CN109409363A - The reverse judgement of text image based on content and bearing calibration - Google Patents
The reverse judgement of text image based on content and bearing calibration Download PDFInfo
- Publication number
- CN109409363A CN109409363A CN201811192521.6A CN201811192521A CN109409363A CN 109409363 A CN109409363 A CN 109409363A CN 201811192521 A CN201811192521 A CN 201811192521A CN 109409363 A CN109409363 A CN 109409363A
- Authority
- CN
- China
- Prior art keywords
- literal line
- cost function
- function value
- text
- row
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/62—Text, e.g. of license plates, overlay texts or captions on TV images
- G06V20/63—Scene text, e.g. street names
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/24—Aligning, centring, orientation detection or correction of the image
- G06V10/242—Aligning, centring, orientation detection or correction of the image by image rotation, e.g. by 90 degrees
Abstract
The present invention provides a kind of reverse judgement of the text image based on content and bearing calibration.Technical solution includes the following steps: S1, carries out row cutting to text image;S2 rejects non-legible row, retains literal line;S3 takes any literal line L, so that the length of literal line L meets the range of setting;S4 carries out optical character identification to literal line L, calculates the cost function value of literal line L;S5, reverse literal line L obtain literal line L1;S6, to literal line L1Optical character identification is carried out, literal line L is calculated1Cost function value;S7, according to the cost function value of literal line L and literal line L1Cost function value, judge whether literal line L overturns;If literal line L is reverse, reverse correction is carried out.The main purposes of the present invention first is that text image that may be present overturns problem after processing text image slant correction, there is high accuracy, be easy to the features such as practical.
Description
Technical field
The present invention relates to text extracting technical field, more particularly, to a kind of judging automatically of overturning of text image and
Bearing calibration.
Background technique
Image is one of information age most important information carrier.In image processing techniques, text image pre-processes and divides
Analysis has application to the industries such as education electronics, secretarial's office, there is important meaning and effect.
In text image pretreatment, the text image of intake can be under normal circumstances inclined;But in subsequent analysis
In, the OCR (Optical Character Recognition, optical character identification) of Text region is such as carried out to text image
In the process, character to be identified is usually but required to be specification, rectify.Therefore, the slant correction of text image, be OCR it
The pretreated important content of preceding text image is printed page analyses and the final OCR such as row cutting, the character locating of text image
The important leverage of accuracy.
Currently, when calculating text image true slope angle, if the true slope angle of the text image of intake compared with
Greatly, calculated text image tilt angle may be the supplementary angle of text image true slope angle.Using this result to text
This image carries out slant correction, and obtained image will be the reverse figure of real text image.
Therefore, before carrying out printed page analysis and OCR using the text image by slant correction, it is necessary to advanced style of writing
The reverse judgement and correction of this image.In consideration of it, the present invention is specifically proposed.
Summary of the invention
It is an object of the present invention to propose a kind of text based on content for there may be the text image that text overturns
The judgement of this image inversion and bearing calibration.This method makes full use of the grammatical characteristic such as word, word, phrase of text, judges text
Whether image overturns, and realizes the correction to reverse text image.
In order to achieve the object, the technical solution of the invention is as follows: a kind of reverse judgement of the text image based on content
And bearing calibration, if text image has been subjected to slant correction, which comprises the following steps:
S1: row cutting is carried out to text image;
S2: rejecting non-legible row, retains literal line;
S3: taking any literal line L, so that the length of literal line L meets the range of setting;
S4: OCR is carried out to literal line L, calculates the cost function value of literal line L;
S5: reverse literal line L obtains literal line L1;
S6: to literal line L1OCR is carried out, literal line L is calculated1Cost function value;
S7: according to the cost function value of literal line L and literal line L1Cost function value, judge whether literal line L overturns;
If literal line L is reverse, reverse correction is carried out.
Compared with prior art, the beneficial effects of the present invention are:
(1) according to the retrieval to existing literature, it is not yet found that closing the money for carrying out reverse judgement and correction to text image
Material, but but through being commonly encountered this problem in actually using.Therefore, the present invention handle text image slant correction after there may be
Text image overturn problem, have innovation and practicality.
(2) present invention takes full advantage of the grammatical characteristic of the different language obtained after OCR.If text diagram is positive
, the literal line handled is positive, the word or character obtained after OCR, can correctly, show as normal
The probability occurred in rule application is high.It is corresponding, literal line obtained in figure is overturned from it, is knot that is reverse, being obtained after OCR
Fruit substantially can be out-of-order word or incorrect character.Attached drawing 2 to attached drawing 13 gives several examples, is all over Britain respectively
Positive literal line, reverse literal line under the different situations such as Chinese character, full Chinese character, the English mixed characters of Chinese and they are right
The ocr result answered.At this point, the cost function value of the positive literal line in positive figure, can be substantially distinguished from reverse in reverse figure
The cost function value of literal line simply can obtain correctly positive figure by contrast judgement.
(3) contrast judgement forward direction figure of the present invention and reverse figure, rather than one width figure of independent judgment, avoid using absolute threshold
Value, thus have the characteristics that it is very steady.Chinese character for example, a certain width forward direction figure, in its processed literal line that may happen to
Or English word, the probability occurred in conventional document is all not too much high, if judging whether to overturn based on some absolute threshold,
It is likely to occur erroneous judgement.But schemed using forward direction and overturn both figures contrast judgement, then the cost function value of positive figure, still has
Very big probability is better than the cost function value of reverse figure, so that correct judging result still can be obtained.
(4) the present invention is based on the average row height of literal line in text diagram (the i.e. subsequent parameter mean_text_ referred to
Lines_height), the length for determining the literal line handled, accuracy and the engineering of calculating time for having taken into account judgement are excellent
Change.On the one hand, longer literal line can more make final result tend to regular situation, thus right wherein the text contained is more
Probability processing result is advantageous, can reduce erroneous judgement.But too long of literal line, the time-consuming for handling needs is longer, to engineer application
It is unfavorable.Therefore, it is necessary to take into account the accuracy of processing result and handling duration.Method of the present invention using default text number, association
It adjusts both above-mentioned.Under the premise of handling duration is acceptable, longer literal line is taken as far as possible.And whether the literal line intercepted is sufficient
Enough long, this method is measurement of being come in using average row height, avoids being suitable for different text diagrams using absolute threshold.
Detailed description of the invention
Fig. 1 is the judgement and bearing calibration flow chart that text image of the present invention overturns;
Fig. 2 is the Chinese character literal line sample graph all over Britain to be processed that the present invention extracts in a positive figure;
Fig. 3 is that the present invention carries out the text sample graph obtained after OCR processing to the literal line of Fig. 2;
Fig. 4 is the literal line sample graph to be processed that the present invention extracts the reverse figure of Fig. 2;
Fig. 5 is that the present invention carries out the text sample graph obtained after OCR processing to the literal line of Fig. 4;
Fig. 6 is the full Chinese character literal line sample graph to be processed that the present invention extracts in a positive figure;
Fig. 7 is that the present invention carries out the text sample graph obtained after OCR processing to the literal line of Fig. 6;
Fig. 8 is the literal line sample graph to be processed that the present invention extracts the reverse figure of Fig. 6;
Fig. 9 is that the present invention carries out the text sample graph obtained after OCR processing to the literal line of Fig. 8;
Figure 10 is the Chinese mixed characters literal line sample graph of English to be processed that the present invention extracts in a positive figure;
Figure 11 is that the present invention carries out the text sample graph obtained after OCR processing to the literal line of Figure 10;
Figure 12 is the literal line sample graph to be processed that the present invention extracts the reverse figure of Figure 10;
Figure 13 is that the present invention carries out the text sample graph obtained after OCR processing to the literal line of Figure 12;
Figure 14 is the figure after one slant correction pretreatment of the present invention;
Figure 15 is after the present invention connects the left and right of Figure 14 progress binaryzation, small scale or the upper and lower foreground point distance of swimming
Figure;
Figure 16 is the present invention to Figure 15, counts the foreground point quantity in the every row of horizontal direction, obtained prospect projection value;
Figure 17 is the statistical probability of highest preceding 128 Chinese characters of probability of occurrence in the students' work exam pool of the invention counted;
Figure 18 is highest preceding 128 Chinese characters of probability of occurrence in the students' work exam pool of the invention counted.
Specific embodiment
Below in conjunction with Fig. 1 in the embodiment of the present invention, technical solution in the embodiment of the present invention carries out clear, complete
Ground description.Obviously, described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.Based on this
Embodiment in invention, every other reality obtained by those of ordinary skill in the art without making creative efforts
Example is applied, shall fall within the protection scope of the present invention.The embodiment of the present invention can according to be defined and covered by claim it is a variety of not
It is embodied with mode.
It should be noted that in the following description, understanding for convenience, giving many details.It is apparent that
Realization of the invention can be without these details.
It should be noted that in the absence of clear limitations or conflicts, each embodiment in the present invention and its
In technical characteristic can be combined with each other and form technical solution.
It should be noted that the present invention is discussed as an example with English and Chinese, but present invention could apply to it
In the word processing of country variants such as his phonetic alphabet text or pictograph.
Firstly, it is necessary to explanation, the text image that the present invention is handled is handled by slant correction;Or text
This image itself be it is proper, do not need carry out slant correction, directly handled using the present invention.
S1: row cutting is carried out to text image;
Carrying out row cutting to text image is the prior art, is realized using existing method, and present invention offer one is excellent
The technical solution of choosing:
Binaryzation is first carried out to text image, for convenience, if background pixel point value is 0, foreground pixel point is taken
Value is 1;The distance of swimming for carrying out foreground pixel point to binary image again connects, i.e., adjacent or distance is less than to the prospect of certain threshold value
Existing background pixel point is revised as foreground pixel point between pixel, forms binaryzation distance of swimming connection figure.As shown, Figure 14
It is connected by binary conversion treatment with the distance of swimming, forms binaryzation distance of swimming connection figure as shown in figure 15.
To binaryzation distance of swimming connection figure, the foreground pixel point quantity of every row in its horizontal direction is counted, it is as shown in figure 16, horizontal
The value of coordinate representation foreground pixel point quantity, ordinate indicate the line label of binaryzation distance of swimming connection figure.To any a line, if
The foreground pixel point quantity of the row is less than the projection threshold value of setting, then it is assumed that the row is blank pixel row, is otherwise non-blank-white picture
Plain row;Projection threshold value is set according to actual conditions.According to the above method, the corresponding text image point of binaryzation distance of swimming connection figure
It is segmented into blank pixel row part and non-blank-white pixel column part, the line number of adjacent non-blank-white pixel column is calculated, obtains non-blank-white
Capable height.At this point, a non-blank-white row may correspond to a character row, a formula row or a figure row.
S2: rejecting non-legible row, retains literal line;
It is high to obtain average literal line for the average value for calculating the height text_lines_height (i) of all non-blank-white rows
Mean_text_lines_height is spent, wherein i is the serial number of non-blank line.
It uses mean_text_lines_height as threshold value, checks the height text_lines_ of each non-blank-white row
height(i).There are three types of situations:
Situation 1: if text_lines_height (i) > > mean_text_lines_height, then it is assumed that the row right and wrong
Literal line, the row may be figure, table, formula etc.;
Situation 2: if text_lines_height (i) < < mean_text_lines_height, think that the row is at this time
Non-legible row, the row may be horizontal line etc.;
Situation 3: if being not belonging to situation 1 and situation 2, then it is assumed that the row is literal line.
Above-mentioned situation 3 is taken, i.e., only takes literal line, subsequent processes are carried out, without the use of the non-text such as figure, table, horizontal line
The part of word row.It should be noted that " being far longer than " operation and situation 2 that how estimate of situation 1 refers to refer to it is " much small
In " operation, it is common knowledge to one skilled in the art, can carries out according to the actual situation customized.
For example, working as text_lines_height (i) >=3*mean_text_lines_height, it can think text_
lines_height(i)>>mean_text_lines_height。
S3 takes any literal line L, so that the length of literal line L meets the range of setting;
A literal line is chosen, if the length for the literal line chosen is greater than the N1 of mean_text_lines_height
Times, and less than N2 times of mean_text_lines_height, then enabling the row is literal line L, carries out subsequent processing, natural
The value of number N1 and N2 rule of thumb obtains, and such as takes N1 ∈ [4,8], N2 ∈ [9,16].If the length for the literal line chosen is small
It in N1, is then supplemented, until meeting the requirements.If the length for the literal line chosen is greater than N2, intercepted, until meeting
It is required that.For example, attached drawing 14 of the invention, when handling literal line " realities of 10. ' following figure is to probe into plane mirror imaging characteristic ' ", this article
Word row is too long, intercepts as " following figure is to probe into plane mirror imaging spy " as shown in Figure 4.
The purpose of above-mentioned way is so that the text in literal line is neither very little, in order to avoid subsequent probabilistic contrastive test
As a result unreliable;Make the text in literal line not many again, in case engineering time-consuming is larger.Our experience, optimal is taken
The length of literal line is 10 times or so of literal line height (i.e. mean_text_lines_height), that is, takes ten Chinese characters, two
The width of ten English alphabets or so calculates the cost function value of positive literal line and reverse literal line at this time, has just had
Stronger otherness.And at this point, engineering time-consuming can generally also receive, it can't be too big.In addition, conventional it is considered that English
Literary word averagely contains 4 letters, considers further that there are blank spaces between English word, 20 English alphabets, it is generally recognized that can
To form 3,4 English words, there is preferable reliability and validity in contrastive test.
S4: OCR is carried out to literal line L, calculates the cost function value of literal line L;
OCR is carried out to literal line L.OCR such as carried out to attached drawing 6 of the invention, the result of OCR as shown in fig. 7, this figure knowledge
Not the result is that all correctly.In addition, the present invention does not use punctuation mark in subsequent processing, therefore ignore to punctuation mark
Recognition result.
The cost function value for calculating literal line L has many schemes optional.An embodiment provided by the invention is: foundation
The result of OCR identification.The cost function value for calculating each character or character combination (word), by all characters or character combination
Cost function value be added to obtain the cost function value of literal line L.Wherein the cost function value of character or character combination and they
The probability used is at positive relationship, and higher using probability, cost function value is bigger;The cost function value of character combination further may be used
With the length with character combination at positive relationship, length is longer, then its cost function value is bigger.For example, in Chinese, according to people
The cost function value of commonly used word is set as 5, be of little use word to the frequency of use statistical conditions of the commonly used word of related character library by people's daily paper
Cost function value be set as 2, the cost function value for the word being not present in character library is set as 0.In English, there is word in day in foundation
The probability often occurred in life sets cost function value, and word present in dictionary, probability of occurrence is high, is set as 5, is not present
Word is set as 0.And in English, and increase its cost function value according to word length, for example, word " absolutely " length
It is 10, its cost function value can be increased 10, and word " he " length is only 2, increased cost function value is 2.According to upper
Content is stated it is found that under normal circumstances, the cost function value of positive literal line can all be greater than the cost function of reverse literal line
Value.
Specifically, the present invention provides a kind of method of preferred cost function value for calculating literal line L, as described below.
Each character identified to OCR following judge: judge that character is suitble to handle by granularity of word
Or using character as granularity processing.Such as English character is suitble to word (rather than alphabetical) for granularity processing, middle text
Symbol is suitble to Chinese character (rather than phrase) for granularity processing.For above-mentioned judging result, literal line L is divided into and is suitble to character
The part handled for granularity or the part handled by granularity of word.Literal line L can specifically be divided into several parts, by
The content of literal line L determines.
The method for carrying out granularity judgement is distinguished and is known preferably by the code word region of the coding codeword distribution of OCR output
Other character, in this way English or Chinese, then the granularity of determining character processing.This is because the constraint relationship of English word is very
By force.Obtain correct English word in normal forward direction text image, and in the reverse figure of mistake, usually will recognise that it is baffled its
Wonderful English alphabet sum number combinatorics on words.For example, Fig. 3 is the recognition result for carrying out OCR to Fig. 2, correct English word is obtained;
Fig. 4 is the text image obtained after overturning to Fig. 2, and Fig. 5 is the recognition result that OCR is carried out to Fig. 4, is identified entirely disorderly
Number and English alphabetic combination.Therefore, it finds, judge whether there is correct English word in the result of OCR, can strongly support
The differentiation of forward direction figure or reverse figure.The processing of word granularity can be used in English text.
And to Chinese, it is substantially based on Chines words processing, it is not necessary to based on Chinese phrase processing.This is because mistake
The Chinese character that OCR comes out in reverse figure is as a result, very maximum probability will appear the Chinese character that is of little use.For example, Fig. 7 is the knowledge for carrying out OCR to Fig. 6
Not as a result, the result identified belongs to more common Chinese character;Fig. 8 is the text image obtained after overturning to Fig. 6, figure
9 be the recognition result that OCR is carried out to Fig. 8, includes multiple Chinese characters being of little use in recognition result.Therefore, Chinese is based at character
Reason effect, that is, guaranteed.Certainly, in Chinese situation, if being based further on Chinese word processing, the accuracy of differentiation can be into one
Step is promoted.
The cost function value that literal line L each section is calculated based on above-mentioned judgement, by the cost function value of each section
Addition obtains the cost function value of literal line L.
To literal line L any portion, if being granularity processing with word (rather than alphabetical), whether several characters are checked
It is combined into legitimate words or fragments of words;The cost function value for calculating each legitimate words or fragments of words, by all generations
Valence functional value is added to obtain the cost function value of the part.
The method that string matching can be used finds legitimate words or fragments of words.Common word can be exported from dictionary,
Form common word library.By taking English as an example, according to the practical engineering experience of inventor, common word library is single using 10,000 English
Word, the success rate of differentiation is very high, and common word can search out.It, can if OCR is higher to the recognition correct rate of letter
Using the accurate matched hash value method of character string, to save matched calculation amount.Particularly, legitimate words, which refer to, combines into syllables correctly
Word, if " other " in English is legitimate words;Fragments of words refers to a part in legitimate words, and such as " er " is
A part of the legitimate words such as " other ", " teacher ".
Calculate the cost function value of legitimate words or fragments of words, can be used word probability of occurrence sub- cost function value and
The sum of sub- cost function value of word length.Word probability of occurrence is higher, then the sub- cost function value of probability of occurrence is bigger;Word
Length is longer, then the sub- cost function value of length is bigger.The sub- cost function value of word probability of occurrence is divided into 5 grades by the present invention, point
It is not assigned a value of 1 to 5 point, the high sub- cost function value value of probability of occurrence is bigger.Word length contribution, then directly made with its length value
For the sub- cost function value of length.
To literal line L any portion, if checking each character that OCR is identified using character as granularity processing,
All cost function values are added to obtain the cost function value of the part by the cost function value for calculating each character.Character
The probability occurred in routine use is bigger, and cost function value is bigger.
By taking Chinese as an example, inventor has counted pupil, the Chinese characters in common use in middle school student's operation exam pool, altogether 8149 Chinese
Word, wherein the statistical probability of maximum preceding 128 Chinese characters of probability of occurrence, as shown in figure 17, abscissa indicate Chinese character serial number, often
The corresponding Chinese character of a serial number, indicates 128 Chinese characters altogether;Ordinate indicates the probability that Chinese character occurs.128 Chinese characters are listed in Figure 18
In.According to the regularity of distribution of the probability of occurrence of the 8149 of statistics Chinese characters, Chinese character can be divided into first 10, it is 50 first, preceding 100
A, preceding 300 and 1000 five grades first, cost function value difference assignment 5,4,3,2,1.Probability of occurrence is except first 1000
Chinese character, be assigned a value of 0.By it, the cost function value of the Chinese character of available OCR result, and the Chinese character that probability of occurrence is higher,
Cost function value value is bigger.
Obviously, a possibility that cost function value of literal line L is bigger, and text image is forward direction is higher.
S5: reverse literal line L obtains literal line L1。
Reverse literal line is i.e. to the corresponding image rotation 180 degree of literal line.Literal line as shown in Figure 10 is obtained by overturning
To literal line shown in Figure 12.
S6: to literal line L1OCR is carried out, literal line L is calculated1Cost function value.
This step is no longer described in detail referring to S4.According to the content of S4 it is found that usually literal line L1Cost function value very
It is small.
S7: according to the cost function value of literal line L and literal line L1Cost function value, judge whether literal line L overturns;
If literal line L is reverse, reverse correction is carried out.
If the cost function value of literal line L is greater than literal line L1Cost function value, then literal line L is positive, no
It needs to overturn;Otherwise it is assumed that literal line L be it is reverse, then carry out reverse correction.
Claims (7)
1. a kind of reverse judgement of text image based on content and bearing calibration, special if text image has been subjected to slant correction
Sign is, comprising the following steps:
S1: row cutting is carried out to text image;
S2: rejecting non-legible row, retains literal line;
S3: taking any literal line L, so that the length of literal line L meets the range of setting;
S4: OCR is carried out to literal line L, calculates the cost function value of literal line L, OCR refers to optical character identification;
S5: reverse literal line L obtains literal line L1;
S6: to literal line L1OCR is carried out, literal line L is calculated1Cost function value;
S7: according to the cost function value of literal line L and literal line L1Cost function value, judge whether literal line L overturns;If
Literal line L is reverse, then carries out reverse correction;
The cost function value of literal line described above, has the characteristic that: for same literal line, positive literal line cost letter
Numerical value is markedly different from reverse literal line cost function value.
2. the reverse judgement of the text image according to claim 1 based on content and bearing calibration, which is characterized in that text
This image carries out row cutting, using following step:
Binaryzation is first carried out to text image;The distance of swimming for carrying out foreground pixel point to binary image again connects, and forms binaryzation
Distance of swimming connection figure;
To binaryzation distance of swimming connection figure, the foreground pixel point quantity of every row in its horizontal direction is counted;To any a line, if should
Capable foreground pixel point quantity is less than the projection threshold value of setting, then it is assumed that the row is blank pixel row, is otherwise non-blank pixel
Row;Projection threshold value is set according to actual conditions;
The line number for calculating adjacent non-blank-white pixel column, obtains the height of non-blank-white row.
3. the reverse judgement of the text image according to claim 2 based on content and bearing calibration, which is characterized in that reject
Non-legible row, the process for retaining literal line are as follows:
The average value for calculating the height text_lines_height (i) of all non-blank-white rows obtains average literal line height
Mean_text_lines_height, wherein i is the serial number of non-blank line;
It uses mean_text_lines_height as threshold value, checks the height text_lines_height of each non-blank-white row
(i), it is handled according still further to following three kinds of situations:
Situation 1: if text_lines_height (i) > > mean_text_lines_height, then it is assumed that the row is non-legible
Row;
Situation 2: if text_lines_height (i) < < mean_text_lines_height, think that the row is non-text at this time
Word row;
Situation 3: if being not belonging to situation 1 and situation 2, then it is assumed that the row is literal line.
4. the reverse judgement of the text image according to claim 3 based on content and bearing calibration, which is characterized in that choose
When any literal line L, if the length for the literal line chosen is greater than N1 times of mean_text_lines_height, and it is less than
N2 times of mean_text_lines_height, then enabling the row is literal line L, and the value of natural number N1 and N2 rule of thumb obtain
?.
5. the reverse judgement of the text image according to claim 4 based on content and bearing calibration, which is characterized in that calculate
The cost function value of literal line L, using following step:
Each character identified to OCR following judge: judge that character is suitble to handle still by granularity of word
Using character as granularity processing;For above-mentioned judging result, literal line L is divided into the portion for being suitble to handle as granularity using character
The part divided or handled using word as granularity;
Based on above-mentioned judgement, the cost function value of literal line L each section is calculated, the cost function value of each section is added
Obtain the cost function value of literal line L;
Wherein, to literal line L any portion, if checking whether several characters are combined into using word as granularity processing
Legitimate words or fragments of words;The cost function value for calculating each legitimate words or fragments of words, by all legitimate words
Or the cost function value of fragments of words is added, and obtains the cost function value of the part;Calculate the generation of legitimate words or fragments of words
Valence functional value uses the sum of the sub- cost function value of word probability of occurrence and the sub- cost function value of word length;Word probability of occurrence
Higher, then the sub- cost function value of probability of occurrence is bigger;Word length is longer, then the sub- cost function value of length is bigger;
Again: to literal line L any portion, if checking each character that OCR is identified using character as granularity processing,
The cost function value for calculating each character is added all character cost function values to obtain the cost function value of the part;
The probability that character occurs in routine use is bigger, and cost function value is bigger.
6. the reverse judgement of the text image according to claim 5 based on content and bearing calibration, which is characterized in that use
The method of string matching finds legitimate words or fragments of words.
7. the reverse judgement of the text image according to claim 5 based on content and bearing calibration, which is characterized in that carry out
Granularity judgement method, preferably by OCR output coding codeword distribution code word region, the character of Division identification, then
Determine the granularity of character processing.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811192521.6A CN109409363B (en) | 2018-10-13 | 2018-10-13 | Content-based text image inversion judgment and correction method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811192521.6A CN109409363B (en) | 2018-10-13 | 2018-10-13 | Content-based text image inversion judgment and correction method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109409363A true CN109409363A (en) | 2019-03-01 |
CN109409363B CN109409363B (en) | 2021-11-12 |
Family
ID=65467144
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811192521.6A Active CN109409363B (en) | 2018-10-13 | 2018-10-13 | Content-based text image inversion judgment and correction method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109409363B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109919155A (en) * | 2019-03-13 | 2019-06-21 | 厦门商集网络科技有限责任公司 | A kind of the inclination angle antidote and terminal of text image |
CN110516667A (en) * | 2019-08-15 | 2019-11-29 | 浙江万朋教育科技股份有限公司 | A kind of text rectification method based on Android |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20140268250A1 (en) * | 2013-03-15 | 2014-09-18 | Mitek Systems, Inc. | Systems and methods for receipt-based mobile image capture |
CN106408038A (en) * | 2016-09-09 | 2017-02-15 | 华南理工大学 | Rotary Chinese character identifying method based on convolution neural network model |
CN107609549A (en) * | 2017-09-20 | 2018-01-19 | 北京工业大学 | The Method for text detection of certificate image under a kind of natural scene |
CN108427950A (en) * | 2018-02-01 | 2018-08-21 | 北京捷通华声科技股份有限公司 | A kind of literal line detection method and device |
-
2018
- 2018-10-13 CN CN201811192521.6A patent/CN109409363B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20140268250A1 (en) * | 2013-03-15 | 2014-09-18 | Mitek Systems, Inc. | Systems and methods for receipt-based mobile image capture |
CN106408038A (en) * | 2016-09-09 | 2017-02-15 | 华南理工大学 | Rotary Chinese character identifying method based on convolution neural network model |
CN107609549A (en) * | 2017-09-20 | 2018-01-19 | 北京工业大学 | The Method for text detection of certificate image under a kind of natural scene |
CN108427950A (en) * | 2018-02-01 | 2018-08-21 | 北京捷通华声科技股份有限公司 | A kind of literal line detection method and device |
Non-Patent Citations (2)
Title |
---|
MARIOS ANTHIMOPOULOS等: "A Pixel-Based Evaluation Method for Text Detection in Color Images", 《IEEE》 * |
王涛等: "基于语义分割技术的任意方向文字识别", 《应用科技》 * |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109919155A (en) * | 2019-03-13 | 2019-06-21 | 厦门商集网络科技有限责任公司 | A kind of the inclination angle antidote and terminal of text image |
CN109919155B (en) * | 2019-03-13 | 2021-03-12 | 厦门商集网络科技有限责任公司 | Inclination angle correction method for text image and terminal |
CN110516667A (en) * | 2019-08-15 | 2019-11-29 | 浙江万朋教育科技股份有限公司 | A kind of text rectification method based on Android |
Also Published As
Publication number | Publication date |
---|---|
CN109409363B (en) | 2021-11-12 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US5664027A (en) | Methods and apparatus for inferring orientation of lines of text | |
Hochberg et al. | Automatic script identification from document images using cluster-based templates | |
US9069753B2 (en) | Determining proximity measurements indicating respective intended inputs | |
US9087047B2 (en) | Text proofreading apparatus and text proofreading method using post-proofreading sentence with highest degree of similarity | |
US6154579A (en) | Confusion matrix based method and system for correcting misrecognized words appearing in documents generated by an optical character recognition technique | |
US8340425B2 (en) | Optical character recognition with two-pass zoning | |
US8725497B2 (en) | System and method for detecting and correcting mismatched Chinese character | |
CN105760359B (en) | Question processing system and method thereof | |
US20190362713A1 (en) | Dynamic extraction of contextually-coherent text blocks | |
WO2024045527A1 (en) | Word/sentence error correction method and device, readable storage medium, and computer program product | |
CN109409363A (en) | The reverse judgement of text image based on content and bearing calibration | |
CN112149680A (en) | Wrong word detection and identification method and device, electronic equipment and storage medium | |
Chaudhuri | Reversed word dictionary and phonetically similar word grouping based spell-checker to Bangla text | |
CN110765107B (en) | Question type identification method and system based on digital coding | |
JPS63146187A (en) | Character recognizing device | |
CN111709228A (en) | Automatic recognition method for repeated errors of words | |
JP4047895B2 (en) | Document proofing apparatus and program storage medium | |
Chaudhuri | Towards Indian language spell-checker design | |
CN113033188B (en) | Tibetan grammar error correction method based on neural network | |
Rathnasena et al. | Summarization based approach for old sinhala text archival search and preservation | |
JP4047894B2 (en) | Document proofing apparatus and program storage medium | |
Guo et al. | SVM Method and Integrating Pinyin and Dictionary Method for Chinese Spelling Errors Detection | |
JP4318223B2 (en) | Document proofing apparatus and program storage medium | |
WO2023026166A1 (en) | System and method for meta-data extraction from documents | |
Meknavin et al. | Progress of combining trigram and winnow in Thai OCR error correction |
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 |