CN106709489A - Processing method and device of character identification - Google Patents
Processing method and device of character identification Download PDFInfo
- Publication number
- CN106709489A CN106709489A CN201510410166.5A CN201510410166A CN106709489A CN 106709489 A CN106709489 A CN 106709489A CN 201510410166 A CN201510410166 A CN 201510410166A CN 106709489 A CN106709489 A CN 106709489A
- Authority
- CN
- China
- Prior art keywords
- character
- page
- row
- altitude range
- current
- 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.)
- Granted
Links
Landscapes
- Character Input (AREA)
Abstract
The invention discloses a processing method and device of character identification, and the method and device are used to improve the character identification effect. The method comprises that characters in a page belong to rows in the page according to height information of the characters in the page, the characters included by each row of the page are obtained, and the height information of each character in the page comprises a ordinate of the character in the page as well as a height range of the character; row correction is carried out on the characters included by each row in the page according to overlap information of the characters in the height range, and row-corrected characters included by each row in the page are obtained; and a semantic analysis model is used to carry out semantic correction on the row-corrected characters included by each row in the page.
Description
Technical field
The present invention relates to field of computer technology, more particularly to a kind for the treatment of method and apparatus of character recognition.
Background technology
Character segmentation and character recognition are optical character identification (English full name：Optical Character
Recognition, english abbreviation：OCR) most important two aspects, the direct shadow of this two parts in technology
The effect and result to character recognition are rung, needs the character that will have split to carry out line direction in OCR technique
On sequence after be presented to user, therefore the line direction ordering techniques of character can be directly influenced and be presented to use
The recognition effect of family viewing.But currently for OCR branch's technology be mainly based upon segmentation after intercharacter
Away to character carrying out simple branch.
Enter the merging and fractionation of line character according to the character pitch after segmentation in the prior art, when not apposition
When occurring after the character of formula carries out typesetting, situations such as the every line character for photographing has very big inclination in the page,
There is larger error to the character recognition on the page, and semanteme is carried out in later use recognition result
It also is difficult to reach accuracy very high during analysis.In addition, being in the prior art according to character to character branch
What spacing was realized, but with environmental change when character block combination is embarked on journey, there is a strong possibility can by other words
Symbol is influenceed, so as to final given recognition effect can be influenceed.
The content of the invention
A kind for the treatment of method and apparatus of character recognition are the embodiment of the invention provides, is known for improving character
Other recognition effect.
In order to solve the above technical problems, the embodiment of the present invention provides following technical scheme：
In a first aspect, the embodiment of the present invention provides a kind of processing method of character recognition, including：
Multiple characters on the page are belonged to the page by the elevation information according to character on the page
On multiple rows on, obtain multiple characters that the every a line on the page includes, the character is in the page
On elevation information include：The altitude range of character ordinate on the page and the character；
According to the overlay information between character on the page on altitude range to each on the page
Multiple characters that row includes enter every trade correction, after obtaining the row correction that the every a line on the page includes
Multiple characters；
Entered using the multiple characters after the row correction that semantic analysis model includes to the every a line on the page
Row missed suppression.
Second aspect, the embodiment of the present invention also provides a kind of processing unit of character recognition, including：
Row splits module, for the elevation information according to character on the page by the multiple words on the page
Symbol is belonged on the multiple rows on the page, obtains multiple characters that the every a line on the page includes,
Elevation information of the character on the page includes：Character ordinate on the page and the character
Altitude range；
Row correction module, for according to the overlay information pair between character on the page on altitude range
Multiple characters that each row on the page includes enter every trade correction, obtain the every a line on the page
Including row correction after multiple characters；
Missed suppression module, for the row included to the every a line on the page using semantic analysis model
Multiple characters after correction carry out missed suppression.
As can be seen from the above technical solutions, the embodiment of the present invention has advantages below：
In embodiments of the present invention, the elevation information first according to character on the page is by the multiple on the page
Character is belonged on the multiple rows on the page, obtains multiple characters that the every a line on the page includes, character
Elevation information on the page includes：The altitude range of ordinate of the character on the page and the character,
Next each row on the page is included according to the overlay information between character on the page on altitude range
Multiple characters enter every trade correction, obtain the multiple characters after the row that every a line on the page includes is corrected,
Multiple characters after the row for finally being included to the every a line on the page using semantic analysis model is corrected carry out language
Justice correction.The all characters on the page are belonged to using elevation information of the character on the page in the present invention
Multiple rows, because the altitude range of the ordinate and character of the character on the page of same a line is all relatively solid
It is fixed, thus according to character elevation information all characters on the page are carried out branch to branch's result be
Accurately, and in the present invention can also be according to the overlay information between character on the page on altitude range
Multiple characters that each row includes are entered with every trade correction, therefore can be detected to difference by overlay information
The row that character should belong in the character typesetting of form, the original of shooting is may also detect that by overlay information
Inclined character is there may be in the beginning page, in addition can also be using semantic analysis model to word in the present invention
Symbol carry out missed suppression, therefore character is entered every trade correction and missed suppression can change because of character typesetting
There is identification mistake caused by inclined character in form difference and shooting parent page, improve character knowledge
Other recognition effect.
Brief description of the drawings
Technical scheme in order to illustrate more clearly the embodiments of the present invention, in being described to embodiment below
The required accompanying drawing for using is briefly described, it should be apparent that, drawings in the following description are only this
Some embodiments of invention, to those skilled in the art, can also obtain according to these accompanying drawings
Other accompanying drawings.
Fig. 1 is a kind of process blocks schematic diagram of the processing method of character recognition provided in an embodiment of the present invention；
Fig. 2-a are that a kind of processing method to character recognition provided in an embodiment of the present invention realizes that scene is shown
It is intended to；
Fig. 2-b are a kind of implementation schematic diagram of the row attribute of correction character provided in an embodiment of the present invention；
Fig. 2-c be use semantic analysis model provided in an embodiment of the present invention carry out before missed suppression one
Plant content of pages schematic diagram；
Fig. 2-d be use semantic analysis model provided in an embodiment of the present invention carry out after missed suppression one
Plant content of pages schematic diagram；
Fig. 3-a are a kind of composition structural representation of the processing unit of character recognition provided in an embodiment of the present invention
Figure；
Fig. 3-b are the composition structural representation that a kind of row provided in an embodiment of the present invention splits module；
Fig. 3-c are a kind of composition structural representation of row correction module provided in an embodiment of the present invention；
Fig. 3-d are that the composition structure of the processing unit of another character recognition provided in an embodiment of the present invention is shown
It is intended to；
Fig. 3-e are that the composition structure of the processing unit of another character recognition provided in an embodiment of the present invention is shown
It is intended to；
Fig. 4 is that the processing method of character recognition provided in an embodiment of the present invention is applied to the composition knot of server
Structure schematic diagram.
Specific embodiment
A kind for the treatment of method and apparatus of character recognition are the embodiment of the invention provides, is known for improving character
Other recognition effect.
To enable that goal of the invention of the invention, feature, advantage are more obvious and understandable, below will
With reference to the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is carried out clear, complete
Ground description, it is clear that the embodiments described below are only a part of embodiment of the invention, and not all
Embodiment.Based on the embodiment in the present invention, the every other implementation that those skilled in the art is obtained
Example, belongs to the scope of protection of the invention.
Term " comprising " and " having " in description and claims of this specification and above-mentioned accompanying drawing
And their any deformation, it is intended that covering is non-exclusive to be included, so as to comprising a series of units
Process, method, system, product or equipment are not necessarily limited to those units, but may include without clearly
It is listing or for these processes, method, product or other intrinsic units of equipment.
It is described in detail individually below.
One embodiment of the processing method of character recognition of the present invention, in specifically can apply to COR technologies,
Refer to shown in Fig. 1, the processing method of the character recognition that one embodiment of the invention is provided can include
Following steps：
101st, the elevation information according to character on the page belongs on the page the multiple characters on the page
On multiple rows, multiple characters that the every a line on the page includes are obtained.
Wherein, elevation information of the character on the page includes：Ordinate of the character on the page and the word
The altitude range of symbol.
In embodiments of the present invention, the character on the page can be partitioned into from the page by OCR technique
Single character and the character obtained after each character is identified, each character on the page is got first
Elevation information, wherein, the elevation information of character refers to：Ordinate of the character on the page and the word
The altitude range of symbol, the elevation information by character on the page can accurately determine character in the page
Fixed coordinates on upper ordinate direction, and minimum point coordinates and highest of the character on ordinate direction
Point coordinates, wherein, the difference between highest point coordinates and minimum point coordinates on ordinate direction is word
The altitude range of symbol.For example, can using the top left corner apex of current page as the origin of coordinates, with
Top left corner apex is as a reference point, measures the top left corner apex of the character on ordinate direction to the coordinate
The distance of origin as character ordinate, in addition the top left corner apex of character be character in ordinate direction
On highest point coordinates, the lower-left angular vertex of character is minimum point coordinates of the character on ordinate direction,
Go out the altitude range of character by the mathematic interpolation between highest point coordinates and minimum point coordinates, it is also possible to claim
Be height size.
In the embodiment of the present invention, the elevation information of all characters on page-out is calculated, then according to the page
The elevation information of upper character enters the fractionation of every trade ownership to the multiple characters on the page, i.e., according to the height of character
Degree information belongs to the multiple characters on the page on the multiple rows on the page, and character is used with prior art
Spacing enter every trade and divide different, the row ownership determination method in the embodiment of the present invention to character is used
The height model of the elevation information of the page, the i.e. ordinate using character on the page and the character where character
Enclose to determine the ownership of row, because the height model of the ordinate and character with the character of a line on the page
Enclose and be all relatively fixed, therefore elevation information according to character carries out to all characters on the page branch and arrives
Branch's result is accurate.Further, in some embodiments of the invention, step 101 is according to word
The elevation information accorded with the page belongs to the multiple characters on the page on the multiple rows on the page, obtains
Multiple characters that every a line on the page includes, specifically may include steps of：
A1, from the multiple characters on the page arbitrarily selection one character as current character, according to current
Elevation information of the character on the page calculates ordinate of the central point of current character on the page；
A2, judge the central point of current character whether in the altitude range of the previous character of current character,
If the central point of current character is in the altitude range of the previous character of current character, current character and
Previous character belongs to same row, if the central point of current character is not in the previous character of current character
Altitude range in, then current character is belonging respectively to two different rows with previous character.
Wherein, the determination methods that a character enters every trade ownership are given in step A1 to A2, for step
The row ownership judgment mode of rapid A1 to A2 is applied to the row ownership determination process to each character on the page,
Wherein, using an optional character in the multiple characters on the page as current character, step A1 and
In step A2 so that the row ownership to current character judges as an example, all characters in the page can be used
To the row ownership judgment mode of current character.Wherein, the central point of current character refers to character in vertical seat
The median of altitude range on mark direction, for example, the altitude range of current character is (y1, y2), then word
Ordinate of the central point of symbol on ordinate direction is (y1+y2)/2, by the ordinate of current character
(y1+y2The altitude range of the previous character of)/2 and current character is judged, such that it is able to judge to work as
The row ownership of the previous character of preceding character and current character.
It should be noted that in some embodiments of the invention, step 101 is according to character on the page
Elevation information the multiple characters on the page are belonged on the page multiple rows on before, the present invention is implemented
The processing method of the character recognition that example is provided can also comprise the following steps：
The multiple original characters on multiple symbolic blocks identification page-out that B1, basis are partitioned into from the page；
B2, all originals according to the altitude range and width range of each original character on the page from the page
Excessive character is weeded out in beginning character or small characters are crossed, the multiple characters on the page are obtained.
Wherein, the original character on the page can be partitioned into single character from the page by OCR technique
And the character obtained after each character is identified, the height of each original character on the page is got first
Degree information and width information, wherein, the height of the character of elevation information and the foregoing teachings description of original character
Degree information is similar, and the width information of original character refers to：Abscissa of the original character on the page and
The width range of the original character, elevation information and width information by original character on the page can be with
Accurately determine original character on the fixed coordinates and abscissa direction on ordinate direction on the page
Fixed coordinates, and altitude range and width range of the original character on ordinate direction.Other step
Excessive character described in B2 refers to the word more than certain numerical value on altitude range and/or width range
Symbol, the mistake small characters described in step B2 are referred on altitude range and/or width range less than certain
The character of numerical value, excessive character is weeded out from all original characters on the page or small characters are crossed, can be with
Obtain the multiple characters on the page described in step 101.
Further, altitude ranges and width range of the above-mentioned steps B2 according to each original character on the page
Excessive character is weeded out from all original characters on the page or small characters are crossed, the multiple on the page is obtained
Character, specifically may include steps of：
B21, calculate the flat of the page respectively according to the altitude range and width range of each original character on the page
Equal character height and average character duration；
B22, altitude range is weeded out from all original characters on the page according to average character height it is more than
The original of original character or altitude range less than M times of average character height of N times of average character height
Beginning character, and from all original characters on the page to weed out width range according to average character duration big
In M times of N times of average character duration of original character or width range less than average character duration
Original character, completes to obtain the character on the page after rejecting, and N is the numerical value more than 1, and M is more than 0
Numerical value less than 1.
Wherein, the average character height of the page refers to the height of all original characters on the page in step B21
Average value, the average character duration of the page refers to the average value of the width of all original characters on the page,
The value of N can be average to need to weed out altitude range in 1.5, i.e. step B22 in step B22
The original character of 1.5 times of character height, it is also desirable to weed out that width range is average character duration 1.5
Times original character, the value of M can be to need to weed out in 0.2, i.e. step B22 in step B22
Altitude range is the original character of 0.2 times of average character height, it is also desirable to which it is average to weed out width range
The original character of 0.2 times of character duration, rejects from all original characters on the page in the manner described above
Fall excessive character or cross small characters, obtain the character on the page described in step 101, wherein N and M
Specific value can be not limited to it is foregoing for example, can be combined with specific application scenarios determine N,
The specific value of M.
In some embodiments of the invention, elevation information of the step 101 according to character on the page is by page
Multiple characters on face are belonged on the multiple rows on the page, obtain the multiple that the every a line on the page includes
After character, the processing method of character recognition provided in an embodiment of the present invention can also comprise the following steps：
C1, according between two neighboring character in the every a line on the page character pitch calculate the page on
Character field segmentation distance per a line；
C2, according to character field split distance multiple characters that the every a line on the page includes are segmented,
Obtain multiple character fields that the every a line on the page includes.
Wherein, in step C1, the width information of each character on the page, the width of character are got first
Information refers to：The width range of abscissa of the character on the page and the character, by character in the page
On width information can accurately determine fixed coordinates of the character on the abscissa direction on the page,
And width range of the character on ordinate direction.Then between two neighboring character in same row
Character pitch is exactly the abscissa difference between latter character and previous character, according to phase in a row
Character pitch between adjacent two characters calculates the character field segmentation distance of each row on the page,
Carried out to how to set character field segmentation distance according to the character pitch between two neighboring character in every a line
Set, wherein, character field be can split into the set of multigroup character in a row, illustrate such as
Under：The content recorded in a row in the page is as follows：
Name：Xiao Ming grade：Two classes of sexes of Third school grade：Man
Then having just in row as above can have 3 character fields, respectively " name：Xiao Ming ", " grade：
Two classes of Third school grade ", " sex：Man ".Predict after character field segmentation distance, split using the character field
Distance is segmented to multiple characters that every a line includes, obtains multiple words that the every a line on the page includes
Symbol section.
Further, step C1 is according to the character pitch between two neighboring character in the every a line on the page
The character field segmentation distance of the every a line on the page is calculated, specifically be may include steps of：
Character pitch in every a line on C11, calculating page-out between two neighboring character；
C12, descending row is carried out according to numerical values recited to the character pitch between two neighboring character in every a line
Row, the median in selection character pitch splits distance as the character field of the every a line on the page.
Wherein, after calculating the character pitch in every a line between two neighboring character, to getting
Have character pitch carries out descending arrangement according to numerical values recited, and selection is in the intercharacter of median in the ranking
Split distance away from as character field, the foundation that character field segmentation distance is split as character field, selection is all
Median in character pitch can accurately get the character in every a line as character field segmentation distance
Section.
Further, step C2 splits multiple words that distance includes to the every a line on the page according to character field
Symbol is segmented, and obtains multiple character fields that the every a line on the page includes, can specifically include following step
Suddenly：
C21, arbitrarily one character of selection as current character, obtains current from the multiple characters on the page
Character pitch between character and adjacent character；
If character pitch between C22, current character and adjacent character less than or equal to character field split away from
From, current character and adjacent character are divided into a character field, if current character and adjacent character it
Between character pitch more than character field split distance, by current character and adjacent character be divided into two it is different
Character field in.
Wherein, a determination methods for character field are given in step C21 to C22, for step C21
Character field judgment mode to C22 is applied to the character field determination process of each character on the page, wherein,
An optional character is used as current character, step C21 and step using in the multiple characters on the page
In C22 so that the character field to current character judges as an example, all characters in the page can use right
The character field judgment mode of current character.
102nd, according to the overlay information between character on the page on altitude range to each the row bag on the page
The multiple characters for including enter every trade correction, obtain the multiple characters after the row correction that the every a line on the page includes.
In embodiments of the present invention, for all characters on the page, obtain on the page between all characters
There is the character of overlap on altitude range, every trade is entered to there is the character for overlapping between character on altitude range
Correction, wherein, on the page between character on altitude range exist overlap may determine that it is big on page-out
The multiple characters in same row are caused, because belonging to the height of multiple characters of same row on the page
Scope is all similar, if because when character produces inclination in the picture of the typesetting of character or the shooting page,
Some characters may be divided into the row of mistake, therefore character in the embodiment of the present invention on to the page is returned
Belong to after multiple rows, multiple characters that each row on the page includes can also be entered according to overlay information
Every trade is corrected, thus correct be likely to occur enter the misjudgment that produces when every trade belongs to character.
In some embodiments of the invention, step 102 according between character on the page on altitude range
Overlay information multiple characters that each row on the page includes are entered every trade correction, specifically can include such as
Lower step：
D1, arbitrarily one character of selection as current character, obtains height from the multiple characters on the page
Scope has overlap multiple characters with the altitude range of current character；
If D2, the altitude range for getting have overlap multiple characters all to belong to the altitude range of current character
In same row, then the row where keeping current character is constant；
If D3, the altitude range for getting have overlap multiple characters to distinguish with the altitude range of current character
Belong to two rows, altitude range has overlap with the altitude range of current character during two rows are calculated respectively
The number of character, the row where current character is defined as into altitude range has with the altitude range of current character
The most row of the number of the character of overlap.
Wherein, the method that a character enters every trade correction is given in step D1 to D2, for step
The row correction judgment mode of D1 to D3 is applied to the determination process to the row correction of each character on the page,
Wherein, using an optional character in the multiple characters on the page, used as current character, step D1 is extremely
In step D3 by taking the row correction to current character as an example, all characters in the page can be used to working as
The row correcting mode of preceding character.Wherein, altitude range has overlap multiple with the altitude range of current character
Character is probably same row, it is also possible to two adjacent rows, is sentenced by way of statistics in step D3
Break and the row that current character should belong to, after can judging entering every trade ownership in former step 101 further
Realization row correction, such that it is able to realize the character recognition of high accuracy.
103rd, the multiple characters after the row included to the every a line on the page using semantic analysis model is corrected enter
Row missed suppression.
In embodiments of the present invention, the multiple characters after the row correction that each row on the page includes are obtained
Afterwards, missed suppression can be carried out to above-mentioned character according to default semantic analysis model, wherein, this hair
The semantic analysis model used in bright embodiment can be word2vec, or HMM etc.,
The further optimization that missed suppression can be realized to character identification result is carried out to character, is more met language
Say the character recognition effect of custom.
In previously described embodiments of the present invention, if performing step C1 and C2, carried out by step 102
After row correction, all include multiple character fields, each character field bag in multiple character fields on the page per a line
Include：Row correction after multiple characters, it is this realize scene under, step 103 use semantic analysis model
Multiple characters after the row correction included to the every a line on the page carry out missed suppression, including：
E1, using semantic analysis model to information and the intersegmental letter of character in character field in the every a line on the page
Breath carries out missed suppression respectively.
Wherein, if having got the character field that the every a line on the page includes, can it is intersegmental to character and
Missed suppression is carried out respectively in character field, and the process that specifically used semantic analysis model carries out missed suppression can
Refering to prior art.
Description by above example to the embodiment of the present invention, first according to character on the page
Elevation information belongs to the multiple characters on the page on the multiple rows on the page, obtains each on the page
Multiple characters that row includes, elevation information of the character on the page includes：Vertical seat of the character on the page
The altitude range of mark and the character, next believes according to the overlap between character on the page on altitude range
Breath enters every trade correction to multiple characters that each row on the page includes, the every a line obtained on the page includes
Row correction after multiple characters, the row for finally being included to the every a line on the page using semantic analysis model
Multiple characters after correction carry out missed suppression.Elevation information in the present invention using character on the page will
All characters on the page belong to multiple rows, due to same a line ordinate of the character on the page and
The altitude range of character is all relatively fixed, thus according to character elevation information to all characters on the page
Carry out branch to branch's result be accurate, and can also be according between character on the page in the present invention
Overlay information on altitude range enters every trade correction to multiple characters that each row includes, therefore can lead to
The row that lap over infomation detection should belong to character in the character typesetting to different-format, is believed by overlapping
Breath there may be inclined character in may also detect that the parent page of shooting, may be used also in the present invention in addition
Missed suppression is carried out to character with using semantic analysis model, therefore enters every trade correction and semantic school to character
Can just change causes because there is inclined character in the form difference of character typesetting and shooting parent page
Identification mistake, improve character recognition recognition effect.
For ease of being better understood from and implementing the such scheme of the embodiment of the present invention, illustrate accordingly should below
It is specifically described with scene.For getting split character block in the embodiment of the present invention, not only
Employ mutual logical relation between character block, and row is not lined up in itself to also allow for character in branch
Problem, i.e., row correction when captured character row is not horizontal.
Due to many people in current OCR technique it is contemplated that Character segmentation and identification division, for final
Display result, split by character row and merged, can be to OCR the semantic analysis that needs to use is combined
Whole structure have very big lifting, especially generation OCR character rows it is not parallel, have various situations such as noise
Under, the embodiment of the present invention can well carry out character row fractionation, and usage scenario is extensive.One kind of the invention
Application scenarios, specifically may include steps of, and refer to as shown in Fig. 2-a.
The single character that step 1, OCR are identified
In the embodiment of the present invention, first by OCR technique to single Character segmentation, identification.First with
Character segmentation method (method such as such as image binaryzation, convex closure) in OCR technique split after it is each
Individual character block, then by the way that relation between logical sum character block is rational single Character segmentation and identifies original
Beginning character.Recognition methods is such as：Based on the matching of convex closure profile, the Gradient Features matching based on gradation of image
Deng, the character and its recognition result of all segmentations under a page have been obtained, based on the above results, connect
Each fritter for having obtained each character by methods such as binaryzation, convex closures to constitute, such as character " small "
The fritter of three parts is obtained, respectively：Zuo Dian, erects hook, right point, is recognized by combining each character block,
Finally identify that this three pieces of characters merge the Chinese character of gained.
Noise on step 2, the removal page
Char_height (is used by the width (being represented with char_width) and height that calculate overall character in the page
Represent) summation, obtain average character duration (char_average_width) on the page and averagely
Character height (char_average_height), then removes from the original character of the page and is more than
The original character of 1.5*char_average_width or 1.5*char_average_height, because in a word
In the page of symbol, it is all based on the distribution of character in most cases, and character is distributed with certain rule
Rule property, or on height wide unanimously (for example：State and), or it is wide consistent (for example：First, two),
Height is consistent (for example such as：！, [etc.), total some larger blocks or less piece can in character recognition
Influence branch's effect of character, such as relatively large character height that can take two rows makes during remerging
The character field that currently must be originally divided into two rows merges into a character field, so as to influence final character recognition
As a result.
Step 3, the calculating page are per a line and character field
First, the preliminary split result of the character branch being calculated on the page.Obtained using in step 1
Rational character distribution, the merging for entering line character according to following logic (is used so as to obtain preliminary character field
Char_section is represented).
(1), to traveling row label belonging to each character：The position of current character can be defined as
(char_x, char_y, char_width, char_height), wherein, char_x is current character upper left corner place
The x coordinate of the page, char_y is the y-coordinate of the page where the current character upper left corner, and char_width is to work as
Preceding character duration, the width range of current character is (char_x, char_x+char_width), char_height
For current character highly, the altitude range of current character is (char_y, char_y+char_height).Profit
With character location information affiliated here, the y-coordinate that can calculate the central point of each character is：
Char_y+char_height/2, if the central point y-coordinate of current character is in the previous character institute of current character
Altitude range in, then current character and previous character are belonging to a line, formulate as follows：
chari+1_y+chari+1_height/2∈[chari_y,chari_y+chari_height]；
Wherein, i=(1,2,3 ... .n), n are total number of characters of current page.
Each character can be belonged on corresponding row according to the above method, thus obtain current
The first walking property distribution of page character.
(2) character field merged into line character is calculated to the affiliated character per a line and splits distance：
After by above, the method for (1) obtains the affiliated character of every a line, calculated currently using following methods
Character pitch in row between connected two characters：
distancek=chari+1_x-(chari_x+chari_width)；
Wherein i=1,2,3 ... m, k=1,2,3 ... ..m-1, m are the number of characters in current line.
Each adjacent character spacing (distance) to being stored is ranked up according to descending, inside selection in
Between value (i.e. distance_sort_middle) as current line character field split distance.
(3) split distance according to character field to be segmented the character of current line, obtain each word of current line
Symbol section：
Character for the character pitch in current line more than distance_sort_middle is split to two words
In symbol section, it is merged into same character field for the character less than distance_sort_middle.Wherein,
Here fractionation is exactly that, separately as new set expression, merging is that current character is merged current character
It is used to build character field in the set for meet condition.
The segment information of character in the affiliated row of character and row under a page has been obtained according to the method described above, so
Step 4 is performed afterwards.
Step 4, character is entered every trade correction
Wherein, to the character correction character row attribute on the page：Entering every trade Attribute transposition process to character
It is middle because shooting angle problem occur this for a line character no longer horizontal direction on, as shown in figure Fig. 2-b,
Each frame is a character distribution, originally belongs to the character of the second row because the outside cause such as shooting angle is marked
The attribute of the first row is noted, the character row that wire frame frame long is outlined with point frame is according to the method in step 3
Obtain, can be as follows using the method for correction in this step 4：
(1) the row attribute of all characters of current page, is obtained in the case where step 3 method is completed, to each character
The row attaching information of current character is calculated as follows：
Current character line range：chari_ range=[chari_y,chari_y+chari_height]；
If chariThe char of _ range and another characterj_ range has overlap, then right
chariThis one-dimensional record array relevant position of _ line_record increases a record value.Finally compare two
It is individually present in row after the number of the character of overlap, takes chari_ line_record the insides numerical value highest institute is right
The current char of behavior for answeringiRow attribute.By taking Fig. 2-b as an example, for last in two rows in the page
Character, chari_ line_record array lengths are 2, are judged using current character line range, can be obtained
It is 5, char to the character number that Fig. 2-b last characters are overlapped in the first rowi_ line_record [0]=5
And the character number for overlapping in a second row is：6, chari_ line_record [1]=6, at this moment in selection array
Numerical value highest 6 is expert at (i.e. the second row) as the row attribute of current character, then last in Fig. 2-b
The row attribute of one character is defined as the second row.
Wherein, i, j=(1,2,3 ... .n), n is total number of characters of current page, chari_ range is one
Individual character y directions distribution, chari_ line_record is the dimension group that initial value is 0,
Number of dimensions is total line number resulting under step 3 method.
Step 5, missed suppression is carried out to information in character field and the intersegmental information of character
For the segmented each character field for having merged, character field can again be entered using semantic analysis model
Row missed suppression, and be that the semantic analysis model of each character content fusion of identification in character field is sentenced
Disconnected, semantic analysis model can be using the technology of current comparative maturity such as：Word2vec, or hidden Ma Er
Section's husband's model etc..So it is corrected for itself wrong part of identification in character field, is such as originally used for depth
Zhen Shi, is erroneously identified as Shen Xun cities etc., intersegmental for character, then can be entered using semantic analysis model
The further missed suppression of row, please refers to as shown in Fig. 2-c and Fig. 2-d respectively, and Fig. 2-c are for before missed suppression
Character field schematic diagram, Fig. 2-d be missed suppression after character field schematic diagram, according to semantic analysis model,
" knot " and " fruit " should belong to a semantic section, therefore can carry out the merging of semantic section.
By the foregoing citing description of this invention, the present invention can utilize character pitch and character institute
In the position of the page, character content and semantic analysis in itself is merged, the fractionation to partial character row is rectified
Just, so that character row segmentation more rationally, logic of language is more met on result is presented.By a large amount of
Experiment test prove that method provided in an embodiment of the present invention can be than other original methods to OCR words
In the segmentation of symbol branch more rationally, the custom of language and is more met in terms of content and in charcter topology arrangement.
It should be noted that for foregoing each method embodiment, in order to be briefly described, therefore by its all table
It is a series of combination of actions to state, but those skilled in the art should know, the present invention does not receive to be retouched
The limitation of the sequence of movement stated because according to the present invention, some steps can using other order or
Carry out simultaneously.Secondly, those skilled in the art should also know, embodiment described in this description
Preferred embodiment is belonged to, necessary to involved action and the module not necessarily present invention.
For ease of preferably implementing the such scheme of the embodiment of the present invention, it is also provided below for implementation
State the relevant apparatus of scheme.
Refer to shown in Fig. 3-a, a kind of processing unit 300 of character recognition provided in an embodiment of the present invention,
Can include：Row splits module 301, row correction module 302, missed suppression module 303, wherein,
Row splits module 301, for the elevation information according to character on the page by the multiple on the page
Character is belonged on the multiple rows on the page, obtains multiple words that the every a line on the page includes
Symbol, elevation information of the character on the page includes：Character ordinate on the page and should
The altitude range of character；
Row correction module 302, for according to the overlay information between character on the page on altitude range
Multiple characters that each row on the page includes are entered with every trade correction, obtains each on the page
Multiple characters after the row correction that row includes；
Missed suppression module 303, for what is included to the every a line on the page using semantic analysis model
Multiple characters after row correction carry out missed suppression.
In some embodiments of the invention, as shown in Fig. 3-b, the row splits module 301, including：
Character center point determining unit 3011, for any selection one from the multiple characters on the page
Individual character calculates described current as current character, the elevation information according to the current character on the page
The central point of character ordinate on the page；
Row judging unit 3012, for judging the central point of the current character whether in the current character
Previous character altitude range in, if the central point of current character is in the previous of the current character
In the altitude range of character, then the current character and the previous character belong to same row, if working as
The central point of preceding character is not in the altitude range of the previous character of the current character, then described current
Character is belonging respectively to two different rows with the previous character.
In some embodiments of the invention, as shown in Fig. 3-c, the row correction module 302, including：
High superposed character determining unit 3021, for arbitrarily being selected from the multiple characters on the page
Used as current character, obtain altitude range has overlap to one character with the altitude range of the current character
Multiple characters；
Row determining unit 3022, if for the altitude range of the altitude range that gets and the current character
The multiple characters for having overlap belong to same row, then the row where keeping the current character is constant；If
The altitude range for getting has overlap multiple characters to be belonging respectively to two with the altitude range of the current character
Individual row, altitude range has overlap word with the altitude range of the current character during two rows are calculated respectively
The number of symbol, the row where the current character is defined as the height of altitude range and the current character
Scope has the most row of the number of the character of overlap.
In some embodiments of the invention, as shown in Fig. 3-d, the processing unit 300 of the character recognition,
Also include：Character denoising module 304, height of the module 301 according to character on the page is split for the row
Before on multiple rows that degree information belongs to the multiple characters on the page on the page, according to from
The multiple symbolic blocks being partitioned on the page identify the multiple original characters on the page；According to institute
State all original characters of the altitude range and width range of each original character on the page from the page
In weed out excessive character or cross small characters, obtain the multiple characters on the page.
In some embodiments of the invention, as shown in Fig. 3-e, the processing unit 300 of the character recognition,
Also include：Character field determining module 305, for the row split module 301 according to character on the page
Elevation information belongs to the multiple characters on the page on the multiple rows on the page, obtains described
After multiple characters that every a line on the page includes, according to two neighboring in the every a line on the page
Character pitch between character calculates the character field segmentation distance of the every a line on the page；According to described
Character field segmentation distance is segmented to multiple characters that the every a line on the page includes, obtains described
Multiple character fields that every a line on the page includes.
In some embodiments of the invention, each character field includes in the multiple character field：Row correction
Character afterwards；
The missed suppression module 303, specifically for using semantic analysis model to each on the page
Information and the intersegmental information of character carry out missed suppression respectively in character field in row.
Description more than to the embodiment of the present invention, the height letter first according to character on the page
Breath belongs to the multiple characters on the page on the multiple rows on the page, and the every a line obtained on the page includes
Multiple characters, elevation information of the character on the page include：Ordinate of the character on the page and should
The altitude range of character, next according to the overlay information between character on the page on altitude range to page
Multiple characters that each row on face includes enter every trade correction, obtain the row school that the every a line on the page includes
Multiple characters after just, after the row for finally being included to the every a line on the page using semantic analysis model is corrected
Multiple characters carry out missed suppression.Using elevation information of the character on the page by the page in the present invention
All characters belong to multiple rows, due to the ordinate and character of the character on the page of same a line
Altitude range is all relatively fixed, therefore elevation information according to character is divided all characters on the page
The branch's result gone is accurate, and in the present invention can also according between character on the page height
Overlay information in scope enters every trade correction to multiple characters that each row includes, therefore can be by overlapping
The row that infomation detection should belong to character in the character typesetting to different-format, also may be used by overlay information
There may be inclined character in the parent page for detecting shooting, can also be utilized in the present invention in addition
Semantic analysis model carries out missed suppression to character, therefore enters every trade correction to character and missed suppression can be with
Change because the form of character typesetting is different and shoots in parent page to exist and recognize caused by inclined character
Mistake, improves the recognition effect of character recognition.
Fig. 4 is that the processing method of character recognition provided in an embodiment of the present invention is applied to a kind of knot of server
Structure schematic diagram, the server 400 be able to can be wrapped because of configuration or performance is different and the larger difference of producing ratio
One or more central processing units (central processing units, CPU) 422 is included (for example, one
Individual or more than one processor) and memory 432, one or more storage application programs 442 or number
According to 444 storage medium 430 (such as one or more mass memory units).Wherein, memory
432 and storage medium 430 can be it is of short duration storage or persistently storage.Store the program in storage medium 430
One or more modules (diagram is not marked) can be included, each module can include in server
Series of instructions operation.Further, central processing unit 422 could be arranged to and storage medium 430
Communication, the series of instructions operation in performing storage medium 430 on server 400.
Server 400 can also include one or more power supplys 426, one or more it is wired or
Radio network interface 450, one or more input/output interfaces 458, and/or, one or one with
Upper operating system 441, such as Windows ServerTM, Mac OS XTM, UnixTM, LinuxTM,
FreeBSDTM etc..
Step in above-described embodiment as performed by server can be based on the character recognition shown in the Fig. 1
Processing method.
In addition it should be noted that, device embodiment described above is only schematical, wherein described
The unit illustrated as separating component can be or may not be it is physically separate, it is aobvious as unit
The part for showing can be or may not be physical location, you can with positioned at a place, or also may be used
To be distributed on multiple NEs.Some or all of mould therein can according to the actual needs be selected
Block realizes the purpose of this embodiment scheme.In addition, in the device embodiment accompanying drawing of present invention offer, mould
Annexation between block represents between them there is communication connection, specifically can be implemented as one or more
Communication bus or holding wire.Those of ordinary skill in the art without creative efforts, i.e.,
It is appreciated that and implements.
Through the above description of the embodiments, it is apparent to those skilled in the art that originally
Invention can add the mode of required common hardware to realize by software, naturally it is also possible to by specialized hardware
Realized including application specific integrated circuit, dedicated cpu, private memory, special components and parts etc..General feelings
Under condition, all functions of being completed by computer program can be realized easily with corresponding hardware, and
And, the particular hardware structure for realizing same function can also be it is diversified, such as analog circuit,
Digital circuit or special circuit etc..But, it is more for the purpose of the present invention in the case of software program realize be more
Good implementation method.Based on such understanding, technical scheme is substantially in other words to existing skill
The part that art contributes can be embodied in the form of software product, computer software product storage
In the storage medium that can read, such as computer floppy disk, USB flash disk, mobile hard disk, read-only storage (ROM,
Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic
Dish or CD etc., including some instructions are used to so that computer equipment (can be personal computer,
Server, or the network equipment etc.) perform method described in each embodiment of the invention.
In sum, the above embodiments are merely illustrative of the technical solutions of the present invention, rather than its limitations；
Although being described in detail to the present invention with reference to above-described embodiment, one of ordinary skill in the art should
Work as understanding：It can still modify to the technical scheme described in the various embodiments described above, or to it
Middle some technical characteristics carry out equivalent；And these modifications or replacement, do not make appropriate technical solution
Essence depart from various embodiments of the present invention technical scheme spirit and scope.
Claims (12)
1. a kind of processing method of character recognition, it is characterised in that including：
Multiple characters on the page are belonged to the page by the elevation information according to character on the page
On multiple rows on, obtain multiple characters that the every a line on the page includes, the character is in the page
On elevation information include：The altitude range of character ordinate on the page and the character；
According to the overlay information between character on the page on altitude range to each on the page
Multiple characters that row includes enter every trade correction, after obtaining the row correction that the every a line on the page includes
Multiple characters；
Entered using the multiple characters after the row correction that semantic analysis model includes to the every a line on the page
Row missed suppression.
2. method according to claim 1, it is characterised in that it is described according to character on the page
Elevation information belongs to the multiple characters on the page on the multiple rows on the page, obtains described
Multiple characters that every a line on the page includes, including：
A character is arbitrarily selected from the multiple characters on the page as current character, according to described
Elevation information of the current character on the page calculates the central point of the current character on the page vertical
Coordinate；
Judge the current character central point whether the current character previous character height model
In enclosing, if the central point of current character is in the altitude range of the previous character of the current character,
The current character and the previous character belong to same row, if the central point of current character is not in institute
State in the altitude range of previous character of current character, then the current character and the previous character
It is belonging respectively to two different rows.
3. method according to claim 1, it is characterised in that described according to character on the page
Between overlay information on altitude range every trade is entered to multiple characters that each row on the page includes
Correction, including：
Arbitrarily one character of selection obtains height as current character from the multiple characters on the page
Scope has overlap multiple characters with the altitude range of the current character；
If the altitude range for getting has overlap multiple characters all to belong to the altitude range of the current character
In same row, then the row where keeping the current character is constant；
If the altitude range for getting has overlap multiple characters to distinguish with the altitude range of the current character
Belong to two rows, altitude range has weight with the altitude range of the current character during two rows are calculated respectively
The number of folded character, altitude range is defined as with the current character by the row where the current character
Altitude range have the most row of the number of the character of overlap.
4. method according to claim 1, it is characterised in that it is described according to character on the page
It is described before on multiple rows that elevation information belongs to the multiple characters on the page on the page
Method also includes：
Multiple symbolic blocks according to being partitioned into from the page identify the multiple original words on the page
Symbol；
The institute of altitude range and width range according to each original character on the page from the page
Have in original character and weed out excessive character or cross small characters, obtain the multiple characters on the page.
5. method according to any one of claim 1 to 4, it is characterised in that described according to word
Accord with multiple rows that the elevation information on the page belongs to the multiple characters on the page on the page
On, obtaining after multiple characters that the every a line on the page includes, methods described also includes：
The page is calculated according to the character pitch between two neighboring character in the every a line on the page
On every a line character field segmentation distance；
Split distance according to the character field to divide multiple characters that the every a line on the page includes
Section, obtains multiple character fields that the every a line on the page includes.
6. method according to claim 5, it is characterised in that each word in the multiple character field
Symbol section includes：Multiple characters after row correction；
Multiple words after the row correction that the use semantic analysis model includes to the every a line on the page
Symbol carries out missed suppression, including：
Using semantic analysis model to information and the intersegmental letter of character in character field in the every a line on the page
Breath carries out missed suppression respectively.
7. a kind of processing unit of character recognition, it is characterised in that including：
Row splits module, for the elevation information according to character on the page by the multiple words on the page
Symbol is belonged on the multiple rows on the page, obtains multiple characters that the every a line on the page includes,
Elevation information of the character on the page includes：Character ordinate on the page and the character
Altitude range；
Row correction module, for according to the overlay information pair between character on the page on altitude range
Multiple characters that each row on the page includes enter every trade correction, obtain the every a line on the page
Including row correction after multiple characters；
Missed suppression module, for the row included to the every a line on the page using semantic analysis model
Multiple characters after correction carry out missed suppression.
8. device according to claim 7, it is characterised in that the row splits module, including：
Character center point determining unit, for arbitrarily selecting a word from the multiple characters on the page
Symbol calculates the current character as current character, the elevation information according to the current character on the page
Central point ordinate on the page；
Row judging unit, for judging the central point of the current character whether before the current character
In one altitude range of character, if the central point of current character is in the previous character of the current character
Altitude range in, then the current character and the previous character belong to same row, if current word
The central point of symbol not in the altitude range of the previous character of the current character, then the current character
Two different rows are belonging respectively to the previous character.
9. device according to claim 7, it is characterised in that the row correction module, including：
High superposed character determining unit, for any selection one from the multiple characters on the page
Used as current character, obtain altitude range has overlap multiple to character with the altitude range of the current character
Character；
Row determining unit, if the altitude range and the altitude range of the current character for getting have weight
Folded multiple characters belong to same row, then the row where keeping the current character is constant；If obtaining
To altitude range there are overlap multiple characters to be belonging respectively to two with the altitude range of the current character
OK, altitude range has overlap character with the altitude range of the current character during two rows are calculated respectively
Number, the row where the current character is defined as the height model of altitude range and the current character
It is with the most row of the number of the character of overlap.
10. device according to claim 7, it is characterised in that the treatment dress of the character recognition
Put, also include：Character denoising module, height of the module according to character on the page is split for the row
Before on multiple rows that information belongs to the multiple characters on the page on the page, according to from institute
State multiple original characters that the multiple symbolic blocks being partitioned on the page are identified on the page；According to described
The altitude range and width range of each original character are from all original characters on the page on the page
Weed out excessive character or cross small characters, obtain the multiple characters on the page.
11. device according to any one of claim 7 to 10, it is characterised in that the character
The processing unit of identification, also includes：Character field determining module, module is split according to character for the row
Multiple characters on the page are belonged to elevation information on the page the multiple rows on the page
On, after obtaining multiple characters that the every a line on the page includes, according to each on the page
In row character pitch between two neighboring character calculate every a line on the page character field split away from
From；Split distance according to the character field to divide multiple characters that the every a line on the page includes
Section, obtains multiple character fields that the every a line on the page includes.
12. devices according to claim 11, it is characterised in that in the multiple character field each
Character field includes：Character after row correction；
The missed suppression module, specifically for using semantic analysis model to the every a line on the page
Information and the intersegmental information of character carry out missed suppression respectively in middle character field.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510410166.5A CN106709489B (en) | 2015-07-13 | 2015-07-13 | Character recognition processing method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510410166.5A CN106709489B (en) | 2015-07-13 | 2015-07-13 | Character recognition processing method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106709489A true CN106709489A (en) | 2017-05-24 |
| CN106709489B CN106709489B (en) | 2020-03-03 |
Family
ID=58898678
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510410166.5A Active CN106709489B (en) | 2015-07-13 | 2015-07-13 | Character recognition processing method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106709489B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107437084A (en) * | 2017-07-24 | 2017-12-05 | 南京晓庄学院 | A kind of character center of gravity localization method of line Handwritten text identification |
| CN108182432A (en) * | 2017-12-28 | 2018-06-19 | 北京百度网讯科技有限公司 | Information processing method and device |
| CN109871743A (en) * | 2018-12-29 | 2019-06-11 | 口碑(上海)信息技术有限公司 | The localization method and device of text data, storage medium, terminal |
| CN110135417A (en) * | 2018-02-09 | 2019-08-16 | 北京世纪好未来教育科技有限公司 | Sample mask method and computer storage medium |
| CN110378347A (en) * | 2019-07-04 | 2019-10-25 | 北京爱医生智慧医疗科技有限公司 | A kind of the key message extracting method and device of medical inspection list |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP3435374B2 (en) * | 1999-10-04 | 2003-08-11 | 沖電気工業株式会社 | Character reading device and character recognition method |
| CN102024139A (en) * | 2009-09-18 | 2011-04-20 | 富士通株式会社 | Device and method for recognizing character strings |
| CN102779275A (en) * | 2012-07-04 | 2012-11-14 | 广州广电运通金融电子股份有限公司 | Paper characteristic identification method and relative device |
| CN103577818A (en) * | 2012-08-07 | 2014-02-12 | 北京百度网讯科技有限公司 | Method and device for recognizing image characters |
| CN104683629A (en) * | 2013-11-26 | 2015-06-03 | 柯尼卡美能达株式会社 | Image forming apparatus, text data embedding method |
-
2015
- 2015-07-13 CN CN201510410166.5A patent/CN106709489B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP3435374B2 (en) * | 1999-10-04 | 2003-08-11 | 沖電気工業株式会社 | Character reading device and character recognition method |
| CN102024139A (en) * | 2009-09-18 | 2011-04-20 | 富士通株式会社 | Device and method for recognizing character strings |
| CN102779275A (en) * | 2012-07-04 | 2012-11-14 | 广州广电运通金融电子股份有限公司 | Paper characteristic identification method and relative device |
| CN103577818A (en) * | 2012-08-07 | 2014-02-12 | 北京百度网讯科技有限公司 | Method and device for recognizing image characters |
| CN104683629A (en) * | 2013-11-26 | 2015-06-03 | 柯尼卡美能达株式会社 | Image forming apparatus, text data embedding method |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107437084A (en) * | 2017-07-24 | 2017-12-05 | 南京晓庄学院 | A kind of character center of gravity localization method of line Handwritten text identification |
| CN107437084B (en) * | 2017-07-24 | 2020-12-08 | 南京晓庄学院 | Character gravity center positioning method for off-line handwritten text recognition |
| CN108182432A (en) * | 2017-12-28 | 2018-06-19 | 北京百度网讯科技有限公司 | Information processing method and device |
| US10963760B2 (en) | 2017-12-28 | 2021-03-30 | Beijing Baidu Netcom Science And Technology Co., Ltd. | Method and apparatus for processing information |
| CN110135417A (en) * | 2018-02-09 | 2019-08-16 | 北京世纪好未来教育科技有限公司 | Sample mask method and computer storage medium |
| CN109871743A (en) * | 2018-12-29 | 2019-06-11 | 口碑(上海)信息技术有限公司 | The localization method and device of text data, storage medium, terminal |
| CN110378347A (en) * | 2019-07-04 | 2019-10-25 | 北京爱医生智慧医疗科技有限公司 | A kind of the key message extracting method and device of medical inspection list |
| CN110378347B (en) * | 2019-07-04 | 2021-10-08 | 北京爱医生智慧医疗科技有限公司 | Method and device for extracting key information of medical examination sheet |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106709489B (en) | 2020-03-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110084292B (en) | Target detection method based on DenseNet and multi-scale feature fusion | |
| CN109816012B (en) | Multi-scale target detection method fusing context information | |
| CN108960211B (en) | Multi-target human body posture detection method and system | |
| CN110991311B (en) | Target detection method based on dense connection deep network | |
| CN106709489A (en) | Processing method and device of character identification | |
| CN106570453B (en) | Method, device and system for pedestrian detection | |
| CN110738207A (en) | character detection method for fusing character area edge information in character image | |
| CN108537824B (en) | Feature map enhanced network structure optimization method based on alternating deconvolution and convolution | |
| CN110738101A (en) | Behavior recognition method and device and computer readable storage medium | |
| CN106960195A (en) | A kind of people counting method and device based on deep learning | |
| CN110598788B (en) | Target detection method, target detection device, electronic equipment and storage medium | |
| CN109492596B (en) | Pedestrian detection method and system based on K-means clustering and regional recommendation network | |
| CN106548169A (en) | Fuzzy literal Enhancement Method and device based on deep neural network | |
| CN105303163B (en) | A kind of method and detection device of target detection | |
| CN109472193A (en) | Method for detecting human face and device | |
| CN114140683A (en) | Aerial image target detection method, equipment and medium | |
| CN109685145A (en) | A kind of small articles detection method based on deep learning and image procossing | |
| Wang et al. | Deep learning model for target detection in remote sensing images fusing multilevel features | |
| CN114783021A (en) | Intelligent detection method, device, equipment and medium for wearing of mask | |
| Hung et al. | Skyline localization for mountain images | |
| CN105023264A (en) | Infrared image remarkable characteristic detection method combining objectivity and background property | |
| CN111027551B (en) | Image processing method, apparatus and medium | |
| CN104463896A (en) | Image corner point detection method and system based on kernel similar region distribution characteristics | |
| CN111797737A (en) | Remote sensing target detection method and device | |
| CN116778386A (en) | Sulfur hexafluoride leakage detection method, device, equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |