en) | Real-time identification method for on-line handwriting sentences | |
| CN103226388B (en) | A kind of handwriting sckeme based on Kinect | |
| CN103065134B (en) | A kind of fingerprint identification device and method with information | |
| US20160343147A1 (en) | Image search system, image search apparatus, and image search method | |
| CN101299236B (en) | Method for recognizing Chinese hand-written phrase | |
| US20140028603A1 (en) | System and method for implementing sliding input of text based upon on-screen soft keyboard on electronic equipment | |
| CN109344793A (en) | Aerial hand-written method, apparatus, equipment and computer readable storage medium for identification | |
| CN101627398B (en) | Radical-based HMM modeling for handwritten east Asian characters | |
| JP2008532176A (en) | Recognition graph | |
| CN103984943A (en) | Scene text identification method based on Bayesian probability frame | |
| CN111950514B (en) | Depth camera-based aerial handwriting recognition system and method | |
| CN103488415B (en) | Personal letter word based on paper medium and pattern recognition system and recognition methods | |
| CN103246891A (en) | Chinese sign language recognition method based on kinect | |
| Joshi et al. | A random forest approach to segmenting and classifying gestures | |
| Li et al. | Hierarchical shape primitive features for online text-independent writer identification | |
| CN109635687A (en) | The Chinese-character text running hand calculated based on timing point set writes automatic Evaluation of Quality method and system | |
| CN103324929A (en) | Handwritten Chinese character recognition method based on substructure learning | |
| Sahoo et al. | Handwritten Bangla word recognition using negative refraction based shape transformation | |
| CN104933408A (en) | Hand gesture recognition method and system | |
| Jin et al. | Visual gesture character string recognition by classification-based segmentation with stroke deletion | |
| CN103488744B (en) | A kind of big data image sorting technique | |
| CN107168633A (en) | A kind of gesture interaction query event construction method based on data dependence |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20190425 Address after: 100080 Floor 11104-2, Building 1, 66 Zhongguancun East Road, Haidian District, Beijing Patentee after: Beijing Zhongke Yueshen Technology Co., Ltd. Address before: 100190 Zhongguancun East Road, Haidian District, Haidian District, Beijing Patentee before: Institute of Automation, Chinese Academy of Sciences |