CN107818321A - A kind of watermark date recognition method for vehicle annual test - Google Patents
A kind of watermark date recognition method for vehicle annual test Download PDFInfo
- Publication number
- CN107818321A CN107818321A CN201710949524.9A CN201710949524A CN107818321A CN 107818321 A CN107818321 A CN 107818321A CN 201710949524 A CN201710949524 A CN 201710949524A CN 107818321 A CN107818321 A CN 107818321A
- Authority
- CN
- China
- Prior art keywords
- watermark
- date
- annual test
- image
- vehicle annual
- 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.)
- Withdrawn
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
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
- G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of watermark date recognition method for vehicle annual test, comprise the following steps：Watermark candidate regions are positioned, and extract watermark candidate region image, identify watermark date and time information, and verify its legitimacy, legal watermark date and time information is exported into recognition result with character string forms, if not conforming to rule preserves vehicle annual test picture, in case artificial enquiry.Present invention is mainly applied to the watermark date recognition in automotive vehicle annual test, the method for efficiently solving conventional watermark date recognition has what just positioning was forbidden, poor anti jamming capability, the problems such as recognition accuracy is low.So as to substituted for existing manual identified mode in vehicle annual test, while manpower has been saved, the accuracy rate of watermark date recognition is effectively improved.
Description
Technical field
It is more particularly to a kind of to be used for vehicle year the present invention relates to the artificial intelligence identification technology field of automotive vehicle annual test
The watermark date recognition method of inspection.
Background technology
With the fast development of social economy and the continuous improvement of living standards of the people, Urban vehicles poputation increases rapidly
It is long.The workload of automotive vehicle annual test also sharply increases therewith.In order to save cost of labor, checkability, traditional vehicle are improved
Many projects in annual test are progressively from manual identified mode to Intelligent Recognition mode transition.Wherein, the watermark of vehicle annual test
Date recognition can be the ageing offer foundation that supervisory organ supervision and inspection person shoots photo, prevent inspector's going through vehicle
History photo is used for annual test then.But the method for conventional watermark date recognition has what just positioning was forbidden, antijamming capability
Difference, the problems such as recognition accuracy is low.
How on the premise of cost of labor is saved, the accuracy rate of watermark date recognition is improved, so as to be effectively vehicle
Inspection and supervision mechanism provides supervision foundation, is the technical problem for being badly in need of solving.
The content of the invention
The purpose of the present invention is：It is proposed a kind of watermark date recognition method for vehicle annual test, this method efficiently and accurately,
Can be in a manner of effectively substituting manual identified, to meet nowadays the needs of to vehicle annual test operating efficiency, accuracy rate.
The technical solution adopted for the present invention to solve the technical problems is：
A kind of watermark date recognition method for vehicle annual test, comprise the following steps：
S1, vehicle annual test picture transformed into hsv color space, give HSV triple channels that threshold value is set, filter out watermark color
Pixel, several point sets are obtained according to connected relation, and calculate their boundary rectangle, the region of the boundary rectangle is water
Print candidate region；
If S2, the watermark candidate region be present, the text objects detection model based on deep learning network is used, according to
It is secondary that the watermark date is detected in each watermark candidate region and records score；If the watermark candidate region is not present, preserve
Vehicle annual test picture, in case artificial examination, end of identification；
If S3, detecting the watermark date, the watermark date position of highest scoring is recorded, and obtains and schemes corresponding to it
Picture；If being not detected by the watermark date, vehicle annual test picture is preserved, in case artificial examination, end of identification；
S4, using the water in the image obtained in the Character segmentation identification Model Identification step S3 based on deep learning network
Print date and time information；
The legitimacy of the watermark date and time information extracted in S5, checking procedure S4；
If S6, step S5 check results are legal, recognition result is exported with character string forms；Otherwise vehicle annual test is preserved
Picture, in case artificial examination.
Further, the obtaining step of the text objects detection model is as follows：
S21, prepare a collection of vehicle annual test picture with watermark as sample；
S22, the watermark candidate region image in samples pictures is obtained using the method described in step S1, and preserved；
S23, mark in the image of the watermark candidate region using rectangle frame position where the watermark date；
S24, deep neural network model, acquisition text objects detection mould are detected using the data training objective marked
Type.
Further, the obtaining step of the Character segmentation identification model is as follows：
S31, the watermark date position marked based on S23 with rectangle, obtain watermark date image, as mark sample；
S32, the position for marking on watermark date image using rectangle frame all characters in the date and classification；
S33, according to the position and classification marked in step S32, calculate on watermark date image belonging to each pixel
Classification, generate a width and watermark date picture size identical label image；
S34, a data set is formed using the watermark date image and the label image, and utilize Training scene segmentation
Deep neural network model, obtain Character segmentation identification model.
Further, the legitimacy of the watermark date and time information includes length legitimacy and numerical value legitimacy；Wherein, length is closed
It must be respectively 4,2,2 character lengths that method, which refers to year, month, day field, and numerical value legitimacy refers to the numerical value of identified year, month, day
Must be in the number range of natural year, month, day.
The beneficial effects of the invention are as follows：Present invention is mainly applied to the watermark date recognition in automotive vehicle annual test, have
Effect solves method inaccurate, the poor anti jamming capability that first positioning be present of conventional watermark date recognition, and recognition accuracy is low etc. asks
Topic.So as to substituted for existing manual identified mode in vehicle annual test, while manpower has been saved, water is effectively improved
Print the accuracy rate of date recognition.
Brief description of the drawings
Fig. 1 is the flow chart of vehicle annual test watermark date recognition of the present invention.
Fig. 2 is the schematic diagram of text objects detection module of the present invention.
Fig. 3 is the schematic diagram of character recognition module of the present invention.
Embodiment
Below in conjunction with accompanying drawing.The present invention will be further described.
The vehicle annual test watermark date recognition method idiographic flow of the present invention is as shown in figure 1, specific implementation step is as follows：
S1, vehicle annual test picture transformed into hsv color space, wherein, H represents tone, interval for [0,180), S
Saturation degree is represented, interval is [0,255], and V represents lightness, and interval is [0,255].According to watermark color to HSV tri-
Passage sets screening scope, filters out the pixel of watermark candidate region.Preferably, by taking watermark red as an example, the screening of H passages
Section be arranged to [0,20) U [160,180), while the screening section of S and V passages is both configured to [30,255].Sieve will be met
The pixel value of condition is selected to be labeled as 255, other pixel values are labeled as 0, obtain covering with vehicle annual test dimension of picture identical two-value
Code figure.Morphological dilations are done to two-value mask figure, the selection of expansive working core is rectangle core, and its length and width is disposed as mask figure width
1 percent.The connected relation of mask figure after expanding is analyzed again, to marking the institute for being a little, according to the connection of 8 neighborhoods
Property, it is divided into several mutually disjunct point sets.The extraneous rectangle of these point sets is calculated respectively, and width is less than mask figure width
/ 10th rectangle is deleted, and remaining rectangular area is all used as watermark candidate region.
The image of watermark candidate region described in S2, acquisition S1.As shown in Fig. 2 using the text based on deep learning network
This target detection model, the watermark date is detected in each candidate region successively.Testing result be expressed as vector [c, s, x, y, w,
H], wherein, c represents classification, there was only a kind of classification, i.e. watermark date in this detection model, therefore it is 1 that c value is permanent；S is represented
Point, span is the decimal between 0 to 1, and the watermark date confidence level that the higher expression of score detects is higher；X, y, w, h points
Not Biao Shi rectangle position where the watermark date top left co-ordinate and width it is high.
Wherein, the acquisition methods of the text objects detection model based on deep learning network are as follows：
Prepare a collection of vehicle annual test picture with watermark as sample.The watermark date is likely to occur in vehicle annual test picture
In the diverse location of watermark region.Therefore, the diversity of watermark location is ensured during selection sample.Obtained using the method described in S1
Watermark candidate region into samples pictures, the image in the region is preserved, it is artificial to reject wherein and do not include the picture of watermark,
Complete the collection of watermark region picture.The position where the watermark date is marked in watermark region picture using rectangle frame.Note
Meaning, the rectangle frame of mark answer complete packet date content containing watermark.If not including the watermark date in some watermark region pictures, do not mark
Position is noted, still retains these samples pictures.Deep neural network model is detected using the data training objective marked, is obtained
Obtain text objects detection model (common knowledge, not repeating hereby).
S3, the score for preserving all watermark dates for detecting of S2, the index of position and place watermark region image.Press
Arranged according to score descending, the watermark date is intercepted from corresponding watermark region image according to the position on the watermark date of highest scoring
Image, for follow-up identification.
S4, as shown in figure 3, using the watermark date obtained by Character segmentation identification Model Identification S3 based on deep learning network
Date and time information in image.The output result of identification model is represented by and watermark date picture size identical gray level image.
In gray-scale map, the gray value of each point only has the possible value of N+1 kinds, and wherein N represents the classification number of character, in addition, background
Class also occupies a kind of gray value.In this programme, the classification of character arrives " 9 " ten numerals and symbol "-" including " 0 ", therefore N takes
It is worth for 11, then the value each put in gray-scale map has 12 kinds of possibility, corresponding 12 labels for training Character segmentation identification model, ash
Angle value represents that the point belongs to background for 0, and gray value 1 to 10 represents that the point belongs to character " 0 " and arrived " 9 " respectively, and gray value 11 represents
Symbol "-".So far the classification of each pixel is predicted, further according to the distribution relation of pixel, you can predict the position of character
Put and classification.Concrete operations are as follows, and the gray-scale map exported first to Character segmentation identification model is respectively with above-mentioned 11 kinds of gray values
Filtering, 11 two-value mask figures are obtained, such as filtering gray value is 3, then the mask value for the point that gray value is 3 is set into 255, its
Remaining point is all set to 0.Morphological dilations, expansive working core selection rectangle core are done to two-value mask figure again, its length and width is disposed as
1 the percent of mask figure width.According to the connectedness of 8 neighborhoods, it is mutually disjunct that the point that mask value is 255 is divided into several
Point set, calculate the number and its boundary rectangle of point in point set, and gray value label corresponding to record.According to the number put in point set
Sort from big to small, boundary rectangle corresponding to preceding ten point sets is the position of watermark date character, and its gray value label identifies
The character class gone out.
Wherein, the acquisition methods of the scene cut model based on deep learning network are as follows：
Based on the watermark date position marked in S2 with rectangle, watermark date image is obtained, as mark sample.Using square
Shape frame marks position and the classification of all characters in the date on watermark date image.Pay attention to, the rectangle frame of mark should be complete
Should not there is any overlapping the rectangle frame of adjacent character when including the content of date character, but marking.For the mark of character class,
Character different in " 9 " and "-" etc. 11 is arrived containing " 0 " altogether in date content, corresponding label is labeled as 1 to 11 successively.
According to character position and classification has been marked, the classification belonging to each pixel on watermark date image is calculated, wherein, rectangle
The pixel of inframe be collectively labeled as belonging to character class label, i.e., the value among 1 to 11.The pixel of outer rectangular frame is the back of the body
Scape, label 0.So one width of generation and watermark date picture size identical label image.Use above-mentioned watermark date image
The data set formed with label image, Training scene segmentation deep neural network model, it is (known to obtain Character segmentation identification model
General knowledge, do not repeat hereby).
The legitimacy of date information in S5, verification recognition result, including length legitimacy and numerical value legitimacy.Based on S4
The position of 10 characters of output and classification, first according to the position of two separator "-", are divided into year, month, day by recognition result
Three fields.Legal year, month, day field must be respectively 4,2,2 character lengths, and year, month, day numerical value must be in nature
In the number range of year, month, day.
If S6, the date information of identification are legal, recognition result is exported with character string forms.
Pay attention to, for days in the problem of watermark date is not present in watermark candidate region, S3 and S5 are not present in S2
Day illegal problem of information, reason are generally that detected vehicle annual test picture does not stamp watermark in itself or picture quality is deposited
In problem, therefore problem picture is preserved in case artificial examination.
The advantages of general principle, principal character and this programme of this programme has been shown and described above.The technology of the industry
Personnel are it should be appreciated that this programme is not restricted to the described embodiments, and the simply explanation described in above-described embodiment and specification is originally
The principle of scheme, on the premise of this programme spirit and scope are not departed from, this programme also has various changes and modifications, these changes
Change and improve and both fall within the range of claimed this programme.This programme be claimed scope by appended claims and its
Equivalent thereof.
Claims (4)
- A kind of 1. watermark date recognition method for vehicle annual test, it is characterised in that comprise the following steps：S1, vehicle annual test picture transformed into hsv color space, give HSV triple channels that threshold value is set, filter out the picture of watermark color Vegetarian refreshments, several point sets are obtained according to connected relation, and calculate their boundary rectangle, the region of the boundary rectangle is that watermark is waited Favored area；If S2, the watermark candidate region be present, the text objects detection model based on deep learning network is used, is existed successively The watermark date is detected in each watermark candidate region and records score；If the watermark candidate region is not present, vehicle is preserved Annual test picture, in case artificial examination, end of identification；If S3, detecting the watermark date, the watermark date position of highest scoring is recorded, and obtains its corresponding image； If being not detected by the watermark date, vehicle annual test picture is preserved, in case artificial examination, end of identification；S4, using the watermark day in the image obtained in the Character segmentation identification Model Identification step S3 based on deep learning network Phase information；The legitimacy of the watermark date and time information extracted in S5, checking procedure S4；If S6, step S5 check results are legal, recognition result is exported with character string forms；Otherwise vehicle annual test figure is preserved Piece, in case artificial examination.
- 2. watermark date recognition method as claimed in claim 1, it is characterised in that the acquisition of the text objects detection model Step is as follows：S21, prepare a collection of vehicle annual test picture with watermark as sample；S22, the watermark candidate region image in samples pictures is obtained using the method described in step S1, and preserved；S23, mark in the image of the watermark candidate region using rectangle frame position where the watermark date；S24, deep neural network model, acquisition text objects detection model are detected using the data training objective marked.
- 3. watermark date recognition method as claimed in claim 2, it is characterised in that the acquisition of the Character segmentation identification model Step is as follows：S31, the watermark date position marked based on S23 with rectangle, obtain watermark date image, as mark sample；S32, the position for marking on watermark date image using rectangle frame all characters in the date and classification；S33, according to the position and classification marked in step S32, calculate the class belonging to each pixel on watermark date image Not, a width and watermark date picture size identical label image are generated；S34, a data set is formed using the watermark date image and the label image, and utilize Training scene segmentation depth Neural network model, obtain Character segmentation identification model.
- 4. watermark date recognition method as claimed in claim 2, it is characterised in that the legitimacy bag of the watermark date and time information Include length legitimacy and numerical value legitimacy；Wherein, it must be respectively 4,2,2 character length that length legitimacy, which refers to year, month, day field, Degree, the numerical value that numerical value legitimacy refers to identified year, month, day must be in the number range of natural year, month, day.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710949524.9A CN107818321A (en) | 2017-10-13 | 2017-10-13 | A kind of watermark date recognition method for vehicle annual test |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710949524.9A CN107818321A (en) | 2017-10-13 | 2017-10-13 | A kind of watermark date recognition method for vehicle annual test |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107818321A true CN107818321A (en) | 2018-03-20 |
Family
ID=61608217
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710949524.9A Withdrawn CN107818321A (en) | 2017-10-13 | 2017-10-13 | A kind of watermark date recognition method for vehicle annual test |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107818321A (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108805519A (en) * | 2018-05-18 | 2018-11-13 | 赵崇标 | Papery schedule electronization generation method, device and electronic agenda table generating method |
| CN110442680A (en) * | 2019-08-05 | 2019-11-12 | 西南财经大学 | The embedded vector generation method of the ideograph of view-based access control model |
| CN110852896A (en) * | 2019-12-22 | 2020-02-28 | 上海眼控科技股份有限公司 | Date validity judgment method, date validity judgment device, date validity judgment equipment and storage medium |
| CN110991488A (en) * | 2019-11-08 | 2020-04-10 | 广州坚和网络科技有限公司 | Image watermark identification method using deep learning model |
| CN110990801A (en) * | 2019-11-29 | 2020-04-10 | 深圳市商汤科技有限公司 | Information verification method and device, electronic equipment and storage medium |
| CN111931721A (en) * | 2020-09-22 | 2020-11-13 | 苏州科达科技股份有限公司 | Method and device for detecting color and number of annual inspection label and electronic equipment |
| CN112907433A (en) * | 2021-03-25 | 2021-06-04 | 苏州科达科技股份有限公司 | Digital watermark embedding method, digital watermark extracting device, digital watermark embedding apparatus, digital watermark extracting apparatus, and digital watermark extracting medium |
-
2017
- 2017-10-13 CN CN201710949524.9A patent/CN107818321A/en not_active Withdrawn
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108805519A (en) * | 2018-05-18 | 2018-11-13 | 赵崇标 | Papery schedule electronization generation method, device and electronic agenda table generating method |
| CN108805519B (en) * | 2018-05-18 | 2021-09-28 | 赵崇标 | Electronic generation method and device for paper schedule and electronic schedule generation method |
| CN110442680A (en) * | 2019-08-05 | 2019-11-12 | 西南财经大学 | The embedded vector generation method of the ideograph of view-based access control model |
| CN110991488A (en) * | 2019-11-08 | 2020-04-10 | 广州坚和网络科技有限公司 | Image watermark identification method using deep learning model |
| CN110991488B (en) * | 2019-11-08 | 2023-10-20 | 广州坚和网络科技有限公司 | Picture watermark identification method using deep learning model |
| CN110990801A (en) * | 2019-11-29 | 2020-04-10 | 深圳市商汤科技有限公司 | Information verification method and device, electronic equipment and storage medium |
| CN110990801B (en) * | 2019-11-29 | 2022-05-17 | 深圳市商汤科技有限公司 | Information verification method and device, electronic equipment and storage medium |
| CN110852896A (en) * | 2019-12-22 | 2020-02-28 | 上海眼控科技股份有限公司 | Date validity judgment method, date validity judgment device, date validity judgment equipment and storage medium |
| CN111931721A (en) * | 2020-09-22 | 2020-11-13 | 苏州科达科技股份有限公司 | Method and device for detecting color and number of annual inspection label and electronic equipment |
| CN111931721B (en) * | 2020-09-22 | 2023-02-28 | 苏州科达科技股份有限公司 | Method and device for detecting color and number of annual inspection label and electronic equipment |
| CN112907433A (en) * | 2021-03-25 | 2021-06-04 | 苏州科达科技股份有限公司 | Digital watermark embedding method, digital watermark extracting device, digital watermark embedding apparatus, digital watermark extracting apparatus, and digital watermark extracting medium |
| CN112907433B (en) * | 2021-03-25 | 2023-06-02 | 苏州科达科技股份有限公司 | Digital watermark embedding method, digital watermark extracting method, digital watermark embedding device, digital watermark extracting device, digital watermark embedding equipment and digital watermark extracting medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107818321A (en) | A kind of watermark date recognition method for vehicle annual test | |
| CN105608456B (en) | A kind of multi-direction Method for text detection based on full convolutional network | |
| CN109284758B (en) | Invoice seal eliminating method and device and computer storage medium | |
| CN103049763B (en) | Context-constraint-based target identification method | |
| CN109784326A (en) | A kind of vehicle chassis detection method based on deep learning | |
| CN110059694A (en) | The intelligent identification Method of lteral data under power industry complex scene | |
| CN111340784B (en) | Mask R-CNN-based image tampering detection method | |
| CN104408449B (en) | Intelligent mobile terminal scene literal processing method | |
| CN106056118A (en) | Recognition and counting method for cells | |
| CN104778470B (en) | Text detection based on component tree and Hough forest and recognition methods | |
| CN108921201B (en) | Dam defect identification and classification method based on feature combination with CNN | |
| CN108520278A (en) | A kind of road surface crack detection method and its evaluation method based on random forest | |
| CN108491797A (en) | A kind of vehicle image precise search method based on big data | |
| CN106610969A (en) | Multimodal information-based video content auditing system and method | |
| CN110598693A (en) | Ship plate identification method based on fast-RCNN | |
| CN104199840B (en) | Intelligent place name identification technology based on statistical model | |
| CN107808375B (en) | Merge the rice disease image detecting method of a variety of context deep learning models | |
| CN107833213A (en) | A kind of Weakly supervised object detecting method based on pseudo- true value adaptive method | |
| CN103093240A (en) | Calligraphy character identifying method | |
| CN107066972B (en) | Natural scene Method for text detection based on multichannel extremal region | |
| CN104732215A (en) | Remote-sensing image coastline extracting method based on information vector machine | |
| CN101702197A (en) | Method for detecting road traffic signs | |
| CN104573685A (en) | Natural scene text detecting method based on extraction of linear structures | |
| CN106919910B (en) | Traffic sign identification method based on HOG-CTH combined features | |
| CN112528997A (en) | Tibetan-Chinese bilingual scene text detection method based on text center region amplification |
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 | ||
| WW01 | Invention patent application withdrawn after publication | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20180320 |