CN105989341A - Character recognition method and device - Google Patents
Character recognition method and device Download PDFInfo
- Publication number
- CN105989341A CN105989341A CN201510086612.1A CN201510086612A CN105989341A CN 105989341 A CN105989341 A CN 105989341A CN 201510086612 A CN201510086612 A CN 201510086612A CN 105989341 A CN105989341 A CN 105989341A
- Authority
- CN
- China
- Prior art keywords
- language
- communication means
- text
- line
- image
- 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.)
- Pending
Links
Abstract
The invention discloses a character recognition method and a character recognition device. The character recognition method includes the following steps that: a plurality of communicated components are extracted from an image containing a character; the communicated components are classified, so that first language communicated components and/or non-first language communicated components can be generated; the first language communicated components are clustered as a first language text line, and the non-first language communicated components are clustered as a non-first language text line; and a first language character and a non-first language character are recognized from the first language text line and the non-first language text line.
Description
Technical field
The present invention relates to image processing field, be specifically related to identify the word in the address doorplate in image
Method and apparatus.
Background technology
Along with the mobile device with shoot function uses more and more general in our daily life
Time, such as mobile phone, digital camera etc., this photo making us obtain natural scene becomes the most square
Just.Address doorplate is our the most highly important information, and people may utilize mobile device
Shooting comprises the photo of address doorplate, with record or the position sharing oneself.Mark at numerical map
In, need to extract the Word message in the address doorplate in a large amount of photos, be identified mark.
Owing to the quantity of photo is the biggest, it is therefore desirable to replace artificial cognition otherwise with automatic knowledge, with fall
Low workload.
Fig. 1 shows the flow chart of a kind of method identifying address doorplate information.As it is shown in figure 1, root
According to the method 100, can detect from photo and extract doorplate figure after step S110 input photo
As (step S120).Fig. 2 a and Fig. 2 b respectively illustrates the photo of input and extracts from this photo
One example of the doorplate image gone out.As shown in figures 2 a and 2b, by step S120, can be from photo
Detect and extract doorplate image.Then, the doorplate image extracted is carried out Text region, with
Identify literal address (step S130).In the example shown in Fig. 2 a and 2b, may recognize that literary composition
Word address " metallurgical North Road 99 ".Finally, literal address output (step S140) that will be identified,
Automatically mark is identified with realize address.
At present, from photo, detect and extract the technology of doorplate image the most ripe, its correctness and
Processing speed can meet current application demand.But, the doorplate image extracted is carried out literary composition
The technology of word identification does not often reach requirement.On the one hand this be due to the Word message in many doorplates
Layout structure is more complicated, is not easy to carry out Text region.On the other hand owing to generally wrapping in doorplate
Word (e.g., Arabic numerals, English character, Chinese character etc.) containing more than one language, and for
Each character, is required to use the identification engine comprising polyglot character to be identified, due to many
The quantity of language character is very big, and this also have impact on the place utilizing character recognition engine to carry out Text region
Reason speed.
Summary of the invention
In view of this, the present invention proposes a kind of character recognition method and device, with to the literary composition in image
Word information is identified.
According to an aspect of the invention, it is provided a kind of character recognition method, including: from comprising literary composition
The image of word extracts multiple communication means；The plurality of communication means is classified, to generate
One language communication means and/or non-first language communication means；Described first language communication means is gathered
Class is first language line of text, and is non-first language literary composition by described non-first language communication means cluster
One's own profession；And identify first from described first language line of text and described non-first language line of text
Spoken and written languages and non-first language word.
According to a further aspect in the invention, it is provided that a kind of character recognition device, including: extraction unit,
Multiple communication means is extracted from the image comprising word；Taxon, to the plurality of communication means
Classify, to generate first language communication means and/or non-first language communication means；Cluster is single
Unit, is first language line of text by described first language communication means cluster, and by described non-first language
Speech communication means cluster is non-first language line of text；And recognition unit, from described first language literary composition
One's own profession and described non-first language line of text identify first language word and non-first language word.
According to technical scheme provided by the present invention, the Word message in the image comprising word can be entered
Row effectively identifies, is particularly suitable for comprising polyglot and having the image of certain layout structure feature.
Accompanying drawing explanation
The embodiments of the present invention are read with reference to the drawings, other spy of the present invention be will be better understood
Seeking peace advantage, accompanying drawing described here is intended merely to schematically illustrate embodiments of the present invention
Purpose, and not all possible enforcement, and be not intended to limit the scope of the present invention.In the accompanying drawings:
Fig. 1 shows the flow chart of the method identifying address doorplate information in prior art；
Photo that Fig. 2 a and Fig. 2 b respectively illustrates input and the doorplate image extracted from this photo
An example；
Fig. 3 shows the flow chart of the character recognition method according to one embodiment of the present invention；
Fig. 4 shows and extracts multiple from the image comprising word according to one embodiment of the present invention
The flow chart of communication means；
Fig. 5 shows and extracts multiple from the image comprising word according to another embodiment of the present invention
The alternative flow chart of communication means；
It is that Fig. 6 shows first language communication means cluster according to one embodiment of the present invention
Non-first language communication means is also clustered the flow process for non-first language line of text by one language text row
Figure；
It is that Fig. 7 shows first language communication means cluster according to another embodiment of the present invention
Non-first language communication means is also clustered replacing for non-first language line of text by one language text row
Select flow chart；
Fig. 8 shows according to one embodiment of the present invention from first language line of text and non-first language
Speech line of text identifies first language word and the flow chart of non-first language word；
Fig. 9 shows and determines according to the layout structure feature of image according to one embodiment of the present invention
The flow chart of the space of a whole page classification that image is residing in multiple space of a whole page classifications；
Figure 10 a to Figure 10 d shows the example of four space of a whole page classifications；
Figure 11 shows the block diagram of the character recognition device according to one embodiment of the present invention；
Figure 12 shows the block diagram of the recognition unit according to one embodiment of the present invention；
Figure 13 shows that the space of a whole page classification according to one embodiment of the present invention determines the frame of subelement
Figure；
Figure 14 shows that the line of text according to one embodiment of the present invention processes the block diagram of subelement；
Figure 15 shows the block diagram of the extraction unit according to one embodiment of the present invention；
Figure 16 shows the block diagram of the cluster cell according to one embodiment of the present invention；
Figure 17 shows the block diagram of the cluster cell according to another embodiment of the present invention；And
Figure 18 shows the computer that can be used for implementing method and apparatus according to embodiments of the present invention
Schematic block diagram.
Detailed description of the invention
Referring now to accompanying drawing, embodiments of the present invention are described in detail.Only it should be noted that following description
It is only exemplary, and is not intended to limit the present invention.Additionally, in the following description, phase will be used
Same drawing reference numeral represents the same or analogous parts in different accompanying drawing.Described below different real
Execute the different characteristic in mode, can be bonded to each other, to form other embodiments in the scope of the invention.
In embodiments of the present invention, it is assumed that utilized technology well known by persons skilled in the art from photograph
Sheet detects and extracts the image-region comprising word, doorplate image as shown in Figure 2 b.And
Various process and operation in embodiments of the present invention are all to do on the basis of obtaining this image
Go out.
Fig. 3 shows the flow chart of the character recognition method according to one embodiment of the present invention.Such as figure
Shown in 3, character recognition method 300 includes step S310 to S340.Literary composition is comprised for acquired
The image of word, in step S310, extracts multiple communication means from the image comprising word.Right
The extraction of communication means can use any applicable mode well known by persons skilled in the art.Extracted
Each communication means can comprise the part of one or more alphabetic character or alphabetic character.
Owing to image generally comprising the word of polyglot, in step s 320, by different language
The multiple communication means extracted are classified, thus is divided into first language to connect communication means
Parts and non-first language communication means.It is appreciated that and does not wraps for only comprising first language word
Image containing other spoken and written languages, after the process of step S320, all interconnecting parts extracted
Part is all classified as first language communication means；And the word for being comprised not is first language literary composition
The image of word, after the process of step S320, all communication means extracted all are classified as
Non-first language communication means.
In step S330, it is first language line of text by first language communication means cluster, and will
Non-first language communication means cluster is non-first language line of text.The cluster of communication means can be used
Any applicable mode well known by persons skilled in the art.Each line of text obtained by after clustered can
Including one or more first language words or non-first language word.
In step S340, respectively from obtained first language line of text and non-first language text
Row identifies first language word and non-first language word, to complete Word message in image
Automatically identify.Available first language character string identification engine and non-first language character string identification engine
From line of text, identify first language word and non-first language word respectively, this will following in detail
State.
According to this embodiment of the present invention, by the word in image being categorized as first language and non-
First language such that it is able to identify the Word message including polyglot from image, such as, Ah
Arabic numbers, English alphabet, Chinese character etc..And, by different spoken and written languages are clustered into respectively
Different line of text are to be identified, it is possible to utilize the identification engine of different language separately to process not
With the line of text of language, and without using the identification engine of multilingual character to process each text
OK, thus improve recognition speed.
Fig. 4 shows and extracts multiple from the image comprising word according to one embodiment of the present invention
The flow chart of communication means.As shown in Figure 4, above-mentioned steps S310 can include sub-step S311 extremely
S313.In sub-step S311, from image, extract connected unit.To connected unit in image
Extract, can be completed by the means of any suitable prior art, be not described in detail in this.Subsequently,
In sub-step S312, calculate the recognition confidence of each connected unit extracted, and in sub-step
In rapid S313, recognition confidence is removed less than the connected unit of predetermined confidence threshold value, and will
The connected unit retained merges, to form multiple communication means.For each image comprising word,
Due to shooting problem or the character area (such as doorplate region) of image, itself there are some dirts sometimes,
And making there are some in extracted connected unit is noise.In order to remove noise, of the present invention
Embodiment has preset confidence threshold value.The recognition confidence of each connected unit by extracting
Compare with default confidence threshold value, thus filter out the connected unit that confidence level is relatively low, remove
Noise, to obtain the higher connected unit of confidence level.
Fig. 5 shows and extracts multiple from the image comprising word according to another embodiment of the present invention
The alternative flow chart of communication means.As it is shown in figure 5, above-mentioned steps S310 can include sub-step S315
To S318.In sub-step S315, from image, extract connected unit.Subsequently, in sub-step
In S316, calculate the stroke width of each connected unit extracted, and calculate all connections
The average stroke width of unit.In sub-step S317, according to average stroke width,
Determine stroke width range.For example it is assumed that calculated average stroke width is SW,
Stroke width range can be defined as 0.5*SW～1.5*SW.Subsequently, in sub-step S318
In, stroke width is not at the connected unit in the range of this and removes, and by not removed connection
Unit merges, to form multiple communication means.Process shown in Fig. 5 is another kind of optional denoising
Mode, it utilizes stroke width, is used as filtercondition, to remove noise.
It addition, for not only including deep background and shallow characters region but also include the mixed image in territory, deep block, the shallow end,
When extracting connected unit from image, image can be carried out front binaryzation and reverse side binaryzation, and will
The result of front binaryzation and reverse side binaryzation is analyzed respectively, to extract connected unit respectively.?
After, the connected unit extracted is merged, to form multiple communication means.
According to an embodiment of the invention, by different language to being carried in above-mentioned steps S320
When the multiple communication means taken out are classified, available the first language including all first language characters
Speech grader is classified.Such as, setting first language is Arabic numerals 0-9, then available bag
Including the grader of 0-9, the communication means extracted is categorized as first language communication means (is Ah
Arabic numbers) and non-first language communication means (not being Arabic numerals).According to the present invention one
Individual embodiment, the character quantity of first language, less than the character quantity of non-first language, so, can subtract
Amount of calculation during subclassification, improves processing speed.It will be understood by those skilled in the art that also can be by non-
First language is categorized as second language and non-first non-second language (by that analogy), it is possible to obtain relatively
Good effect, its concrete mode is similar to the above, is not described in detail in this.
It is that Fig. 6 shows first language communication means cluster according to one embodiment of the present invention
Non-first language communication means is also clustered the flow process for non-first language line of text by one language text row
Figure.As shown in Figure 6, above-mentioned steps S330 can include sub-step S331 to S334.In sub-step
In S331, compare the horizontal interval between first language communication means and vertical spacing, and according to than
Relatively result is that each first language communication means arranges level marks or vertical labelling.Specifically, for
Each first language communication means, compares between the level of its first language communication means adjacent with level
Every and the size of vertical spacing with the most adjacent first language communication means.If horizontal interval is relatively
Little, illustrate the compactest, then level marks is set for this communication means, otherwise then
Vertical labelling is set.Similarly, in sub-step S332, according to non-first language communication means it
Between horizontal interval and the comparative result of vertical spacing, for each non-first language communication means, water is set
Flat labelling or vertically labelling.
Then, in sub-step S333, will have first language communication means and the tool of level marks
The first language communication means having vertical labelling clusters respectively as the horizontal line of text of first language and first
The vertical line of text of language.In sub-step S334, the non-first language with level marks is connected
Parts and the non-first language communication means with vertical labelling cluster respectively as non-first language level
Line of text and the vertical line of text of non-first language.
It is that Fig. 7 shows first language communication means cluster according to another embodiment of the present invention
Non-first language communication means is also clustered replacing for non-first language line of text by one language text row
Select flow chart.As it is shown in fig. 7, above-mentioned steps S330 can include sub-step S335 to S338.?
In sub-step S335, compare the horizontal interval between first language communication means and vertical spacing with pre-
If the size of threshold value, and be that first language communication means arranges level marks and perpendicular according to comparative result
Straight labelling.Specifically, for each first language communication means, first language adjacent with level by it
Horizontal interval between speech communication means compares with the threshold value preset, and by it with the most adjacent
Vertical spacing between first language communication means compares with this threshold value.If horizontal interval is less than
This threshold value, then arrange level marks, if horizontal interval is more than this threshold value, is then not provided with level marks.
Equally, if vertical spacing is less than this threshold value, then vertical labelling is set, if vertical spacing is more than being somebody's turn to do
Threshold value, then be not provided with vertical labelling.Similarly, in sub-step S336, relatively non-first language
The size of horizontal interval between communication means and vertical spacing and the threshold value preset, and according to comparing knot
Fruit arranges level marks and vertical labelling for non-first language communication means.
It is appreciated that for some communication means, is likely to be due to its horizontal interval and vertical spacing is the least
In this threshold value preset, and it is provided with level marks and vertical labelling simultaneously.In sub-step S337
In, not only there is level marks but also there is first language communication means and the Fei of vertical labelling for each
One language communication means, according to its horizontal interval with similar communication means and the comparison knot of vertical spacing
Really, one of its level marks and vertical labelling are removed.That is, if first language communication means is in level
Closer to another first language communication means on direction, then retain its level marks, otherwise, then retain
Its vertical labelling.Subsequently, in sub-step S338, the first language with level marks is connected
Parts and the first language communication means with vertical labelling cluster respectively as the horizontal text of first language
Row and the vertical line of text of first language, and will have non-first language communication means and the tool of level marks
The non-first language communication means having vertical labelling cluster respectively as the horizontal line of text of non-first language and
The vertical line of text of non-first language.
By above-mentioned process, available through language classification the difference that determines Text region direction
Line of text, but also need to determine the recognition sequence between different line of text.To this end, present applicant proposes one
Plant the space of a whole page classification being determined pending image by default space of a whole page template, so that it is determined that different text
Recognition sequence between row.
Fig. 8 shows according to one embodiment of the present invention from first language line of text and non-first language
Speech line of text identifies first language word and the flow chart of non-first language word.Such as Fig. 8 institute
Showing, above-mentioned steps S340 can include sub-step S341 to S343.In sub-step S341, according to
First language line of text and non-first language line of text calculate the layout structure feature of image.Subsequently, exist
In sub-step S342, according to the layout structure feature calculated, determine that this image is known many
Which space of a whole page classification is individual space of a whole page classification belong to.After determining space of a whole page classification, i.e. determine not identical text
Processing sequence between one's own profession.In sub-step S343, process it according to the space of a whole page classification of this image
In first language line of text and non-first language line of text, to identify first language word and Fei
One spoken and written languages.
According to an embodiment of the invention, the version of the image calculated in above-mentioned sub-step S341
Area structure character comprises the steps that the geometric properties of line of text the longest in first language line of text and identification are put
The geometric properties of line of text the longest in reliability and non-first language line of text.
Specifically, the geometric properties of line of text can include the coboundary of this article one's own profession, lower boundary, the left side
Average the ratio of width to height of communication means in boundary, right margin, this article one's own profession and/or adjacent communication means
Equispaced.
According to one embodiment of present invention, the layout structure feature of image comprises the steps that first language is
The recognition confidence (P) of long article one's own profession, 6 dimension geometric properties (tops of first language long article one's own profession
Boundary-y0, lower boundary-y1, left margin-x0, the relative position of right margin-x1, communication means average
The ratio of width to height, the equispaced of adjacent communication means) and 6 dimension geometry of second language long article one's own profession
Feature (coboundary-y0, lower boundary-y1, left margin-x0, the relative position of right margin-x1, connection
Average the ratio of width to height of parts, the equispaced of adjacent communication means).Wherein, first language long article
The recognition confidence (P) of one's own profession refers to the average identification of all candidate's communication means in this article one's own profession
Confidence level, its calculation is as follows:
P=(PCC1+PCC2+……+PCCM)/M
The number of candidate's communication means during wherein M represents the long article one's own profession of first language.Except this it
Outward, this average recognition confidence can also use other calculations, such as: all in this article one's own profession
Based on width weighed the average recognition confidence of candidate's communication means, it may be assumed that
Wherein wCCMRepresenting the width of m-th candidate's communication means in this article one's own profession, w represents this article
The width of one's own profession.
Fig. 9 shows and determines according to the layout structure feature of image according to one embodiment of the present invention
The flow chart of the space of a whole page classification that image is residing in multiple space of a whole page classifications.As it is shown in figure 9, above-mentioned sub-step
Rapid S342 can include sub-step S342a to S342b.In sub-step S342a, according to image
Layout structure feature, calculates each confidence level probability in this image and multiple space of a whole page classifications.Subsequently,
In sub-step S342b, the space of a whole page classification of the confidence level maximum probability with this image is defined as its version
Face classification.According to an embodiment, image and the confidence level probability of each space of a whole page classification by trained really
Fixed discriminant function is calculated by the nonlinear transformation that threshold value is limited, and this discriminant function is permissible
It is linear or nonlinear.
According to an embodiment of the invention, available multiple training samples, according to layout structure
Difference, pre-defined N kind space of a whole page classification.Figure 10 a to Figure 10 d shows four space of a whole page classifications
Example.Utilize multiple known training sample, their layout structure feature can be calculated, thus can
To train the grader of a N class, can be linear classifier or Nonlinear Classifier, such as,
Classical Linear SVM grader in discrimination model.
The sample belonging to each space of a whole page classification is regarded as the positive sample of the category, and other all samples are seen
Become the negative sample of the category, it is possible to training obtains a linear discriminant function, just to reach to distinguish,
The purpose of negative sample, is shown below.
fi(x)=wi Tx+bi, i=1 ..., N
Wherein, x represents layout structure characteristic vector, coefficient wiAnd biBe in linear classifier with version
Relevant trained of face classification i and the parameter that determines, fiX () represents that space of a whole page classification i is tied about the space of a whole page
The linear function of structure feature x.
Then this linear discriminant function is converted by following Sigmoid and is converted to confidence level probability:
Wherein, parameter alpha is positive number, and parameter beta is real number, is all default coefficient, can by experiment really
Fixed.Such as, α and β can be respectively set to 1 and 0.Each classification can share identical parameter
α and β.PiX () represents the confidence level probability of image and space of a whole page classification i.
Parameter w of above-mentioned N class graderi、bi, α and β can pass through training side known in the art
Method obtains.The LibSVM tool kit increased income can be used to carry out the training of N class Linear SVM.
In practical operation, the layout structure feature of the image calculated can be input to above-mentioned training
N class grader, so that the classification of classification confidence output probability with maximum is defined as this figure
As space of a whole page classification residing in known multiple space of a whole page classifications.
According to an embodiment of the invention, above-mentioned sub-step S343 comprises the steps that and is utilized respectively
One language character string identification engine and non-first language character string identification engine identification first language word
With non-first language word.Further, a predeterminable corrected threshold.For the image of input, work as institute
When the maximum confidence probability obtained is less than this corrected threshold preset, then can change and be respectively used to identify
The character string identification engine of first language line of text and/or non-first language line of text.Additionally, also may be used
Preset another corrected threshold.For the image of input, when all in the first language line of text that it is the longest
When the average recognition confidence of communication means is respectively less than this another corrected threshold preset, adjustable first
The line of text identification direction of language text row and/or non-first language line of text.Thus, can be by pre-
If the adjusting thresholds processing mode to image, to improve accuracy.
Figure 11 shows the block diagram of the character recognition device according to one embodiment of the present invention.Such as figure
Shown in 11, character recognition device 1100 comprises the steps that extraction unit 1110, taxon 1120, gathers
Class unit 1130 and recognition unit 1140.Extraction unit 1110 can extract from the image comprising word
Multiple communication means.Multiple communication means that extraction unit 1110 extracts can be entered by taxon 1120
Row classification, to generate first language communication means and/or non-first language communication means.Cluster cell
The first language communication means cluster that taxon 1120 can be generated by 1130 is first language text
OK, and by non-first language communication means cluster as non-first language line of text.Recognition unit 1140
The first language line of text that can be clustered from cluster cell 1130 and non-first language line of text identify
Go out first language word and non-first language word.
Figure 12 shows the block diagram of the recognition unit according to one embodiment of the present invention.Such as Figure 12
Shown in, recognition unit 1140 comprises the steps that layout structure feature calculation subelement 1141, space of a whole page classification
Determine that subelement 1142 and line of text process subelement 1143.Layout structure feature calculation subelement
The 1141 first language line of text that can be clustered according to cluster cell 1130 and non-first language line of text
Calculate the layout structure feature of image.Space of a whole page classification determines that subelement 1142 can be special according to layout structure
Levy the layout structure feature of the image that computation subunit 1141 is calculated, determine that image is at multiple spaces of a whole page
Space of a whole page classification residing in classification.Line of text processes subelement 1143 can determine son according to space of a whole page classification
Unit 1142 processes first language line of text and non-first language literary composition for the space of a whole page classification that this image determines
One's own profession, to identify first language word and non-first language word.
According to an embodiment of the invention, the layout structure feature of image comprises the steps that first language
The geometric properties of line of text the longest in line of text and recognition confidence and non-first language line of text
In the geometric properties of the longest line of text.
Figure 13 shows that the space of a whole page classification according to one embodiment of the present invention determines the frame of subelement
Figure.As shown in figure 13, space of a whole page classification determines that subelement 1142 comprises the steps that confidence level probability calculation mould
Block 1142a and space of a whole page category determination module 1142b.Confidence level probability evaluation entity 1142a can basis
The layout structure feature of image, calculates this image and each space of a whole page class in the multiple space of a whole page classifications preset
Other confidence level probability.Space of a whole page category determination module 1142b can be by with the confidence level probability of this image
Big space of a whole page classification is defined as the space of a whole page classification of this image.
Figure 14 shows that the line of text according to one embodiment of the present invention processes the block diagram of subelement.
As shown in figure 14, line of text processes subelement 1143 and comprises the steps that first language character string identification engine
1143a, non-first language character string identification engine 1143b, identification engine handover module 1143c and knowledge
Other direction handover module 1143d.First language character string identification engine 1143a and non-first language word
Symbol string identifies that engine 1143b can be respectively used to identify first language word and non-first language word.Right
In the image inputted, when its maximum confidence probability is less than the first predetermined corrected threshold, identify
Engine handover module 1143c can change and be respectively used to identify first language line of text and/or non-first language
The character string identification engine of speech line of text.For the image inputted, when the first language literary composition that it is the longest
When in one's own profession, the average recognition confidence of all communication means is respectively less than the second predetermined corrected threshold, know
Other direction handover module 1143d adjustable first language line of text and/or non-first language line of text
Line of text identification direction.
Figure 15 shows the block diagram of the extraction unit according to one embodiment of the present invention.Such as Figure 15
Shown in, extraction unit 1110 comprises the steps that extraction subelement 1111, recognition confidence computation subunit
1112, stroke width calculation subelement 1113, scope determine subelement 1114 and merge son list
Unit 1115.Extract subelement 1111 and can extract connected unit from image.Recognition confidence calculates son
Unit 1112 can calculate the identification confidence extracting each connected unit that subelement 1111 is extracted
Degree.Stroke width calculation subelement 1113 can calculate extract subelement 1111 extracted every
The stroke width of individual connected unit and average stroke width.Scope determines subelement 1114
Stroke width range can be determined according to the average stroke width calculated.Merge son single
Recognition confidence can be less than connected unit and the stroke width of predetermined confidence threshold value by unit 1115
The connected unit that degree is not in stroke width range removes, and is closed by the connected unit retained
And, to form multiple communication means.
Figure 16 shows the block diagram of the cluster cell according to one embodiment of the present invention.Such as Figure 16
Shown in, cluster cell 1130 comprises the steps that and compares subelement 1131, labelling subelement 1132 and cluster
Subelement 1133.Relatively subelement 1131 can by the horizontal interval between first language communication means with
Vertical spacing compares, and by the horizontal interval between non-first language communication means and vertical spacing
Compare.Labelling subelement 1132 can be each according to the comparative result comparing subelement 1131
One language communication means and non-first language communication means arrange level marks or vertical labelling.Cluster
Unit 1133 can will have the first language communication means of level marks and non-first language communication means
Cluster is the horizontal line of text of first language and the horizontal line of text of non-first language respectively, and will have vertically
The first language communication means of labelling and non-first language communication means cluster respectively erects into first language
Straight line of text and the vertical line of text of non-first language.
Figure 17 shows the block diagram of the cluster cell according to another embodiment of the present invention.Such as Figure 17
Shown in, cluster cell 1130 comprises the steps that and compares subelement 1135, labelling subelement 1136, labelling
Remove subelement 1137 and cluster subelement 1138.First language can be connected by relatively subelement 1135
Between the level between horizontal interval and vertical spacing and non-first language communication means between parts
Every comparing with default threshold value respectively with vertical spacing.Labelling subelement 1136 can be according to comparing
The comparative result of subelement 1135 is first language communication means and the setting of non-first language communication means
Level marks and vertical labelling.For being marked with the first language of level marks and vertical both labellings even
Each in logical parts and non-first language communication means, labelling removes subelement 1137 can be according to it
With horizontal interval and the size of vertical spacing of similar communication means, remove its level marks and vertically mark
One of note.Cluster subelement 1138 can have first language communication means and the Fei of level marks
One language communication means clusters respectively as the horizontal line of text of first language and the horizontal text of non-first language
OK, and by the first language communication means and non-first language communication means with vertical labelling gather respectively
Class is the vertical line of text of first language and the vertical line of text of non-first language.
It will be understood by those skilled in the art that character recognition method provided by the present invention and device both may be used
For identifying the doorplate Word message in the doorplate image-region in captured photo it can also be used to identify
Word message in any image comprising word, be particularly suited for having certain layout structure feature,
Image that belong to certain space of a whole page type, that comprise polyglot word.
It addition, still need here it is noted that in said apparatus each building block can pass through software,
The mode of firmware, hardware or a combination thereof configures.Configure spendable specific means or mode for this
Known to skilled person, do not repeat them here.In the case of being realized by software or firmware,
From storage medium or network to computer (such as general shown in Figure 18 with specialized hardware structure
Computer 1800) install constitute this software program, this computer when being provided with various program,
It is able to carry out various functions etc..
Figure 18 shows the computer that can be used for implementing method and apparatus according to embodiments of the present invention
Schematic block diagram.
In figure 18, CPU (CPU) 1801 is according in read only memory (ROM) 1802
The program stored or the program being loaded into random access memory (RAM) 1803 from storage part 1808
Perform various process.In RAM 1803, perform various always according to needs storage as CPU 1801
Data required during process etc..CPU 1801, ROM 1802 and RAM 1803 are via bus
1804 are connected to each other.Input/output interface 1805 is also connected to bus 1804.
Components described below is connected to input/output interface 1805: importation 1806 (includes keyboard, Mus
Mark etc.), output part 1807 (include display, such as cathode ray tube (CRT), liquid crystal
Show device (LCD) etc., and speaker etc.), storage part 1808 (including hard disk etc.), communications portion
1809 (including NIC such as LAN card, modem etc.).Communications portion 1809 warp
Communication process is performed by network such as the Internet.As required, driver 1810 can be connected to defeated
Enter/output interface 1805.Detachable media 1811 such as disk, CD, magneto-optic disk, quasiconductor are deposited
Reservoir etc. can be installed in driver 1810 as required so that the computer read out
Program is installed to store in part 1808 as required.
In the case of realizing above-mentioned series of processes by software, it is situated between from network such as the Internet or storage
Matter such as detachable media 1811 installs the program constituting software.
It will be understood by those of skill in the art that this storage medium is not limited to its shown in Figure 18
In have program stored therein and equipment distributes the detachable media of the program that provides a user with separately
1811.The example of detachable media 1811 comprises disk (comprising floppy disk (registered trade mark)), CD (comprises
Compact disc read-only memory (CD-ROM) and digital universal disc (DVD)), magneto-optic disk (comprise mini-disk
(MD) (registered trade mark)) and semiconductor memory.Or, storage medium can be ROM 1802,
Hard disk of comprising etc. in storage part 1808, wherein computer program stored, and with comprise setting of they
For being distributed to user together.
The present invention also proposes the program product that a kind of storage has the instruction code of machine-readable.Described finger
When making code be read by machine and perform, the above-mentioned method according to embodiment of the present invention can be performed.
Correspondingly, for carrying the depositing of program product that above-mentioned storage has the instruction code of machine-readable
Storage media is intended to be included within the scope of the present invention.Described storage medium include but not limited to floppy disk, CD,
Magneto-optic disk, storage card, memory stick etc..
It should be noted that, the method for the present invention be not limited to specifications described in time sequencing hold
OK, it is also possible to sequentially, in parallel or independently perform according to other order.Therefore, this explanation
The technical scope of the present invention is not construed as limiting by the execution sequence of the method described in book.
The description of embodiment each to the present invention is to be more fully understood that the present invention above, and it is only
Exemplary, and be not intended to limit the invention.It should be noted that in the above description, for one
Kind embodiment description and/or the feature illustrated can be in same or similar mode one or more
Other embodiment individual uses, combined with the feature in other embodiment, or it is real to substitute other
Execute the feature in mode.It will be understood by those skilled in the art that in the inventive concept without departing from the present invention
In the case of, the variations and modifications carried out for embodiment described above, belong to this
In the range of invention.
To sum up, in an embodiment according to the present invention, the invention provides following technical scheme.
Scheme 1, a kind of character recognition method, including:
Multiple communication means is extracted from the image comprising word；
The plurality of communication means is classified, to generate first language communication means and/or non-
One language communication means；
It is first language line of text by described first language communication means cluster, and by described non-first language
Speech communication means cluster is non-first language line of text；And
First language is identified from described first language line of text and described non-first language line of text
Word and non-first language word.
Scheme 2, method as described in scheme 1, wherein from described first language line of text and described non-
First language line of text identifies first language word and non-first language word includes:
Described image is calculated according to described first language line of text and described non-first language line of text
Layout structure feature；
Layout structure feature according to described image, determines that described image is residing in multiple space of a whole page classifications
Space of a whole page classification；And
Space of a whole page classification according to described image processes described first language line of text and described non-first language
Speech line of text, to identify first language word and non-first language word.
Scheme 3, method as described in scheme 2, the layout structure feature of wherein said image includes:
The geometric properties of line of text the longest in first language line of text and recognition confidence and non-first language
The geometric properties of line of text the longest in speech line of text.
Scheme 4, method as described in scheme 3, wherein the geometric properties of line of text includes this article one's own profession
Coboundary, lower boundary, left margin, right margin, communication means in this article one's own profession average wide high
Ratio and/or the equispaced of adjacent communication means.
Scheme 5, method as according to any one of scheme 2-4, wherein according to the space of a whole page of described image
Architectural feature, determines that the space of a whole page classification residing for described image is in multiple space of a whole page classifications includes:
Layout structure feature according to described image, calculates in described image and the plurality of space of a whole page classification
Each confidence level probability；And
Space of a whole page classification with the confidence level maximum probability of described image is defined as the space of a whole page of described image
Classification.
Scheme 6, method as described in scheme 5, wherein said image is general with the confidence level of each classification
Rate is calculated by the nonlinear transformation that threshold value is limited by the trained discriminant function determined, described in sentence
Other function is linear or nonlinear.
Scheme 7, method as described in scheme 5 or 6, wherein according to the space of a whole page class other places of described image
Manage described first language line of text and described non-first language line of text, to identify first language word
Include with non-first language word:
It is utilized respectively first language character string identification engine and non-first language character string identification engine is known
Other first language word and non-first language word；And
Wherein, for maximum confidence probability less than the image of the first predetermined corrected threshold, change and divide
Yong Yu not identify the character string identification engine of first language line of text and/or non-first language line of text；
And
Wherein, for the average recognition confidence of all communication means in the longest first language line of text
It is respectively less than the image of the second predetermined corrected threshold, adjusts first language line of text and/or non-first language
The line of text identification direction of speech line of text.
Scheme 8, method as according to any one of scheme 1-7, wherein from the image comprising word
Extract multiple communication means to include:
Connected unit is extracted from described image；
Calculate the recognition confidence of each connected unit extracted；And
Recognition confidence is removed less than the connected unit of predetermined confidence threshold value, and by not removed
Connected unit merges, to form the plurality of communication means.
Scheme 9, method as according to any one of scheme 1-8, wherein from the image comprising word
Extract multiple communication means to include:
Connected unit is extracted from described image；
Calculate the stroke width of each connected unit extracted and average stroke width；
According to average stroke width, determine stroke width range；And
The connected unit being not at by stroke width in described stroke width range removes, and
Not removed connected unit is merged, to form the plurality of communication means.
Scheme 10, method as according to any one of scheme 1-9, wherein connect described first language
Parts cluster is first language line of text, and by described non-first language communication means cluster for non-first
Language text row includes:
According to the horizontal interval between first language communication means and the comparative result of vertical spacing and
Horizontal interval between non-first language communication means and the comparative result of vertical spacing, for each first
Language communication means and non-first language communication means arrange level marks or vertical labelling；
The first language communication means and non-first language communication means with level marks are gathered respectively
Class is the horizontal line of text of first language and the horizontal line of text of non-first language；And
The first language communication means and non-first language communication means with vertical labelling are gathered respectively
Class is the vertical line of text of first language and the vertical line of text of non-first language.
Scheme 11, method as according to any one of scheme 1-9, wherein connect described first language
Parts cluster is first language line of text, and by described non-first language communication means cluster for non-first
Language text row includes:
According to the horizontal interval between first language communication means with preset threshold value comparative result with
And the comparative result of the horizontal interval between non-first language communication means and described default threshold value, for
First language communication means and non-first language communication means arrange level marks；
Comparison knot according to the vertical spacing between first language communication means with described default threshold value
Vertical spacing between fruit and non-first language communication means and the comparison knot of described default threshold value
Really, vertical labelling is set for first language communication means and non-first language communication means；
For being marked with the first language communication means and non-first of level marks and vertical both labellings
Each in language communication means, according to its horizontal interval with similar communication means and vertical spacing
Comparative result, removes one of its level marks and vertical labelling；And
The first language communication means and non-first language communication means with level marks are gathered respectively
Class is the horizontal line of text of first language and the horizontal line of text of non-first language, and will have vertical labelling
First language communication means and non-first language communication means cluster respectively as the vertical text of first language
Row and the vertical line of text of non-first language.
Scheme 12, method as according to any one of scheme 1-11, the wherein number of characters of first language
Amount is less than the character quantity of non-first language.
Scheme 13, a kind of character recognition device, including:
Extraction unit, extracts multiple communication means from the image comprising word；
Taxon, classifies to the plurality of communication means, to generate first language communication means
And/or non-first language communication means；
Cluster cell, is first language line of text by described first language communication means cluster, and by institute
Stating non-first language communication means cluster is non-first language line of text；And
Recognition unit, identifies from described first language line of text and described non-first language line of text
First language word and non-first language word.
Scheme 14, device as described in scheme 13, wherein said recognition unit includes:
Layout structure feature calculation subelement, according to described first language line of text and described non-first language
Speech line of text calculates the layout structure feature of described image；
Space of a whole page classification determines subelement, according to the layout structure feature of described image, determines described image
Space of a whole page classification residing in multiple space of a whole page classifications；And
Line of text processes subelement, processes described first language text according to the space of a whole page classification of described image
Row and described non-first language line of text, to identify first language word and non-first language word.
Scheme 15, device as described in scheme 14, the layout structure feature of wherein said image includes:
The geometric properties of line of text the longest in first language line of text and recognition confidence and non-first language
The geometric properties of line of text the longest in speech line of text.
Scheme 16, device as described in scheme 14 or 15, wherein said space of a whole page classification determines that son is single
Unit includes:
Confidence level probability evaluation entity, according to the layout structure feature of described image, calculates described image
With each confidence level probability in the plurality of space of a whole page classification；And
Space of a whole page category determination module, determines the space of a whole page classification of the confidence level maximum probability with described image
Space of a whole page classification for described image.
Scheme 17, device as described in scheme 16, wherein said line of text processes subelement and includes:
First language character string identification engine and non-first language character string identification engine, be respectively used to know
Other first language word and non-first language word；
Identify engine handover module, for maximum confidence probability less than the first predetermined corrected threshold
Image, changes and is respectively used to identify first language line of text and/or the character of non-first language line of text
String identifies engine；And
Identify direction handover module, for all communication means flat in the longest first language line of text
All recognition confidences are respectively less than the image of the second predetermined corrected threshold, adjust first language line of text and
/ or the line of text identification direction of non-first language line of text.
Scheme 18, device as according to any one of scheme 13-17, wherein said extraction unit includes:
Extract subelement, from described image, extract connected unit；
Recognition confidence computation subunit, calculates the recognition confidence of each connected unit extracted；
Stroke width calculation subelement, calculates the stroke width of each connected unit extracted
Degree and average stroke width；
Scope determines subelement, according to average stroke width, determines stroke width range；
And
Merge subelement, recognition confidence is less than connected unit and the character of predetermined confidence threshold value
The connected unit that stroke width is not in described stroke width range removes, and by not removed
Connected unit merges, to form the plurality of communication means.
Scheme 19, device as according to any one of scheme 13-18, wherein said cluster cell includes:
Relatively subelement, compares the horizontal interval between first language communication means with vertical spacing
Relatively, and by the horizontal interval between non-first language communication means compare with vertical spacing；
Labelling subelement, is each first language interconnecting part according to the described comparative result comparing subelement
Part and non-first language communication means arrange level marks or vertical labelling；And
Cluster subelement, connects the first language communication means with level marks with non-first language
Parts cluster as the horizontal line of text of first language and the horizontal line of text of non-first language respectively, will have perpendicular
First language communication means and the non-first language communication means of straight labelling cluster respectively as first language
Vertically line of text and the vertical line of text of non-first language.
Scheme 20, device as according to any one of scheme 13-18, wherein said cluster cell includes:
Relatively subelement, by the horizontal interval between first language communication means and vertical spacing and non-
Horizontal interval and vertical spacing between first language communication means compare with default threshold value respectively
Relatively；
Labelling subelement, according to the described comparative result comparing subelement be first language communication means and
Non-first language communication means arranges level marks and vertical labelling；
Labelling removes subelement, for being marked with the first language of level marks and vertical both labellings even
Each, according between its with the level of similar communication means in logical parts and non-first language communication means
Every the size with vertical spacing, remove one of its level marks and vertical labelling；And
Cluster subelement, connects the first language communication means with level marks with non-first language
Parts cluster as the horizontal line of text of first language and the horizontal line of text of non-first language respectively, will have perpendicular
First language communication means and the non-first language communication means of straight labelling cluster respectively as first language
Vertically line of text and the vertical line of text of non-first language.
Claims (10)
1. a character recognition method, including:
Multiple communication means is extracted from the image comprising word；
The plurality of communication means is classified, to generate first language communication means and/or non-
One language communication means；
It is first language line of text by described first language communication means cluster, and by described non-first language
Speech communication means cluster is non-first language line of text；And
First language is identified from described first language line of text and described non-first language line of text
Word and non-first language word.
2. the method for claim 1, wherein from described first language line of text and described non-
First language line of text identifies first language word and non-first language word includes:
Described image is calculated according to described first language line of text and described non-first language line of text
Layout structure feature；
Layout structure feature according to described image, determines that described image is residing in multiple space of a whole page classifications
Space of a whole page classification；And
Space of a whole page classification according to described image processes described first language line of text and described non-first language
Speech line of text, to identify first language word and non-first language word.
3. method as claimed in claim 2, the layout structure feature of wherein said image includes:
The geometric properties of line of text the longest in first language line of text and recognition confidence and non-first language
The geometric properties of line of text the longest in speech line of text.
4. method as claimed in claim 2 or claim 3, wherein the layout structure according to described image is special
Levy, determine that the space of a whole page classification residing for described image is in multiple space of a whole page classifications includes:
Layout structure feature according to described image, calculates in described image and the plurality of space of a whole page classification
Each confidence level probability；And
Space of a whole page classification with the confidence level maximum probability of described image is defined as the space of a whole page of described image
Classification.
5. method as claimed in claim 4, wherein processes institute according to the space of a whole page classification of described image
State first language line of text and described non-first language line of text, to identify first language word and non-
First language word includes:
It is utilized respectively first language character string identification engine and non-first language character string identification engine is known
Other first language word and non-first language word；And
Wherein, for maximum confidence probability less than the image of the first predetermined corrected threshold, change and divide
Yong Yu not identify the character string identification engine of first language line of text and/or non-first language line of text；
And
Wherein, for the average recognition confidence of all communication means in the longest first language line of text
It is respectively less than the image of the second predetermined corrected threshold, adjusts first language line of text and/or non-first language
The line of text identification direction of speech line of text.
6. the method as according to any one of claim 1-3, wherein carries from the image comprising word
Take multiple communication means to include:
Connected unit is extracted from described image；
Calculate the recognition confidence of each connected unit extracted；And
Recognition confidence is removed less than the connected unit of predetermined confidence threshold value, and by not removed
Connected unit merges, to form the plurality of communication means.
7. the method as according to any one of claim 1-3, wherein carries from the image comprising word
Take multiple communication means to include:
Connected unit is extracted from described image；
Calculate the stroke width of each connected unit extracted and average stroke width；
According to average stroke width, determine stroke width range；And
The connected unit being not at by stroke width in described stroke width range removes, and
Not removed connected unit is merged, to form the plurality of communication means.
8. the method as according to any one of claim 1-3, wherein by described first language interconnecting part
Part cluster is first language line of text, and is non-first language by described non-first language communication means cluster
Speech line of text includes:
According to the horizontal interval between first language communication means and the comparative result of vertical spacing and
Horizontal interval between non-first language communication means and the comparative result of vertical spacing, for each first
Language communication means and non-first language communication means arrange level marks or vertical labelling；
The first language communication means and non-first language communication means with level marks are gathered respectively
Class is the horizontal line of text of first language and the horizontal line of text of non-first language；And
The first language communication means and non-first language communication means with vertical labelling are gathered respectively
Class is the vertical line of text of first language and the vertical line of text of non-first language.
9. the method as according to any one of claim 1-3, wherein by described first language interconnecting part
Part cluster is first language line of text, and is non-first language by described non-first language communication means cluster
Speech line of text includes:
According to the horizontal interval between first language communication means with preset threshold value comparative result with
And the comparative result of the horizontal interval between non-first language communication means and described default threshold value, for
First language communication means and non-first language communication means arrange level marks；
Comparison knot according to the vertical spacing between first language communication means with described default threshold value
Vertical spacing between fruit and non-first language communication means and the comparison knot of described default threshold value
Really, vertical labelling is set for first language communication means and non-first language communication means；
For being marked with the first language communication means and non-first of level marks and vertical both labellings
Each in language communication means, according to its horizontal interval with similar communication means and vertical spacing
Comparative result, removes one of its level marks and vertical labelling；And
The first language communication means and non-first language communication means with level marks are gathered respectively
Class is the horizontal line of text of first language and the horizontal line of text of non-first language, and will have vertical labelling
First language communication means and non-first language communication means cluster respectively as the vertical text of first language
Row and the vertical line of text of non-first language.
10. a character recognition device, including:
Extraction unit, extracts multiple communication means from the image comprising word；
Taxon, classifies to the plurality of communication means, to generate first language communication means
And/or non-first language communication means；
Cluster cell, is first language line of text by described first language communication means cluster, and by institute
Stating non-first language communication means cluster is non-first language line of text；And
Recognition unit, identifies from described first language line of text and described non-first language line of text
First language word and non-first language word.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510086612.1A CN105989341A (en) | 2015-02-17 | 2015-02-17 | Character recognition method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510086612.1A CN105989341A (en) | 2015-02-17 | 2015-02-17 | Character recognition method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN105989341A true CN105989341A (en) | 2016-10-05 |
Family
ID=57038481
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510086612.1A Pending CN105989341A (en) | 2015-02-17 | 2015-02-17 | Character recognition method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105989341A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107610696A (en) * | 2017-08-09 | 2018-01-19 | 上海斐讯数据通信技术有限公司 | The router control method and router of conversation message based on user's input |
| CN107886093A (en) * | 2017-11-07 | 2018-04-06 | 广东工业大学 | A kind of character detection method, system, equipment and computer-readable storage medium |
| CN108305209A (en) * | 2017-01-12 | 2018-07-20 | 富士通株式会社 | Character deformation method and Character deformation equipment |
| CN108875758A (en) * | 2017-05-09 | 2018-11-23 | 富士通株式会社 | Information processing method and device and information detecting method and device |
| CN109685055A (en) * | 2018-12-26 | 2019-04-26 | 北京金山数字娱乐科技有限公司 | Text filed detection method and device in a kind of image |
| CN109919037A (en) * | 2019-02-01 | 2019-06-21 | 汉王科技股份有限公司 | A kind of text positioning method and device, text recognition method and device |
| CN109977762A (en) * | 2019-02-01 | 2019-07-05 | 汉王科技股份有限公司 | A kind of text positioning method and device, text recognition method and device |
| CN110458167A (en) * | 2019-08-20 | 2019-11-15 | 浙江工业大学 | A kind of metalwork surface curvature line of text antidote |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2004246546A (en) * | 2003-02-13 | 2004-09-02 | Ricoh Co Ltd | Image processing method, program used for execution of method, and image processing apparatus |
| CN1808468A (en) * | 2005-01-17 | 2006-07-26 | 佳能信息技术（北京）有限公司 | Optical character recognition method and system |
| CN101751567A (en) * | 2008-12-12 | 2010-06-23 | 汉王科技股份有限公司 | Quick text recognition method |
| CN101833648A (en) * | 2009-03-13 | 2010-09-15 | 汉王科技股份有限公司 | Method for correcting text image |
| CN102194117A (en) * | 2010-03-05 | 2011-09-21 | 北京大学 | Method and device for detecting page direction of document |
| CN102332096A (en) * | 2011-10-17 | 2012-01-25 | 中国科学院自动化研究所 | Video caption text extraction and identification method |
| CN102890783A (en) * | 2011-07-20 | 2013-01-23 | 富士通株式会社 | Method and device for recognizing direction of character in image block |
| CN103455806A (en) * | 2012-05-31 | 2013-12-18 | 富士通株式会社 | Document processing device, document processing method and scanner |
| CN103544475A (en) * | 2013-09-23 | 2014-01-29 | 方正国际软件有限公司 | Method and system for recognizing layout types |
-
2015
- 2015-02-17 CN CN201510086612.1A patent/CN105989341A/en active Pending
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2004246546A (en) * | 2003-02-13 | 2004-09-02 | Ricoh Co Ltd | Image processing method, program used for execution of method, and image processing apparatus |
| CN1808468A (en) * | 2005-01-17 | 2006-07-26 | 佳能信息技术（北京）有限公司 | Optical character recognition method and system |
| CN101751567A (en) * | 2008-12-12 | 2010-06-23 | 汉王科技股份有限公司 | Quick text recognition method |
| CN101833648A (en) * | 2009-03-13 | 2010-09-15 | 汉王科技股份有限公司 | Method for correcting text image |
| CN102194117A (en) * | 2010-03-05 | 2011-09-21 | 北京大学 | Method and device for detecting page direction of document |
| CN102890783A (en) * | 2011-07-20 | 2013-01-23 | 富士通株式会社 | Method and device for recognizing direction of character in image block |
| CN102332096A (en) * | 2011-10-17 | 2012-01-25 | 中国科学院自动化研究所 | Video caption text extraction and identification method |
| CN103455806A (en) * | 2012-05-31 | 2013-12-18 | 富士通株式会社 | Document processing device, document processing method and scanner |
| CN103544475A (en) * | 2013-09-23 | 2014-01-29 | 方正国际软件有限公司 | Method and system for recognizing layout types |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108305209B (en) * | 2017-01-12 | 2021-06-04 | 富士通株式会社 | Character deformation method and character deformation apparatus |
| CN108305209A (en) * | 2017-01-12 | 2018-07-20 | 富士通株式会社 | Character deformation method and Character deformation equipment |
| CN108875758A (en) * | 2017-05-09 | 2018-11-23 | 富士通株式会社 | Information processing method and device and information detecting method and device |
| CN108875758B (en) * | 2017-05-09 | 2022-01-11 | 富士通株式会社 | Information processing method and device, and information detection method and device |
| CN107610696A (en) * | 2017-08-09 | 2018-01-19 | 上海斐讯数据通信技术有限公司 | The router control method and router of conversation message based on user's input |
| CN107886093A (en) * | 2017-11-07 | 2018-04-06 | 广东工业大学 | A kind of character detection method, system, equipment and computer-readable storage medium |
| CN109685055A (en) * | 2018-12-26 | 2019-04-26 | 北京金山数字娱乐科技有限公司 | Text filed detection method and device in a kind of image |
| CN109685055B (en) * | 2018-12-26 | 2021-11-12 | 北京金山数字娱乐科技有限公司 | Method and device for detecting text area in image |
| CN109977762A (en) * | 2019-02-01 | 2019-07-05 | 汉王科技股份有限公司 | A kind of text positioning method and device, text recognition method and device |
| CN109919037B (en) * | 2019-02-01 | 2021-09-07 | 汉王科技股份有限公司 | Text positioning method and device and text recognition method and device |
| CN109919037A (en) * | 2019-02-01 | 2019-06-21 | 汉王科技股份有限公司 | A kind of text positioning method and device, text recognition method and device |
| CN110458167A (en) * | 2019-08-20 | 2019-11-15 | 浙江工业大学 | A kind of metalwork surface curvature line of text antidote |
| CN110458167B (en) * | 2019-08-20 | 2022-02-15 | 浙江工业大学 | Metal piece surface bending text line correction method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105989341A (en) | Character recognition method and device | |
| CN104463101B (en) | Answer recognition methods and system for character property examination question | |
| CN107301171A (en) | A kind of text emotion analysis method and system learnt based on sentiment dictionary | |
| CN107391760A (en) | User interest recognition methods, device and computer-readable recording medium | |
| CN112926405A (en) | Method, system, equipment and storage medium for detecting wearing of safety helmet | |
| CN106446526A (en) | Electronic medical record entity relation extraction method and apparatus | |
| CN109800749A (en) | A kind of character recognition method and device | |
| CN108959265A (en) | Cross-domain texts sensibility classification method, device, computer equipment and storage medium | |
| CN111046886A (en) | Automatic identification method, device and equipment for number plate and computer readable storage medium | |
| CN105117740A (en) | Font identification method and device | |
| CN109583401A (en) | It is a kind of automatically generate answer search topic method and user equipment | |
| CN109189965A (en) | Pictograph search method and system | |
| CN112528934A (en) | Improved YOLOv3 traffic sign detection method based on multi-scale feature layer | |
| CN108090099B (en) | Text processing method and device | |
| CN109800309A (en) | Classroom Discourse genre classification methods and device | |
| CN108345633A (en) | A kind of natural language processing method and device | |
| CN108363717A (en) | A kind of recognition detection method and device of data security levels | |
| CN106227836B (en) | Unsupervised joint visual concept learning system and unsupervised joint visual concept learning method based on images and characters | |
| CN106909600A (en) | The collection method and device of user context information | |
| CN112800225A (en) | Microblog comment emotion classification method and system | |
| CN107507161A (en) | A kind of evaluation method and system for writing tidiness | |
| CN108920451A (en) | Text emotion analysis method based on dynamic threshold and multi-categorizer | |
| CN108681532A (en) | A kind of sentiment analysis method towards Chinese microblogging | |
| CN104347071A (en) | Method and system for generating oral test reference answer | |
| CN111159977A (en) | Method and device for determining standard human phenotype expression |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20161005 |