CN108154144A - A kind of name of vessel character locating method and system based on image - Google Patents
A kind of name of vessel character locating method and system based on image Download PDFInfo
- Publication number
- CN108154144A CN108154144A CN201810031744.8A CN201810031744A CN108154144A CN 108154144 A CN108154144 A CN 108154144A CN 201810031744 A CN201810031744 A CN 201810031744A CN 108154144 A CN108154144 A CN 108154144A
- Authority
- CN
- China
- Prior art keywords
- image
- region
- name
- character
- value
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- 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
- G06V10/22—Image preprocessing by selection of a specific region containing or referencing a pattern; Locating or processing of specific regions to guide the detection or recognition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
- G06F18/2148—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the process organisation or structure, e.g. boosting cascade
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Cheminformatics (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Character Input (AREA)
Abstract
The invention discloses a kind of name of vessel character locating method based on image, this method carries out contrast enhancing pretreatment after first obtaining ship image, and MSER search is carried out, and using MSER regions as name of vessel word candidate region to pretreated image；Screening and filtering is carried out to candidate regions using priori later, is obtained after meeting the candidate regions of priori, candidate regions are made with stroke width transformation, stroke width mean variance threshold value is set, filters some candidate regions for not meeting stroke width feature；A non-legible grader of word is finally trained, final character area is obtained with two graders.The present invention also proposes a kind of system for detecting automatically and positioning name of vessel character in ship image, using Computer Image Processing, computer vision technique and mode identification technology, ship name of vessel character is carried out the basis that fixation and recognition is manually verified as boat brake tube reason, to improve the boat lock efficiency of management.
Description
Technical field
It navigates lock management domain the invention belongs to image procossing and inland river, and in particular to a kind of name of vessel character locating method and be
System.
Background technology
With the booming and huge advantage of inland water transport, Shiplock management work is more aobvious important, and the thing followed is modern
Change Video Supervision Technique to be also widely used in daily management.Compare verification register information in boat brake tube reason is with ship information
The important link of management, it can be that information verification is saved very that the name of vessel automatic identification positioning in monitor video is identified in the link
More manpowers and resource.Therefore, name of vessel character automatic identification location technology can accelerate inland river boat lock modern management process.
Name of vessel character locating challenge based on image mostlys come from that name of vessel font is various, and name of vessel position size is not known,
And image background information is complex, complicated background information mainly includes the character interference such as container text in addition to name of vessel
Word.In addition night illumination deficiency and heavy rain foggy weather etc. will also result in image certain interference and noise, increase for image procossing
Add difficulty.
Invention content
The technical problems to be solved by the invention are：The problem of for background technology, proposes a kind of based on image
Name of vessel character locating method and system, to ship name of vessel character carry out fixation and recognition.
The present invention uses following technical scheme to solve above-mentioned technical problem：
The present invention proposes a kind of name of vessel character locating method based on image, for detecting and positioning automatically in ship image
Name of vessel character, including step：
1), ship image is pre-processed；
2) most stable extremal region MSER search, is carried out to pretreated image, and using MSER regions as name of vessel word
Accord with candidate region；
3) screening and filtering, is carried out to candidate region using the priori including geometry, size, obtains meeting priori
The character candidates area of knowledge；
4) stroke width transformation, is done to the character candidates region for meeting priori, by setting stroke width mean value side
Poor threshold value obtains the character candidates region for meeting stroke width feature；
5), one non-legible grader of word of training is classified to obtain final with two graders to character candidates region
Character area.
Further, the name of vessel character locating method proposed by the invention based on image, in the step 1), to ship figure
Refer to as carrying out pretreatment：Ship image is converted to gray level image, and carries out contrast enhancing pretreatment.
Further, the name of vessel character locating method proposed by the invention based on image is to use Retinex in step 1)
Algorithm carries out contrast enhancing pretreatment；It is specific as follows：
Original image S (x, y) is regarded as to the product of light image L (x, y) and albedo image R (x, y), i.e. S (x, y)
=R (x, y) * L (x, y)；Image is transformed into log-domain：
S (x, y)=log S (x, y),
L (x, y)=log L (x, y),
R (x, y)=log R (x, y),
S=r+l；
Reflecting component noise is removed by normalization：
Index contravariant is asked to change to real number field the reflecting component after denoising and obtain enhanced image.
Further, the name of vessel character locating method proposed by the invention based on image, the step 2) are specific as follows：
Threshold value is taken to carry out binary conversion treatment to image, threshold value is incremented by successively from 0 to 255, in obtained all bianry images
In, if certain connected region variation in image, less than threshold value, which is thus referred to as maximum stable extremal region, connected region
The mathematical definition of variation for q (i)=| Qi+Δ-Qi-Δ|/|Qi|, wherein, QiRepresent a certain connected region when threshold value is i, Δ is
The change of gray threshold, q (i) are that threshold value is i time domains QiChange rate, i.e., when q (i) be local minimum when QiFor
Maximum stable extremal region.
Further, the name of vessel character locating method proposed by the invention based on image, in the step 3), priori
Including：Region area, region rectangle degree, the ratio of width to height of boundary rectangle and boundary rectangle height；Region is a point set, comprising this
Each pixel point coordinates, boundary rectangle are to take minimum enclosed rectangle to the region in region；Region area is picture in the region
Vegetarian refreshments number；Region rectangle degree is region area and boundary rectangle area ratio, and closer to 1, which more connects the value
Nearly rectangle；The ratio between boundary rectangle the ratio of width to height, that is, boundary rectangle width and height；
To more than priori given threshold, the word candidate region for meeting priori can be obtained.
Further, the name of vessel character locating method proposed by the invention based on image, in the step 4), by each
Candidate region all carries out stroke width feature extraction as piece image, first carries out edge detection to image using Canny operators,
The direction gradient value of each edge pixel point is obtained, if edge pixel point p direction gradients value is dp, from point p along gradient direction dpGo out
Hair finds pixel q, judges this direction gradient dqWith dpWhether meet：If in the presence of the point q for the condition that meets,
Then the stroke width value of pixel is between p and q on the pathIf there is no the point q for the condition that meets, which is given up
It abandons；
Such circular treatment obtains the stroke width value of each edge pixel point of the image, so as to calculate image pen
Draw the standard rate of width, i.e. the ratio between standard deviation and mean value；
By setting a standard rate threshold value, the word candidate region for meeting stroke width feature is obtained.
Further, the name of vessel character locating method proposed by the invention based on image, in the step 5),
Cascade non-legible two grader of word is to carry out grader based on local binary feature and Adaboost algorithm
Cascade, the training of the grader are divided into two steps：The training of Weak Classifier and grader cascade；
Wherein, the training process of Weak Classifier is as follows：
It sorts for the characteristic value of all training samples of each feature calculation, and by characteristic value；To tactic each
Element calculates four indexs, the weight and T of whole word samples0, weight and the minimum T of whole non-text samples1, in this element
The weight and S of printed words sheet above0, the weight and S of non-text samples before this element1；Choose currentElement characteristic value and it before
For a value between one characteristic value as threshold value, the error in classification of the threshold value is e=min (S1+(T0-S0),S0+(T1-S1))；
Using the threshold value of error minimum as optimal threshold, Weak Classifier is obtained；
It is as follows that several Weak Classifiers are cascaded into strong classifier process：
If training library sample number is N, wherein word sample is N0, non-text samples number is N1, maximum iteration T, just
Beginningization sample weights are 1/N；All samples of first time repetitive exercise obtain first Weak Classifier；
Improve the sample weights misidentified in previous step；Using misclassification sample and new samples as next Weak Classifier
Training sample；The new Weak Classifier of repetition training obtains T optimal Weak Classifier h after T takes turns iteration1(x),…,hT
(x)；
Weak Classifier is combined into strong classifier as follows：
Wherein, εtRepresent the error rate of each grader, αtFor Weak Classifier weight, the error rate of the value and grader has
It closes.
The present invention also proposes a kind of system for detecting automatically and positioning name of vessel character in ship image, including：
Image pre-processing module, for being pre-processed to ship image；
Character candidates area determining module, is configured to perform following steps：
First, most stable extremal region MSER search is carried out to pretreated image, and using MSER regions as name of vessel
Character candidates region；Secondly, screening and filtering is carried out to candidate region using the priori including geometry, size, obtained
Meet the character candidates area of priori；Then, stroke width transformation is done to the character candidates region for meeting priori, passes through
Stroke width mean variance threshold value is set, obtains the character candidates region for meeting stroke width feature；
The non-legible classifier modules of word classify character candidates region with two graders to obtain final literal field
Domain.
Further, the above-mentioned system for detecting automatically and positioning name of vessel character in ship image of the present invention, word are non-
Word classifier modules include Weak Classifier training unit, grader concatenation unit, wherein：
Weak Classifier training unit sorts for the characteristic value of all training samples of each feature calculation, and by characteristic value,
Four indexs, the weight and T of whole word samples are calculated to tactic each element0, the weight of whole non-text samples
With minimum T1, in the weight and S of this element printed words sheet above0, the weight and S of non-text samples before this element1；It chooses current
Elemental characteristic value and a value before it between a characteristic value are used as threshold value, and the error in classification of the threshold value is e=min (S1+
(T0-S0),S0+(T1-S1))；Using the threshold value of error minimum as optimal threshold, Weak Classifier is obtained；
Several Weak Classifiers are cascaded by force by grader concatenation unit based on local binary feature and Adaboost algorithm
Grader.
The present invention compared with prior art, has following technique effect using above technical scheme：
The present invention using high definition snapshot camera as sensor, integrated use Computer Image Processing, computer vision technique and
Mode identification technology carries out fixation and recognition to ship name of vessel character, as the basis that boat brake tube reason is manually verified, improves boat brake tube
Manage efficiency.
Description of the drawings
Fig. 1 is the method for the present invention flow chart.
Fig. 2 is the stroke width schematic diagram in the present invention.
Fig. 3 is the Adaboost cascade classifier schematic diagrames in the present invention.
Specific embodiment
Technical scheme of the present invention is described in further detail below in conjunction with the accompanying drawings：
Those skilled in the art of the present technique are it is understood that unless otherwise defined, all terms used herein are (including skill
Art term and scientific terminology) there is the meaning identical with the general understanding of the those of ordinary skill in fields of the present invention.Also
It should be understood that those terms such as defined in the general dictionary should be understood that with in the context of the prior art
The consistent meaning of meaning, and unless defined as here, will not be explained with the meaning of idealization or too formal.
The present invention proposes a kind of name of vessel character locating method based on image, and this method is first obtained after ship image to image
Carry out contrast enhancing pretreatment；Most stable extremal region (MSER) is carried out to pretreated image to search for, and by MSER areas
Domain is as name of vessel word candidate region；Screening and filtering is carried out to candidate regions using priori later, wherein priori includes
Character zone the ratio of width to height, width, height, rectangular degree etc.；It obtains after meeting the candidate regions of priori, it is wide to make stroke to candidate regions
Degree transformation, sets stroke width mean variance threshold value, filters some candidate regions for not meeting stroke width feature；Finally train
One non-legible grader of word, final character area is obtained with two graders.
As shown in Figure 1, a kind of name of vessel character locating method based on image disclosed by the invention, includes the following steps：
1) ship image is converted to gray level image and carries out contrast enhancing pretreatment using Retinex algorithm；
2) most stable extremal region (MSER) is carried out to pretreated image to search for, and using MSER regions as name of vessel word
Accord with candidate region；
3) screening and filtering is carried out to candidate region using prioris such as geometry, sizes, obtains meeting priori
Character candidates area；
4) stroke width transformation is done to the word candidate region for meeting priori, sets stroke width mean variance threshold
Value, obtains the character candidates region for meeting stroke width feature；
5) one non-legible grader of word of training, classifies character candidates region with two graders to obtain final text
Block domain.
Specifically, in the step 1), ship image is converted to gray level image and is compared using Retinex algorithm
Degree enhancing pretreatment：
Ship image is obtained by camera first, and converts images into gray-scale map；Then using Retinex algorithm into
Row contrast enhancement processing, Retinex algorithm can go back original image well for greasy weather, backlight scene.
It is as follows that Retinex algorithm promotes contrast step；Original image S (x, y) can regard light image L (x, y) as
With the product of albedo image R (x, y), i.e. S (x, y)=R (x, y) * L (x, y)；Image is transformed into log-domain：
S (x, y)=log S (x, y),
L (x, y)=log L (x, y),
R (x, y)=log R (x, y),
S=r+l；
The noise source of Retinex theory hypothesis images is different in each position reflectivity of image, therefore removes reflectogram
The noise of picture can go back original image, and reflecting component noise can be removed by normalizing,After denoising
Reflecting component seek index, contravariant changes to real number field and obtains enhanced image.
In the step 2), most stable extremal region (MSER) is carried out to pretreated image and is searched for, and by MSER areas
Domain is as name of vessel character candidates region：
Most stable extremal region algorithm has stronger affine transformation invariance to image.The extraction of most stable extremal region
Method is as follows；Threshold value is taken to carry out binary conversion treatment to a width gray level image (gray value 0-255), threshold value is passed successively from 0 to 255
Increase, in obtained all bianry images, certain connected regions in image vary less, even without variation, then the region
Thus referred to as maximum stable extremal region, connected region variation mathematical definition for q (i)=| Qi+Δ-Qi-Δ|/|Qi|。
In the step 3), screening and filtering is carried out to candidate region using prioris such as geometry, sizes, is accorded with
Close the character candidates area of priori：
Priori mainly includes, region area, region rectangle degree, the ratio of width to height of boundary rectangle and boundary rectangle height
Deng；Region is a point set, and comprising pixel point coordinates each in the region, boundary rectangle is exactly that minimum external square is taken to the region
Shape；Region area is pixel number in the region；Region rectangle degree is region area and boundary rectangle area ratio, is somebody's turn to do
Value is closer to 1, and the region shape is closer to rectangle；The ratio between boundary rectangle the ratio of width to height, that is, boundary rectangle width and height；To more than
Priori given threshold can obtain the word candidate region for meeting priori.
In the step 4), stroke width transformation is done to the word candidate region for meeting priori, sets stroke width
Mean variance threshold value obtains the character candidates region for meeting stroke width feature：
Stroke width feature substantially belongs to the exclusive feature of word, it is however generally that unified text all has unified pen
Width is drawn, sees attached drawing 2.Stroke width calculating process is as follows, each candidate region carries out stroke width as piece image
Feature extraction first carries out edge detection to image using Canny operators, obtains the direction gradient value of each edge pixel point, if
Edge pixel point p direction gradients value is dp, from point p along gradient direction dpSet off in search pixel q, this direction gradient dqWith dp
It is substantially oppositeIf in the presence of the point q for the condition that meets, on the path between p and q pixel stroke width
It is worth and isIf there is no the point q for the condition that meets, which gives up；Each edge pixel point of the image is obtained in this way
Stroke width value, so as to calculate the standard rate of image stroke width；Set standard rate (standard deviation and a mean value
The ratio between) threshold value, obtain the word candidate region for meeting stroke width feature.
In the step 5), one cascade non-legible two grader of word of training, with two graders to character candidates region
Classified to obtain final character area：
Cascade non-legible two grader of word is to carry out grader based on local binary feature and Adaboost algorithm
Cascade, cascade classifier schematic diagram such as attached drawing 3.The training of the grader is largely divided into two steps, the training and classification of Weak Classifier
Device cascades；
The training process of Weak Classifier is as follows：For the characteristic value of all training samples of each feature calculation, and by feature
Value sequence；Four indexs, the weight and T of whole word samples are calculated to tactic each element0, whole non-text samples
Weight and minimum T0, in the weight and S of this element printed words sheet above0, the weight and S of non-text samples before this element1；Choosing
Taking currentElement characteristic value and a value before it between a characteristic value, the error in classification of the threshold value is e=as threshold value
min(S1+(T0-S0),S0+(T1-S1))；Using the threshold value of error minimum as optimal threshold, Weak Classifier is obtained.
It is as follows that several Weak Classifiers are cascaded into strong classifier process：Training library sample number is N, and wherein word sample is
N0It is N with non-text samples number1, maximum iteration T, initialization sample weight is 1/N；All samples of first time repetitive exercise
This training obtains first Weak Classifier；Improve the sample weights misidentified in previous step；By misclassification sample and new samples
Training sample as next Weak Classifier；The new Weak Classifier of repetition training, T obtain T optimal weak typings after taking turns iteration
Device；Weak Classifier is combined into strong classifier as follows：
Below with a specific example to the technical program explanation for example：
The first step：Ship image is converted to gray level image and carries out contrast enhancing pretreatment using Retinex algorithm.
Second step：To pretreated image carry out most stable extremal region (MSER) search for, and using MSER regions as
Name of vessel character candidates region.
Third walks：Screening and filtering is carried out to candidate region using prioris such as geometry, sizes, obtains meeting priori
The character candidates area of knowledge.
4th step：Stroke width transformation, setting stroke width mean value side are done to the word candidate region for meeting priori
Poor threshold value obtains the character candidates region for meeting stroke width feature.
5th step：One non-legible grader of word of training, is classified to obtain with two graders to character candidates region
Final character area.
The above is only some embodiments of the present invention, it is noted that for the ordinary skill people of the art
For member, various improvements and modifications may be made without departing from the principle of the present invention, these improvements and modifications also should
It is considered as protection scope of the present invention.
Claims (10)
1. a kind of name of vessel character locating method based on image, for detecting and positioning name of vessel character in ship image automatically,
It is characterized in that, including step：
1), ship image is pre-processed；
2) most stable extremal region MSER search, is carried out to pretreated image, and is waited MSER regions as name of vessel character
Favored area；
3) screening and filtering, is carried out to candidate region using the priori including geometry, size, obtains meeting priori
Character candidates area；
4) stroke width transformation, is done to the character candidates region for meeting priori, by setting stroke width mean variance threshold
Value, obtains the character candidates region for meeting stroke width feature；
5), one non-legible grader of word of training, classifies character candidates region with two graders to obtain final word
Region.
2. the name of vessel character locating method according to claim 1 based on image, which is characterized in that in the step 1),
Pretreatment is carried out to ship image to refer to：Ship image is converted to gray level image, and carries out contrast enhancing pretreatment.
3. the name of vessel character locating method according to claim 2 based on image, which is characterized in that be to use in step 1)
Retinex algorithm carries out contrast enhancing pretreatment；It is specific as follows：
Original image S (x, y) is regarded as to the product of light image L (x, y) and albedo image R (x, y), i.e. S (x, y)=R
(x,y)*L(x,y)；Image is transformed into log-domain：
S (x, y)=logS (x, y),
L (x, y)=logL (x, y),
R (x, y)=logR (x, y),
S=r+l；
Reflecting component noise is removed by normalization：
Index contravariant is asked to change to real number field the reflecting component after denoising and obtain enhanced image.
4. the name of vessel character locating method according to claim 1 based on image, which is characterized in that the step 2) is specific
It is as follows：
Threshold value is taken to carry out binary conversion treatment to image, threshold value is incremented by successively from 0 to 255, in obtained all bianry images, if
Certain connected region variation in image is less than threshold value, then the region is thus referred to as maximum stable extremal region, connected region variation
Mathematical definition for q (i)=| Qi+Δ-Qi-Δ|/|Qi|, wherein QiRepresent a certain connected region when threshold value is i, Δ is gray scale threshold
The change of value, q (i) are that threshold value is i time domains QiChange rate, i.e., when q (i) be local minimum when QiIt is maximum steady
Determine extremal region.
5. the name of vessel character locating method according to claim 1 based on image, which is characterized in that in the step 3),
Priori includes：Region area, region rectangle degree, the ratio of width to height of boundary rectangle and boundary rectangle height；Region is a point
Collection, comprising pixel point coordinates each in the region, boundary rectangle is to take minimum enclosed rectangle to the region；Region area is should
Pixel number in region；Region rectangle degree is region area and boundary rectangle area ratio, and the value is closer to 1, the region
Shape is closer to rectangle；The ratio between boundary rectangle the ratio of width to height, that is, boundary rectangle width and height；
To more than priori given threshold, the word candidate region for meeting priori can be obtained.
6. the name of vessel character locating method according to claim 1 based on image, which is characterized in that in the step 4),
Stroke width feature extraction is carried out using each candidate region as piece image, first image is carried out using Canny operators
Edge detection obtains the direction gradient value of each edge pixel point, if edge pixel point p direction gradients value is dp, from point p along ladder
Spend direction dpSet off in search pixel q judges this direction gradient dqWith dpWhether meet：Meet if existing
The point q of condition, then the stroke width value of pixel is between p and q on the pathIf there is no the point q for the condition that meets,
The path is given up；
Such circular treatment obtains the stroke width value of each edge pixel point of the image, wide so as to calculate image stroke
The standard rate of degree, i.e. the ratio between standard deviation and mean value；
By setting a standard rate threshold value, the word candidate region for meeting stroke width feature is obtained.
7. the name of vessel character locating method according to claim 1 based on image, which is characterized in that in the step 5),
Cascade non-legible two grader of word is the cascade that grader is carried out based on local binary feature and Adaboost algorithm,
The training of the grader is divided into two steps：The training of Weak Classifier and grader cascade.
8. the name of vessel character locating method according to claim 7 based on image, which is characterized in that wherein, Weak Classifier
Training process it is as follows：
It sorts for the characteristic value of all training samples of each feature calculation, and by characteristic value；To tactic each element
Calculate four indexs, the weight and T of whole word samples0, weight and the minimum T of whole non-text samples1, in this element above
The weight and S of printed words sheet0, the weight and S of non-text samples before this element1；Choose currentElement characteristic value and one before it
For a value between characteristic value as threshold value, the error in classification of the threshold value is e=min (S1+(T0-S0),S0+(T1-S1))；It will be accidentally
Poor minimum threshold value obtains Weak Classifier as optimal threshold；
It is as follows that several Weak Classifiers are cascaded into strong classifier process：
If training library sample number is N, wherein word sample is N0, non-text samples number is N1, maximum iteration T, initialization
Sample weights are 1/N；All samples of first time repetitive exercise obtain first Weak Classifier；
Improve the sample weights misidentified in previous step；Using misclassification sample and new samples as the instruction of next Weak Classifier
Practice sample；The new Weak Classifier of repetition training obtains T optimal Weak Classifier h after T takes turns iteration1(x),…,hT(x)；
Weak Classifier is combined into strong classifier as follows：
Wherein, εtRepresent the error rate of each grader, αtFor Weak Classifier weight, the value is related with the error rate of grader.
9. a kind of system for detecting automatically and positioning name of vessel character in ship image, which is characterized in that including：
Image pre-processing module, for being pre-processed to ship image；
Character candidates area determining module, is configured to perform following steps：
First, most stable extremal region MSER search is carried out to pretreated image, and using MSER regions as name of vessel character
Candidate region；Secondly, screening and filtering is carried out to candidate region using the priori including geometry, size, is met
The character candidates area of priori；Then, stroke width transformation is done to the character candidates region for meeting priori, passes through setting
Stroke width mean variance threshold value obtains the character candidates region for meeting stroke width feature；
The non-legible classifier modules of word classify character candidates region with two graders to obtain final character area.
10. a kind of system for detecting automatically and positioning name of vessel character in ship image according to claim 9, special
Sign is that the non-legible classifier modules of word include Weak Classifier training unit, grader concatenation unit, wherein：
Weak Classifier training unit sorts for the characteristic value of all training samples of each feature calculation, and by characteristic value, to suitable
Each element of sequence arrangement calculates four indexs, the weight and T of whole word samples0, the weight of whole non-text samples and most
Small T1, in the weight and S of this element printed words sheet above0, the weight and S of non-text samples before this element1；Choose currentElement
Characteristic value and a value before it between a characteristic value are used as threshold value, and the error in classification of the threshold value is e=min (S1+(T0-
S0),S0+(T1-S1))；Using the threshold value of error minimum as optimal threshold, Weak Classifier is obtained；
Several Weak Classifiers are cascaded into strong classification by grader concatenation unit based on local binary feature and Adaboost algorithm
Device.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810031744.8A CN108154144A (en) | 2018-01-12 | 2018-01-12 | A kind of name of vessel character locating method and system based on image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810031744.8A CN108154144A (en) | 2018-01-12 | 2018-01-12 | A kind of name of vessel character locating method and system based on image |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108154144A true CN108154144A (en) | 2018-06-12 |
Family
ID=62461539
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810031744.8A Pending CN108154144A (en) | 2018-01-12 | 2018-01-12 | A kind of name of vessel character locating method and system based on image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108154144A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110197181A (en) * | 2019-05-31 | 2019-09-03 | 烽火通信科技股份有限公司 | A kind of cable character detection method and system based on OCR |
| CN110766002A (en) * | 2019-10-08 | 2020-02-07 | 浙江大学 | Ship name character region detection method based on deep learning |
| CN111027544A (en) * | 2019-11-29 | 2020-04-17 | 武汉虹信技术服务有限责任公司 | MSER license plate positioning method and system based on visual saliency detection |
| CN111767909A (en) * | 2020-05-12 | 2020-10-13 | 合肥联宝信息技术有限公司 | Character recognition method and device and computer readable storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106529380A (en) * | 2015-09-15 | 2017-03-22 | 阿里巴巴集团控股有限公司 | Image identification method and device |
| CN107045634A (en) * | 2017-05-02 | 2017-08-15 | 电子科技大学 | A kind of text positioning method based on maximum stable extremal region and stroke width |
| CN107563377A (en) * | 2017-08-30 | 2018-01-09 | 江苏实达迪美数据处理有限公司 | It is a kind of to detect localization method using the certificate key area of edge and character area |
-
2018
- 2018-01-12 CN CN201810031744.8A patent/CN108154144A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106529380A (en) * | 2015-09-15 | 2017-03-22 | 阿里巴巴集团控股有限公司 | Image identification method and device |
| CN107045634A (en) * | 2017-05-02 | 2017-08-15 | 电子科技大学 | A kind of text positioning method based on maximum stable extremal region and stroke width |
| CN107563377A (en) * | 2017-08-30 | 2018-01-09 | 江苏实达迪美数据处理有限公司 | It is a kind of to detect localization method using the certificate key area of edge and character area |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110197181A (en) * | 2019-05-31 | 2019-09-03 | 烽火通信科技股份有限公司 | A kind of cable character detection method and system based on OCR |
| CN110197181B (en) * | 2019-05-31 | 2021-04-30 | 烽火通信科技股份有限公司 | Cable character detection method and system based on OCR |
| CN110766002A (en) * | 2019-10-08 | 2020-02-07 | 浙江大学 | Ship name character region detection method based on deep learning |
| CN110766002B (en) * | 2019-10-08 | 2022-03-22 | 浙江大学 | Ship name character region detection method based on deep learning |
| CN111027544A (en) * | 2019-11-29 | 2020-04-17 | 武汉虹信技术服务有限责任公司 | MSER license plate positioning method and system based on visual saliency detection |
| CN111027544B (en) * | 2019-11-29 | 2023-09-29 | 武汉虹信技术服务有限责任公司 | MSER license plate positioning method and system based on visual saliency detection |
| CN111767909A (en) * | 2020-05-12 | 2020-10-13 | 合肥联宝信息技术有限公司 | Character recognition method and device and computer readable storage medium |
| CN111767909B (en) * | 2020-05-12 | 2022-02-01 | 合肥联宝信息技术有限公司 | Character recognition method and device and computer readable storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108256493A (en) | A kind of traffic scene character identification system and recognition methods based on Vehicular video | |
| Kakani et al. | Improved OCR based automatic vehicle number plate recognition using features trained neural network | |
| US10817741B2 (en) | Word segmentation system, method and device | |
| CN108154144A (en) | A kind of name of vessel character locating method and system based on image | |
| Zang et al. | Traffic sign detection based on cascaded convolutional neural networks | |
| US20140079316A1 (en) | Segmentation co-clustering | |
| Huang et al. | Text detection and recognition in natural scene images | |
| Gilly et al. | A survey on license plate recognition systems | |
| Hatolkar et al. | A survey on road traffic sign recognition system using convolution neural network | |
| Elbamby et al. | Real-time automatic multi-style license plate detection in videos | |
| Adytia et al. | Indonesian license plate detection and identification using deep learning | |
| Sahu et al. | A comparative analysis of deep learning approach for automatic number plate recognition | |
| Jiang et al. | An efficient and unified recognition method for multiple license plates in unconstrained scenarios | |
| CN113989806A (en) | Extensible CRNN bank card number identification method | |
| El-Said | Shadow aware license plate recognition system | |
| Ghoshal et al. | An improved scene text and document image binarization scheme | |
| Xavier et al. | Automatic detection and recognition of text in traffic sign boards based on word recognizer | |
| Khan et al. | Text detection and recognition on traffic panel in roadside imagery | |
| Sugiyono et al. | Extracting information from vehicle registration plate using OCR tesseract | |
| Mutholib et al. | Optimization of ANPR algorithm on Android mobile phone | |
| Saidallah et al. | Classification and comparison of license plates localization algorithms | |
| Sathya et al. | Vehicle license plate recognition (vlpr) | |
| Maske et al. | A Review on Detection and Recognition of Text | |
| Khan et al. | Character segmentation on traffic panels using visual appearance from roadside imagery | |
| CN109886265A (en) | A kind of car door limiter detection method based on Adaboost and template matching |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180612 |