CN1278271C - Identifying method for fragmentary printed digits - Google Patents
Identifying method for fragmentary printed digits Download PDFInfo
- Publication number
- CN1278271C CN1278271C CN 200410025042 CN200410025042A CN1278271C CN 1278271 C CN1278271 C CN 1278271C CN 200410025042 CN200410025042 CN 200410025042 CN 200410025042 A CN200410025042 A CN 200410025042A CN 1278271 C CN1278271 C CN 1278271C
- Authority
- CN
- China
- Prior art keywords
- character
- profile
- ratio
- model
- max
- 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.)
- Expired - Fee Related
Links
Images
Landscapes
- Character Discrimination (AREA)
Abstract
The present invention relates to an identifying method for fragmentary printed numeric characters, which is used for the technical field of image identification. The present invention has the following steps: firstly, an integral character contour is resolved into an upper, a lower, a left and a right local contours which are aggregately denoted by using the position coordinates of contour pixel points from a framing mask to the character contour; secondly, one time of discrete differentiation is carried out respectively for the position coordinates of the contour pixel points of the upper, the left and the right contours, and element collecting structural characteristics are extracted form residual upper, left and right contours according to the tendency variation of a contour curve of the analysis of the discrete differentiation. In addition, the left and the right contours are used for accounting the width of characters, and the upper and the lower contours are used for accounting the height of the characters; the height ratio of the characters is accounted and the number of strokes is accounted within the range of 0.5H; the contour structural characteristics of the characters are combined with statistical characteristics to establish models for numeric characters, and then a structured statement identifying method is adopted to incomplete the numeric characters. The method can accurately identify bottom-incomplete and complete numeric characters and enhances the accuracy of identifying the numeric characters in practical application.
Description
Technical field
What the present invention relates to is a kind of character recognition method, particularly a kind of recognition methods of incomplete printing digital character.Be used for field of image recognition.
Background technology
Character recognition technology extensively applies to each engineering field in recent years, and the research of literal identification has already obtained huge achievement, and the literal recognition correct rate of block letter is up to 99%.But, existing the character of part incompleteness in the practical application, incomplete character and complete character mix, and identification has caused difficulty to literal.
Find by literature search, people such as Pasquale Foggia are at " Image and Vision Computing " (1999,17 (9), 701-711.) " Combining statisticaland structural approaches for handwritten character description " (" handwritten character description that statistics combines with structural approach ") of delivering on (" image and vision computer "), the method that is proposed in this article, from handwritten character, extract earlier structural motif, with the method for statistics structural motif is carried out feature description then, adopt neural network to carry out handwritten form identification at last.
The technology that this article relates to mainly exists following defective and deficiency: (1) algorithm is only applicable to the literal identification of handwritten form, though can effectively overcome the character distortion of handwritten form, powerless for the distortion of incomplete character; (2) stroke of incomplete character is lost and can be reduced the original structural motif of character; (3) stroke of incomplete character is lost and also can be changed character structural motif originally simultaneously, makes to distort when with statistical method primitive feature being described; (4) recognition methods of neural network must be determined the dimension of input feature value in advance, but the dimension of the different meeting of the incomplete degree effect characteristics vector of incomplete character.
Summary of the invention
The objective of the invention is to overcome the deficiency in the existing character recognition technology, a kind of recognition methods of incomplete printing digital character is provided, make it carry out accurate recognition the printing digital character of incompleteness.
The present invention is achieved by the following technical solutions, and the inventive method is as follows: earlier whole character outline is decomposed into upper and lower, four local configurations in a left side and the right side, and represents profile with framing mask to the position coordinates set of the wire-frame image vegetarian refreshments of character outline.Position coordinates to last, left and right profile wire-frame image vegetarian refreshments carries out discrete differential one time respectively, trend according to a discrete differential analysis contour curve changes, define 5 structure element of sets, comprise vertically tiltedly (L), right tiltedly (R), circular arc (C) and sudden change (P) of (V), a left side, and from remainingly go up, extraction element of set architectural feature a left side and the right profile.In addition, utilize the width (W) of left and right sides profile statistics character, bottom profiled statistics character height (H) in the utilization, statistics stroke number in the 0.5H scope at last.The contour structure feature and the statistical nature of character combine, and for after numerical character sets up model, adopt the incomplete numerical character of structured statement recognition methods.
Though the numerical character of bottom incompleteness makes character lose many features owing to lost the important stroke of bottom, also make Partial Feature become unstable simultaneously, identification has caused very big difficulty to literal.But when the broken partial section of character was no more than the 0.5H of original characters, human eye still can identify exactly.The incomplete character of this explanation still remains with enough remaining features.Extract these residual stable features, adopt rational recognition strategy can realize the identification of the incomplete numerical character in bottom.
Below the inventive method is further described, method step is as follows:
(1) character outline is decomposed
The contour feature that the overall profile of character is decomposed into top, bottom, left side and right side four direction is described.When making bottom profile damaged, be unlikely to have influence on the top feature, and can from the contour feature of the left and right sides, extract the valuable information of part.
The left side profile (LP (k), k=1,2 ... M) be defined as the horizontal direction coordinate figure of character leftmost side boundary pixel point.
LP(i)＝min{x|P(x，y)∈C，y＝i} i＝1，2…M (1)
(x, y) coordinate is that (x, pixel y), C are represented the set of character pixels point to P in the formula in the presentation video.In like manner, right lateral contours (RP (k), k=1,2 ... M) be defined as the horizontal direction coordinate figure of character rightmost side boundary pixel point.
RP(i)＝max{x|P(x，y)∈C，y＝i} i＝1，2…M (2)
Correspondingly, the top profile (TP (k), k=1,2 ... N) be defined as the vertical direction coordinate figure of the highest border of character pixel.Bottom profile (BP (k), k=1,2 ... N) be defined as the vertical direction coordinate figure of the minimum boundary pixel point of character.
TP(j)＝min{y|P(x，y)∈C，x＝j} j＝1，2…N (3)
BP(j)＝max{y|P(x，y)∈C，x＝j} j＝1，2…N (4)
(2) profile single order discrete differential
In order to describe the profile varying feature, the single order differential of definition four direction profile:
LPD＝LP(i+1)-LP(i)
RPD＝RP(i+1)-RP(i) (5)
TPD＝TP(j+1)-TP(j)
BPD＝BP(j+1)-BP(j)
I=1 in the formula, 2 ... M-1, j=1,2 ... N-1.
(3) the structure element of set feature extraction on each profile
Constitute the basic primitive of character outline according to the variation tendency definition of character outline.Basic primitive has 5 and is respectively: vertically tiltedly (L), right tiltedly (R), circular arc (C) and sudden change (P) of (V), a left side.Define above-mentioned basic primitive:
(a) vertical
Definition: suppose SL, SV and SR represent that respectively certain side profile single order differential value greater than zero, equals zero and minus number, if SR=0, SL=0 then is structure V.
(b) left side tiltedly
Definition: suppose SL, SV and SR represent that respectively certain side profile single order differential value greater than zero, equals zero and minus number, if SR=0, the big threshold value LT of SL then is structure L.
(c) tiltedly right
Definition: suppose SL, SV and SR represent that respectively certain side profile single order differential value greater than zero, equals zero and minus number, if SL=0, the big threshold value RT of SR then is structure R.
(d) circular arc
Definition: suppose SL, SV and SR represent that respectively certain side profile single order differential value greater than zero, equals zero and minus number, and greater than threshold value RT, the big threshold value LT of SL then is a structure C as if SR.
(e) sudden change
Continuous character outline, the variable quantity of its single order differential value is smaller, and when character outline was undergone mutation, its single order differential value was relatively large.Therefore, definition: then character outline has sudden change when the single order differential value of profile surpasses threshold value PT, is structure P.Structure P is divided into several continuous curves with character outline, and each bar curve extracts architectural feature independently of one another.
According to above-mentioned definition, consider the interference pixel that exists on the character outline, adopt threshold technology to detect element of set:
Suppose that PD (k) represents the single order differential of certain side profile, k=1,2 ... K, SL, SV and SR are respectively detected PD (k) greater than zero, equal zero and minus number, and PT, RT and LT are positive integer, then
If | PD (k) |＞=PT then detects structural mutation (P) at the k place; The effective range that detects mutation structure P is at x ∈ [ST, N-ST+1], y ∈ [ST, M-ST+1], and wherein ST represents the width of stroke.This mainly is under the serious interference situation, when the contour edge smooth treatment is not ideal enough, and may detected pseudomutation primitive.
If SL＜LT, SR＜RT then detects structure and is vertical (V);
If SL＞LT, SR＜RT then detects structure and is a left side oblique (L);
If SL＜LT, SR＞RT then detects structure and is right tiltedly (R);
If SL＞LT, SR＞RT, then detecting structure is circular arc (C).
Detected element of set is kept among separately the Vector Groups LS and RS in accordance with the order from top to bottom on the profile of the left and right sides; The top detects element of set and is kept among another Vector Groups TS by from left to right order.On the profile of top, i structure element of set of the Vector Groups of TS (i) expression top contour structure element of set, Tn represents the element of set number that the top profile is total; On the profile of left side, i structure element of set of the Vector Groups of LS (i) expression left side contour structure element of set, Ln represents the total element of set number of left side profile; On the right lateral contours, i structure element of set of the Vector Groups of RS (i) expression right lateral contours structure element of set, Rn represents the element of set number that right lateral contours is total.
(4) statistical nature of profile
Adopt above-mentioned structural motif also to be not enough to accurately identification incompleteness and complete numeral, introducing has the strong mutual not profile statistical nature of property with architectural feature.
(a) character height and maximum character duration W
MaxThe ratio
The breadth extreme of character is
The height of single character is
In the practical application, incomplete character and complete character mix, and quantity is less relatively.Therefore, though the bottom incompleteness causes the height of incomplete character accurately to estimate, in same character area, character boundary fixing, it highly near equating, can adopt the medium filtering of single character height to estimate character height,
H＝med{h
1，h
2，…h
m} (8)
M is the character sum in the character area.
The depth-width ratio of character is
Ratio＝H/W
max (9)
This feature is mainly used in discriminating digit 1.When Ratio 〉=2.5, be numeral 1.
(b) stroke number of vertical direction
In the scope of character 0.5H, scan the stroke number of every row pixel from top to bottom, get stroke number purpose maximal value S
MaxThis feature is mainly used in difference numeral 0 and 8, works as S
Max, be character 8 at 2 〉=2 o'clock; Otherwise be character 0.
(5) set up the model of 10 numerical characters
The model of 10 numerical characters is as follows:
The model of " 0 " character: Ratio 〉=2.5, TS (1)=C, Size (LS)=Size (RS)=1, S
Max＜2
The model of " 1 " character: Ratio＜2.5
The model of " 2 " character: Ratio 〉=2.5, TS (1)=C, LS (1) ≠ C, LS (Ln-1)=P, LS (Ln)=L
The model of " 3 " character: Ratio 〉=2.5, TS (1)=C, LS (1) ≠ C, P ∈ LS, LS (Ln) ≠ L; Or Ratio 〉=2.5, TS (1)=V, RS (1)=C
The model of " 4 " character: Ratio 〉=2.5, TS (1)=L, RS (1)=V
The model of " 5 " character: Ratio 〉=2.5, V ∈ TS, P ∈ RS;
The model of " 6 " character: Ratio 〉=2.5, TS (1)=C, P ∈ RS, Size (LS)=1; Or Ratio 〉=2.5, TS (1)=L, V RS
The model of " 7 " character: Ratio 〉=2.5, TS (1)=V, P ∈ LS, Size (RS)=1;
The model of " 8 " character: Ratio 〉=2.5, TS (1)=C, Size (LS)=Size (RS)=1, S
Max〉=2
The model of " 9 " character: Ratio 〉=2.5, TS (1)=C, LS (1)=C, LS (2)=P
(6) adopt the structured statement recognition methods to discern incomplete numerical character
Adopt the structured statement recognition methods,,,, realize the literal identification of incomplete numerical character with Model Matching according to defined numerical character model with architectural feature and the statistical nature that extracts on the target text.
The present invention has overcome owing to numerical character bottom stroke is lost the problem that causes character to discern, and also can realize accurate identification to complete numerical character.This literal image partitioning method has following advantage: (1) extracts the remaining feature of structural motif and profile statistics from the profile of top, the right and left from the numerical character of bottom incompleteness.(2), set up the structural model of numerical character according to above-mentioned feature.(3) can accurately discern the incomplete numerical character in bottom.(4) simultaneously, also accurate identification be can realize, reliability and recognition accuracy that recognizer is lost, is out of shape character stroke improved for complete numerical character.
Description of drawings
Fig. 1 outline definition synoptic diagram
Five structure element of sets of Fig. 2 synoptic diagram
Embodiment
As shown in Figure 1, be outline definition synoptic diagram of the present invention, the contour feature that wherein overall profile of character is decomposed into top, bottom, left side and right side four direction is described.When making bottom profile damaged, be unlikely to have influence on the top feature, and can from the contour feature of the left and right sides, extract the valuable information of part.
The left side profile (LP (k), k=1,2 ... M) be defined as the horizontal direction coordinate figure of character leftmost side boundary pixel point.
LP(i)＝min{x|P(x，y)∈C，y＝i} i＝1，2…M
(x, y) coordinate is that (x, pixel y), C are represented the set of character pixels point to P in the formula in the presentation video.In like manner, right lateral contours (RP (k), k=1,2 ... M) be defined as the horizontal direction coordinate figure of character rightmost side boundary pixel point.
RP(i)＝max{x|P(x，y)∈C，y＝i} i＝1，2…M
Correspondingly, the top profile (TP (k), k=1,2 ... N) be defined as the vertical direction coordinate figure of the highest border of character pixel.Bottom profile (BP (k), k=1,2 ... N) be defined as the vertical direction coordinate figure of the minimum boundary pixel point of character.
TP(j)＝min{y|P(x，y)∈C，x＝j} j＝1，2…N
BP(j)＝max{y|P(x，y)∈C，x＝j} j＝1，2…N
In the call number literal identification of library collection,, be subjected to the influence of spine space constraint because call number is attached on the spine; when call number is made up of two row or the character string more than two row; the character regular meeting of second row is folding, and after the camera shooting obtained image, the bottom of character just can be lost.The existence of this character has seriously reduced the accuracy of call number identification.
Content in conjunction with the inventive method provides following examples, and is specific as follows:
(1) contour feature that the overall profile of character is decomposed into top, bottom, left side and right side four direction is described, and explains with the position coordinates of wire-frame image vegetarian refreshments.
(2) each profile being carried out the single order discrete differential calculates.
(3) analyze the contour curve variation tendency according to the single order discrete differential, extract the structure element of set of each profile, and set up corresponding primitive Vector Groups.Parameter PT=6 when primitive extracts, LT=3, RT=3, ST=3.
(4) statistical nature of extraction profile comprises the depth-width ratio of character and the maximum stroke number on the column direction in the 0.5H scope.
(5), set up the model of 10 numerical characters according to above-mentioned character outline architectural feature and statistical nature.
(6) adopt the incomplete numerical character of structured statement recognition methods identification.
Adopt said method that the incomplete numerical character in the call number is discerned, its accuracy is 91.8%, and the recognition correct rate of complete character is 97.6%, thereby has guaranteed that call number identification has higher accuracy.
Claims (2)
1, a kind of recognition methods of incomplete printing digital character, it is characterized in that, whole character outline is decomposed into, down, a left side and right four local configurations, and represent profile to the set of the position coordinates of the wire-frame image vegetarian refreshments of character outline with framing mask, respectively to last, the position coordinates of the wire-frame image vegetarian refreshments on a left side and the right side carries out the single order discrete differential, trend according to a discrete differential analysis contour curve changes, define 5 structure element of sets, comprise vertical V, a left side is L tiltedly, right tiltedly R, circular arc C and sudden change P, and on remaining, extract the element of set architectural feature in the left and right profile, in addition, utilize the width W of left and right sides profile statistics character, bottom profiled statistics character height H in the utilization, in the 0.5H scope, add up the stroke number at last, the contour structure feature and the statistical nature of character combine, and for after numerical character sets up model, adopt the incomplete numerical character of structured statement recognition methods.Comprise following steps:
(1) character outline is decomposed, and the left side profile is defined as the horizontal direction coordinate figure of character leftmost side boundary pixel point,
LP(i)＝min{x|P(x，y)∈C，y＝i}i＝1，2…M
Right lateral contours is defined as the horizontal direction coordinate figure of character rightmost side boundary pixel point,
RP(i)＝max{x|P(x，y)∈C，y＝i}i＝1，2…M
Correspondingly, the top profile is defined as the vertical direction coordinate figure of the highest border of character pixel,
TP(j)＝min{y|P(x，y)∈C，x＝j}j＝1，2…N
Bottom profile is defined as the vertical direction coordinate figure of the minimum boundary pixel point of character,
BP(j)＝max{y|P(x，y)∈C，x＝j}j＝1，2…N；
(2) profile single order discrete differential, the single order differential LPD of definition left profile, the single order differential RPD of right profile, the single order differential TPD of top profile, the single order differential BPD of bottom profile:
LPD＝LP(i+1)-LP(i)
RPD＝RP(i+1)-RP(i)
TPD＝TP(j+1)-TP(j)
BPD＝BP(j+1)-BP(j)
I=1 in the formula, 2 ... M-1, j=1,2 ... N-1;
(3) the structure element of set feature extraction on each profile supposes that PD (k) represents the single order differential of certain side profile, k=1, and 2 ... K, SL, SV and SR are respectively detected PD (k) greater than zero, equal zero and minus number, and PT, RT and LT are positive integer, then
If | PD (k) |＞=PT then detects structural mutation P at the k place; The effective range that detects mutation structure P is at x ∈ [ST, N-ST+1], y ∈ [ST, M-ST+1], and wherein ST represents the width of stroke.
If SL＜LT, SR＜RT, then detecting structure is vertical V;
If SL＞LT, SR＜RT then detects structure and is the oblique L in a left side;
If SL＜LT, SR＞RT then detects structure and is right tiltedly R;
If SL＞LT, SR＞RT, then detecting structure is circular arc C;
Detected element of set is kept among separately the Vector Groups LS and RS in accordance with the order from top to bottom on the profile of the left and right sides; The top detects element of set and is kept among another Vector Groups TS by from left to right order, on the profile of top, and i structure element of set of the Vector Groups of TS (i) expression top contour structure element of set, Tn represents the element of set number that the top profile has; On the profile of left side, i structure element of set of the Vector Groups of LS (i) expression left side contour structure element of set, Ln represents the total element of set number of left side profile; On the right lateral contours, i structure element of set of the Vector Groups of RS (i) expression right lateral contours structure element of set, Rn represents the element of set number that right lateral contours is total;
(4) statistical nature of profile extracts, a. character height and maximum character duration W
MaxThe ratio
The breadth extreme of character is:
The height of single character is:
In the practical application, in same character area, character boundary is fixed, and it highly near equating, adopts the medium filtering of single character height to estimate character height,
H＝med{h
1，h
2，…h
m}
M is the character sum in the character area;
The depth-width ratio of character is:
Ratio＝H/W
max
This feature is mainly used in discriminating digit 1, when Ratio 〉=2.5, is numeral 1;
B. the stroke number of vertical direction
In the scope of character 0.5H, scan the stroke number of every row pixel from top to bottom, get stroke number purpose maximal value S
Max, this feature is mainly used in difference numeral 0 and 8, works as S
Max2 〉=2 o'clock, be character 8, otherwise be character 0;
(5) set up the model of numerical character,
The model of " 0 " character: Ratio 〉=2.5, TS (1)=C, Size (LS)=Size (RS)=1, S
Max＜2
The model of " 1 " character: Ratio＜2.5
The model of " 2 " character: Ratio 〉=2.5, TS (1)=C, LS (1) ≠ C, LS (Ln-1)=P, LS (Ln)=L
The model of " 3 " character: Ratio 〉=2.5, TS (1)=C, LS (1) ≠ C, P ∈ LS, LS (Ln) ≠ L; Or Ratio 〉=2.5, TS (1)=V, RS (1)=C
The model of " 4 " character: Ratio 〉=2.5, TS (1)=L, RS (1)=V
The model of " 5 " character: Ratio 〉=2.5, V ∈ TS, P ∈ RS;
The model of " 6 " character: Ratio 〉=2.5, TS (1)=C, P ∈ RS, Size (LS)=1; Or Ratio 〉=2.5, TS (1)=L,
The model of " 7 " character: Ratio 〉=2.5, TS (1)=V, P ∈ LS, Size (RS)=1;
The model of " 8 " character: Ratio 〉=2.5, TS (1)=C, Size (LS)=Size (RS)=1, S
Max〉=2
The model of " 9 " character: Ratio 〉=2.5, TS (1)=C, LS (1)=C, LS (2)=P
(6) adopt structured statement recognition methods identification literal.
2, the recognition methods of incomplete printing digital character according to claim 1 is characterized in that, described employing structured statement recognition methods identification literal is specially:
Adopt the structured statement recognition methods,,, realize literal identification with Model Matching with architectural feature and the statistical nature that extracts on the target text.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 200410025042 CN1278271C (en) | 2004-06-10 | 2004-06-10 | Identifying method for fragmentary printed digits |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 200410025042 CN1278271C (en) | 2004-06-10 | 2004-06-10 | Identifying method for fragmentary printed digits |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN1584921A CN1584921A (en) | 2005-02-23 |
| CN1278271C true CN1278271C (en) | 2006-10-04 |
Family
ID=34601115
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN 200410025042 Expired - Fee Related CN1278271C (en) | 2004-06-10 | 2004-06-10 | Identifying method for fragmentary printed digits |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN1278271C (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2006252400A (en) * | 2005-03-14 | 2006-09-21 | Keyence Corp | Image processor and method of generating registration data in image processing |
| JP4738469B2 (en) * | 2008-10-29 | 2011-08-03 | 株式会社東芝 | Image processing apparatus, image processing program, and image processing method |
| CN101916378B (en) * | 2010-07-20 | 2013-01-09 | 青岛海信网络科技股份有限公司 | Method and device for recognizing confusable character |
| CN103996057B (en) * | 2014-06-12 | 2017-09-12 | 武汉科技大学 | Real-time Handwritten Numeral Recognition Method based on multi-feature fusion |
| CN106709484B (en) * | 2015-11-13 | 2022-02-22 | 国网吉林省电力有限公司检修公司 | Digital identification method of digital instrument |
-
2004
- 2004-06-10 CN CN 200410025042 patent/CN1278271C/en not_active Expired - Fee Related
Also Published As
| Publication number | Publication date |
|---|---|
| CN1584921A (en) | 2005-02-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN100345153C (en) | Man face image identifying method based on man face geometric size normalization | |
| CN108920992A (en) | A kind of positioning and recognition methods of the medical label bar code based on deep learning | |
| CN101030258A (en) | Dynamic character discriminating method of digital instrument based on BP nerve network | |
| CN101034481A (en) | Method for automatically generating portrait painting | |
| CN101059870A (en) | Image cutting method based on attribute histogram | |
| CN1737824A (en) | Set up the method and apparatus of deterioration dictionary | |
| CN1655178A (en) | Local localization using fast image match | |
| CN1710593A (en) | Hand-characteristic mix-together identifying method based on characteristic relation measure | |
| CN103295009B (en) | Based on the license plate character recognition method of Stroke decomposition | |
| CN101064008A (en) | Method for recognizing print form italic character | |
| CN105809166A (en) | Vehicle license plate recognition method, device and system | |
| CN107944451B (en) | Line segmentation method and system for ancient Tibetan book documents | |
| CN108133216A (en) | The charactron Recognition of Reading method that achievable decimal point based on machine vision is read | |
| CN1438603A (en) | 2-D bar-code automatic reading method based on general office apparatus | |
| CN1278271C (en) | Identifying method for fragmentary printed digits | |
| CN1920856A (en) | Computer assisted calligraphic works distinguishing method between true and false | |
| CN1837853A (en) | Seam eliminating method for mosaic of remote sensing image | |
| CN101034440A (en) | Identification method for spherical fruit and vegetables | |
| CN106815810B (en) | Method and device for determining fitting boundary | |
| CN1265324C (en) | Words and image dividing method on the basis of adjacent edge point distance statistics | |
| CN109858484B (en) | Multi-class transformation license plate correction method based on deflection evaluation | |
| CN1545067A (en) | A method for compressing digitalized archive file using computer | |
| CN1881211A (en) | Graphic retrieve method | |
| CN1489745A (en) | Method for identifying image and device for realizing same | |
| CN111914847B (en) | OCR (optical character recognition) method and system based on template matching |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| C17 | Cessation of patent right | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20061004 |