CN109815764A - The read method and system of machine sensible information inside image - Google Patents
The read method and system of machine sensible information inside image Download PDFInfo
- Publication number
- CN109815764A CN109815764A CN201910041344.XA CN201910041344A CN109815764A CN 109815764 A CN109815764 A CN 109815764A CN 201910041344 A CN201910041344 A CN 201910041344A CN 109815764 A CN109815764 A CN 109815764A
- Authority
- CN
- China
- Prior art keywords
- information
- image
- acquisition
- feature value
- module
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Landscapes
- Image Analysis (AREA)
- Character Discrimination (AREA)
Abstract
The read method and system of machine sensible information inside image judge to carry out Image Acquisition with the presence or absence of Chinese letter figure code to be processed in visual field according to preset time and image feature value extracts, the image feature value of extraction is compared with Standard Eigenvalue library;When there is the image feature value extracted in Standard Eigenvalue library, extract the dimensional information and orientation information of image, establish the coordinate system of reading information, mode, which is distinguished, according to information carries out image feature value parsing and to the image progress arda sampling of acquisition, lookup characteristics of image format information area and the characteristics of image format information for reading storage;Recognition obtains characteristics of image format information area's information of hum pattern code and obtains precoded form information;According to the instruction of the related prior information in characteristics of image format information area, demodulation and the acquisition of information of corresponding precoding information are carried out, and carries out corresponding error-correcting decoding and information decoding, obtains the original encoded information in image.
Description
Technical field
The present embodiments relate to technical field of image information processing, and in particular to a kind of reading of image inside machine sensible information
Take method and system.
Background technique
During the rapid development of information technology, information collection and acquisition of information problem be information technology research and development always with
Key problem in application process.Currently, one kind is to pass through image pickup there are two types of information automation acquisition major ways
The mode of the heat transfer agents such as head, Temperature Humidity Sensor, microphone acquisition, converts the signal in sensing equipment information collection region to
Data information, and be further processed in such a way that pattern recognition and classification extracts feature and be converted into information system
The valuable information that can be utilized.For example by common camera etc., the scene that license plate carries out automatic identification is clapped, actually
It is after Image Acquisition is returned system, directly handling image and carrying out OCR identification (is substantially that image segmentation and classification isotype are known
Processing is not carried out) and extracts license plate number information, and characteristic information is obtained by so-called intelligent device, carries out relevant Information procession,
Processing is all similar process.
Another kind is by Automatic Identification & Data Collections such as bar code, two dimensional code, RFID, the characteristics of this kind of technology
It is to obtain the storage in standardized data carrier by devices such as the image collecting devices such as camera, radio frequency identification interrogators
Information, to directly carry out information processing or processing.Bar code, two dimensional code, RFID are a kind of data mediums for carrying information, for
For two dimensional code, substantially two dimensional code is a kind of image code carrier, can be by coded image also by the reading method of standard
It originally was original information.
From the point of view of state-of-the-art, most widely used at present, the type collection technology of distribution most extensively, the most mature is
Image information collecting technology, the mobile phone in current industry camera, traffic camera and public hand, early have become acquisition and
Handle the most important information terminal of information.And using bar code, two dimensional code as the automatic identification technology of representative, serve the logistics in the whole world
Commerce and trade already exceed 40 years, by taking the bar code on commodity as an example, by acquisition bar code information complete commodity clearing, make an inventory and
Retrospect etc. is the prevailing model of commodity retail.
In recent years, planar bar code technology becomes the skills such as automatic identification technology and commodity sign, mobile marketing and product back-tracing
The most important hot spot technology of art and application development.Planar bar code technology, especially using Chinese letter co, QR code as the matrix type two-dimension of representative
Code, essence are a kind of special structured graphics, which has macroscopic view by the two dimensional code and other two dimensional codes to be distinguished
It the characteristic graphics frame of feature and is distributed in graphics frame, the data with specific information arrangement mode and coding definition
Sequence of modules or group are combined into.The data acquisition modes of two dimensional code are the image collecting devices such as camera, are adopted by analysis processing
Collect the image obtained, find the pattern image of two dimensional code in the picture, to detect picture position, the orientation etc. for obtaining two dimensional code
Essential information, and then out of, two dimensional code graphics frame, using information arrangement mode defined in two dimensional code standard, obtain two dimension
Information coding matrix in code.
As the extensive use of two dimensional code, especially two dimensional code are direct in the classes such as commodity, mobile marketing, advertisement and consumer
Correlation can directly acquire the fast-ripenin and large-scale application of the application model of the New Two Dimensional code technology of information by network,
In traditional two dimensional code, increase figure, color etc., to change the human eye vision effect of two dimensional code, makes two dimensional code and packaging
Or the scheme that combines together of graphic designs has begun and propagates its belief on a large scale, and has certain entangle in particular with planar bar code technology itself
Wrong function, the superposed graph LoGo in image in 2 D code, or change functional graphic shape or data module in image in 2 D code
Shape, the application for increasing two dimensional code by aesthetic feeling of the design are very common.However, these technical solutions are the problem of can't steer clear of:
The first, these schemes (are entangled with only the graphical information redundancy (functional graphic) and information redundancy of two dimensional code
It is wrong), it is still the package design member such as LoGo of the technical solution in traditional planar bar code technology frame, two dimensional code and commodity or product
Element or pictorial element are still separation, cannot be known as integrated.
The second, these existing technical solutions are all the recognition performance (software types that can be distinguished to reduce two dimensional code
Limited, software distinguishes average speed is slow, environmental requirement improves etc.) it is cost, if excessively considering graphical visual effect, no
The evitable two dimensional code that will cause is difficult to distinguish.It is because while to have beautified two dimensional code, but is actually to be folded on two dimensional code
Add the noise of figure, the recognition performance decline of these two dimensional codes is very big.
On the other hand from the point of view of, current image pattern recognition achieves development at full speed, with deep neural network
The maturation and chip of technology software and hardware, the equipment with image steganalysis function, the especially personal mobile communications such as mobile phone
Equipment has been increasingly becoming mainstream, original graphic pattern discrimination extracted by the realtime graphic identification to denounce and graphic feature and
Processing problem is gradually resolved, and also has become current development in image recognition and the intelligent image processing of mobile terminal
One of mainstream commercial trends.
However, the intelligent image identification technology of existing mobile terminal is faced with following challenge:
The first, the intelligent image identification technology of existing mobile terminal substantially belongs to the model of pattern recognition and classification technology
Farmland determines the characteristic value of identification object (actually image pattern) by modes such as machine learning and deep learnings, and provides
Most probable identification classification results.Due in mobile terminal real-time and calculate storage resource limitation, relevant classification calculate with
The extraction of characteristic value is all more limited, to mean that relevant pattern-recognition accuracy and efficiency are more limited, is easy to appear
Mistake.
The second, the intelligent image identification technology of existing mobile terminal can only extract the characteristic of division information of image,
The fingerprint feature information of image can only be obtained using deep neural network chip on the basis of training set, even if using same
Belong to the OCR technique of mode identification technology, the error rate of the text information of acquisition is still higher, therefore for needing to carry out precisely individual
The occasion of identification is still not applicable.
To sum up, a kind of reading technical solution of new image inside machine sensible information is needed.
Summary of the invention
For this purpose, the embodiment of the present invention provides the read method and system of a kind of image inside machine sensible information, solve by taking the photograph
Pattern-recognition is carried out after image procossing can only be obtained as first-class Image Acquisition sensing device, extracts the classification letter of article, image
The defect that can not precisely identify is ceased, by the judgement to known or unknown graph image or specific region feature, obtains image mark
Relevant regularly arranged data module is known, so that the information collection efficiency and the degree of reliability of information acquisition system be greatly improved.
To achieve the goals above, embodiments of the present invention provide the following technical solutions: a kind of machine-readable letter in image inside
The read method of breath, comprising:
1) judge in visual field with the presence or absence of Chinese letter figure code to be processed:
A, in visual field, Image Acquisition is carried out according to preset time, image feature value extraction is carried out to the image of acquisition, it will
The Standard Eigenvalue library stored in the image feature value of extraction and local recognition system is compared;
B, when there is the described image characteristic value extracted in the Standard Eigenvalue library, extract the dimensional information of image with
Orientation information establishes the coordinate system of reading information, carries out image feature value parsing according to information recognition mode prepared in advance;
C, mode is distinguished according to the information and arda sampling is carried out to the image of acquisition, search characteristics of image format information area
And read the characteristics of image format information of storage;
2) recognition obtains characteristics of image format information area's information of hum pattern code and obtains precoded form information;
3) according to the described image characteristic format information area related prior information instruction, carry out corresponding precoding information
Demodulation and acquisition of information, and carry out corresponding error-correcting decoding and information decoding, obtain the original encoded information in image.
As the preferred embodiment of the read method of machine sensible information inside image, in the step a, all kinds of machines can be passed through
Study, deep learning algorithm carry out image feature value extraction.
As the preferred embodiment of the read method of machine sensible information inside image, in the step b, mode is distinguished by information
The meaning of the dimensional information and orientation information is preset.
As the preferred embodiment of the read method of machine sensible information inside image, the precoded form information includes error correction etc.
Grade, information type and message length;The specific image that the Chinese letter figure code uses picture, text or Logo to identify for people, it is described
Specific image includes icon, trade mark or pattern.
As the preferred embodiment of the read method of machine sensible information inside image, when the image obtained in step b by decoding
When characteristic value and the inconsistent image feature value of the middle acquisition of step a), then the image for giving up this acquisition terminates decoding and continues back at
Step a) resurveys image.
As the preferred embodiment of the read method of machine sensible information inside image, to the mark for comparison in local recognition system
Quasi- characteristic value library is timed by network system or not timing synchronized update.
The embodiment of the present invention also provides a kind of reading system of image inside machine sensible information, comprising:
Judgment module, for judging in visual field with the presence or absence of Chinese letter figure code to be processed;
Image capture module, for carrying out Image Acquisition according to preset time in visual field;
Characteristic extracting module, for carrying out image feature value extraction to the image of acquisition;
Comparison module, for carrying out the Standard Eigenvalue library stored in the image feature value of extraction and local recognition system
It compares;
Coordinate construct module, for when in Standard Eigenvalue library exist extract image feature value when, according to extract image
Dimensional information and orientation information establish the coordinate system of reading information;
Characteristic value parsing module, for carrying out image feature value parsing according to information recognition mode prepared in advance;
Arda sampling module carries out arda sampling to the image of acquisition for distinguishing mode according to information;
Characteristics of image format information read module, for search characteristics of image format information area and read storage image it is special
Levy format information;
Precoded form information extraction modules, for distinguishing the characteristics of image format information area information of acquisition hum pattern code simultaneously
Obtain precoded form information;
Precoding information extraction module, for the instruction of the related prior information according to characteristics of image format information area, into
The demodulation of the corresponding precoding information of row and acquisition of information;
Raw information extraction module, for obtaining the original encoded information in image by error-correcting decoding and information decoding.
As the preferred embodiment of the readings system of machine sensible information inside image, pass through in characteristic extracting module machine learning,
Deep learning algorithm carries out image feature value extraction;Coordinate constructs in module and distinguishes mode to dimensional information and orientation by information
The meaning of information is preset.
It is pre- in precoded form information extraction modules as the preferred embodiment of the reading system of machine sensible information inside image
Encoding format information includes error-correction level, information type and message length.
It further include update module as the preferred embodiment of the reading system of machine sensible information inside image, for local knowledge
The Standard Eigenvalue library for comparison in read apparatus is timed by network system or not timing synchronized update.
Embodiments of the present invention have the advantages that by utilizing image to graphic pattern discrimination, orientation and judgement
Specific information realizes being perfectly combined for two dimensional code and image.At present in two dimensional code and image co-registration, the view finding of Chinese letter figure code is
By to figure identify and position and information extraction carry out, without the view finding figure of traditional two dimensional code.View finding figure
Shape is factor that is most eye-catching and influencing image display effect, and using image as two dimensional code view finding figure, judgement is the technical program
It is no there are the information position of Chinese letter figure code and Chinese letter figure code, solve the problems, such as that two dimensional code with image is merging for two figures;
Realizing the letter to image procossing and pattern-recognition in such a way that image corresponding position increases precoding information
Breath verifying and verification illustrate image if the identification classification of image is extracted the characteristic value stored in characteristic value and symbol and is not inconsistent
The content of identification be it is vicious, need to re-start Image Acquisition, it is inevitable with pattern-recognition to avoid image procossing
Misjudgement and wrong classification problem, improve the accuracy of image recognition;
Chinese letter figure code is substantially a kind of special two dimensional code, can provide more, more fine grained on image technique
Coding and mark to text information, network address, verification information etc. various information may be implemented in the information of degree, compared to
Traditional image recognition, the information content provided is more thinner, in combination with the advantage with prior information and edge calculations, Neng Gouji
The big efficiency and effect for promoting image steganalysis and obtaining information;
Image is beautiful, saves printing area, simplifies man-machine interface, and the Chinese believes that figure code is exactly graph image first, next is only
Two dimensional code, therefore the region for individually printing two dimensional code is not needed, save printing area, while image recognition and automatic identification
It is fused, it does not need as at present, image recognition needs camera, and recognition two dimensional code needs dedicated recognizer two
Equipment a, but only that camera or a recognizing apparatus.
Detailed description of the invention
It, below will be to embodiment party in order to illustrate more clearly of embodiments of the present invention or technical solution in the prior art
Formula or attached drawing needed to be used in the description of the prior art are briefly described.It should be evident that the accompanying drawings in the following description is only
It is merely exemplary, it for those of ordinary skill in the art, without creative efforts, can also basis
The attached drawing of offer, which is extended, obtains other implementation attached drawings.
Fig. 1 is the read method flow chart of machine sensible information inside a kind of image provided in an embodiment of the present invention;
Fig. 2 is the reading system schematic of machine sensible information inside a kind of image provided in an embodiment of the present invention;
Fig. 3 practices example diagram for an image co-registration information based on the technical program;
Fig. 4 is the sample figure of traditional graphical information redundancy and information redundancy that two dimensional code is utilized;
In figure: 1, judgment module;2, image capture module;3, characteristic extracting module;4, comparison module;5, coordinate constructs mould
Block;6, characteristic value parsing module;7, arda sampling module;8, characteristics of image format information read module;9, precoded form is believed
Cease extraction module;10, precoding information extraction module;11, raw information extraction module;12, update module.
Specific embodiment
Embodiments of the present invention are illustrated by particular specific embodiment below, those skilled in the art can be by this explanation
Content disclosed by book is understood other advantages and efficacy of the present invention easily, it is clear that described embodiment is the present invention one
Section Example, instead of all the embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art are not doing
Every other embodiment obtained under the premise of creative work out, shall fall within the protection scope of the present invention.
Referring to Fig. 1, the present embodiment provides a kind of read methods of machine sensible information inside image, comprising the following steps:
S1: judge in visual field with the presence or absence of Chinese letter figure code to be processed:
S101: in visual field, Image Acquisition is carried out according to preset time, image feature value is carried out to the image of acquisition and is mentioned
It takes, the Standard Eigenvalue library stored in the image feature value of extraction and local recognition system is compared;
S102: when there is the described image characteristic value extracted in the Standard Eigenvalue library, the scale letter of image is extracted
Breath and orientation information, establish the coordinate system of reading information, carry out image feature value according to information recognition mode prepared in advance
Parsing;
S103: mode is distinguished according to the information, arda sampling is carried out to the image of acquisition, search characteristics of image format letter
Breath area and the characteristics of image format information for reading storage;
S2: recognition obtains characteristics of image format information area's information of hum pattern code and obtains precoded form information;
S3: according to the instruction of the related prior information of the described image characteristic format information area, corresponding precoding letter is carried out
The demodulation of breath and acquisition of information, and corresponding error-correcting decoding and information decoding are carried out, obtain the original encoded information in image.
Inside image in one embodiment of the read method of machine sensible information, in the step a, pass through machine learning, depth
It spends learning algorithm and carries out image feature value extraction.Deep learning is that more abstract high-rise expression is formed by combination low-level feature
Attribute classification or feature, to find that the distributed nature of data indicates, concept the grinding derived from artificial neural network of deep learning
Study carefully, the multilayer perceptron containing more hidden layers is exactly a kind of deep learning structure.
The present embodiment is not intended to limit specific deep learning algorithm types.Supervised learning algorithm, supervised can be used
Learning algorithm is made of a target variable or outcome variable (or dependent variable).These variables are by a series of known predictors
(independent variable) prediction.Using this number of variables, the function that input value is mapped to desired output can be generated.
This training process can be continued for, until model obtains desired accuracy on the training data.The example of supervised study
Have: recurrence, decision tree, random forest, K- nearest neighbor algorithm, logistic regression etc..
In non-supervisory formula learning algorithm, to predict or estimate without any target variable or outcome variable.Non-supervisory formula
Practising secondary can use association algorithm and K- mean algorithm.
Nitrification enhancement training machine carries out decision.Machine, which is placed in one, can allow it to train by trial and error repeatedly certainly
In oneself environment.Machine is learnt from past experience, and attempts to make accurately using the most thorough knowledge of understanding
Business judges.The example of intensified learning has Markovian decision process.
Inside image in one embodiment of the read method of machine sensible information, in the step b, mode is distinguished by information
The meaning of the dimensional information and orientation information is preset.The dimensional information and orientation information for extracting image, are to build
The coordinate system of vertical reading information, so that the information recognition in Chinese letter figure code is returned, specific characteristic value value and its row
Column and acquisition modes are can to carry out predetermined, and the information of specific symbol distinguishes mode, i.e., how relevant information is compiled
Make can be specified by technical standard and application standard in figure.
Inside image in one embodiment of the read method of machine sensible information, the precoded form information includes error correction etc.
Grade, information type and message length.Specifically, obtaining the characteristics of image format information area information of hum pattern code by recognition, obtain
Take relevant information precoded form information, such as error-correction level, information type, message length relevant information, information precoding
Format information can be reduced to for decoding significant format information.
Specifically, the specific image that the Chinese letter figure code uses picture, text or Logo to identify for people, the specific image
Including icon, trade mark or pattern.The Chinese believes that figure code is equipped with the information area, according to feature of image and the information generated according to relevant criterion
Matrix, the matrix are that continuously or discontinuously, information-bearing units are nominal regular figure, and such as name is round, rectangular, triangle
The mutual relationship of shape, heart etc., relevant information-bearing units is relatively fixed, for the arrangement mode of a certain type, has
Determining arrangement order is carried out from different meanings, relevant definition by different technical solutions or different technologies and application standard
Definition has opposite independence.
The information content of the information area is unrelated to image itself and relevant in application in addition to machine-readable including needing to carry out
It can also include the feature decision information of image except precoding information, such as specific image type indexes, image appearance rope
Draw, image texture index etc. information relevant to image, thus distinguish the Chinese letter figure code when, provide to image steganalysis
Result verification, to avoid the occurrence of error code as far as possible.
The information content of the information area further includes according to image information and input letter other than image information and input information
Breath, the redundancy generated using specific error correction algorithm, the algorithm of use include being not limited to Reed Solomon algorithm, BCH
The nonblock codes such as the linear block codes such as algorithm and LDPC code.
Inside image in one embodiment of the read method of machine sensible information, when the image obtained in step b by decoding
When characteristic value and the inconsistent image feature value of the middle acquisition of step a), then the image for giving up this acquisition terminates decoding and continues back at
Step a) resurveys image.Due to by that can be corrected to the image feature value of acquisition, if the figure after error correction calculations
Characteristic value after extracting as characteristic value with the global feature by image is inconsistent, then it is wrong to be likely to occur decoding for explanation
Accidentally, it needs to discard this recognition process and resurveys image.
Inside image in one embodiment of the read method of machine sensible information, to the mark for comparison in local recognition system
Quasi- characteristic value library is timed by network system or not timing synchronized update.The Standard Eigenvalue library of comparison can be in recognition
It is stored, can also be updated by network system in system, characteristic value is calculated after Image Acquisition and is compared, pair of comparison
It is stored in symbol as can be, is also possible to be updated in system.
Referring to fig. 2, the embodiment of the present invention also provides a kind of reading system of image inside machine sensible information, comprising:
Judgment module 1, for judging in visual field with the presence or absence of Chinese letter figure code to be processed;
Image capture module 2, for carrying out Image Acquisition according to preset time in visual field;
Characteristic extracting module 3, for carrying out image feature value extraction to the image of acquisition;
Comparison module 4, Standard Eigenvalue library for that will store in the image feature value of extraction and local recognition system into
Row compares;
Coordinate construct module 5, for when in Standard Eigenvalue library exist extract image feature value when, according to extract image
Dimensional information and orientation information establish the coordinate system of reading information;
Characteristic value parsing module 6, for carrying out image feature value parsing according to information recognition mode prepared in advance;
Arda sampling module 7 carries out arda sampling to the image of acquisition for distinguishing mode according to information;
Characteristics of image format information read module 8, for searching characteristics of image format information area and reading the image of storage
Characteristic format information;
Precoded form information extraction modules 9, for distinguishing the characteristics of image format information area information for obtaining hum pattern code
And obtain precoded form information;
Precoding information extraction module 10, for the instruction of the related prior information according to characteristics of image format information area,
Carry out demodulation and the acquisition of information of corresponding precoding information;
Raw information extraction module 11, for obtaining the original coding letter in image by error-correcting decoding and information decoding
Breath.
Pass through inside image in one embodiment of the readings system of machine sensible information, in characteristic extracting module machine learning,
Deep learning algorithm carries out image feature value extraction;Coordinate constructs in module and distinguishes mode to dimensional information and orientation by information
The meaning of information is preset.
It is pre- in precoded form information extraction modules inside image in one embodiment of the reading system of machine sensible information
Encoding format information includes error-correction level, information type and message length.
It further include update module 12 inside image in one embodiment of the reading system of machine sensible information, for local
The Standard Eigenvalue library for comparison in recognition system is timed by network system or not timing synchronized update.
Specifically, conventional solution, can only acquire image, pattern-recognition is carried out, pumpkin and west can be at most told
The difference of melon, and when being identified due to the design of correlation classifier, it is inevitably easy to appear mistake and (imagines lower green peel
Pumpkin).In addition, any pattern recognition system can only pick out such as 330 milliliters of mineral water, but filled by image recognition
It sets, it is impossible to which picking out this is the 330 milliliters of mineral waters produced the year before or the mineral water just produced, in shape packaging not
It may be respectively.For example judge that whether having the recognizer of pumpkin and watermelon can be in visual field judges acquisition system acquisition figure
Color in shape is yellow or cyan, and yellow is exactly pumpkin, and cyan is exactly watermelon, if not considering shape, characteristics of image
Value can set yellow as 1, cyan 0, if adding appearance problem, so that it may which, along with ellipse is 1, circle is 0, then southern
Melon characteristic value is 11, watermelon 00.At this point, if in melon system on sale, so that it may 2 binary systems be used to whether there is as image
Characteristic value.In this process, it is determined that whether there may be the Chinese to believe figure code in visual field.
Please referring to Fig. 3 on the basis of comparison diagram 4, according to an embodiment of the present invention in technical solution a practice process,
Characteristics of image format information area is the region of 5 modules signal in the upper right corner, wherein two blocks of front are for verifying whether
Whether, there are the characteristic value of symbol, subsequent part storage is relevant message encoding format information, such as module arrangement, have
Error correction etc., if legend is 3 blocks.Relevant message length and the meaning of each representative can be carried out not by different standards
Same definition.And error-correcting decoding (if there is) and letter are carried out to the information area according to the definition of Its Relevant Technology Standards or application standard
Breath decoding.The detailed process of error-correcting decoding and information decoding can be realized using the general-purpose algorithm of the prior art.
By relevant last 3, obtain in the symbol that there are 38 data module (circular modes in figure
Block), and relevant information storage means is in the red of mark and blue region (due to that can not illustrate specific using artwork master
Color) along mark edge arranged counterclockwise, range marker is outer peripheral to be equidistant, and the storage of preceding 20 modules is basic
Information, rear 18 modules are protections code system (such as Bose-Chaudhuri Hocquenghem error correction codes), allow 9 module errors occur in symbol at this time.For this purpose,
Polar coordinate system can be established centered on identifying center of gravity, according to the relationship between relevant normal place and symbol module, established
Grid is sampled, relevant module status is fetched to come.Hereafter, error correction information decoding is carried out, in case of module status mistake,
It can be restored by error correcting code.Later, information decoding is carried out, original encoded information is restored.It will by the technical program
Image judges whether there is the information position of Chinese letter figure code and Chinese letter figure code, solves two dimensional code as two dimensional code view finding figure
It is that two figures merge problem with image, avoids image procossing and inevitably misjudge with pattern-recognition and asked with mistake classification
Topic improves the accuracy of image recognition, is capable of the efficiency and effect of significant increase image steganalysis acquisition information, image beauty
It sees, saves printing area, simplify man-machine interface.
Although above having used general explanation and specific embodiment, the present invention is described in detail, at this
On the basis of invention, it can be made some modifications or improvements, this will be apparent to those skilled in the art.Therefore,
These modifications or improvements without departing from theon the basis of the spirit of the present invention are fallen within the scope of the claimed invention.
Claims (10)
1. the read method of machine sensible information inside image, which comprises the steps of:
1) judge in visual field with the presence or absence of Chinese letter figure code to be processed:
A, in visual field, Image Acquisition is carried out according to preset time, image feature value extraction is carried out to the image of acquisition, will be extracted
Image feature value and local recognition system in the Standard Eigenvalue library that stores be compared;
B, when there is the described image characteristic value extracted in the Standard Eigenvalue library, the dimensional information and orientation of image are extracted
Information establishes the coordinate system of reading information, carries out image feature value parsing according to information recognition mode prepared in advance;
C, mode is distinguished according to the information and arda sampling is carried out to the image of acquisition, searched characteristics of image format information area and read
Take the characteristics of image format information of storage;
2) recognition obtains characteristics of image format information area's information of hum pattern code and obtains precoded form information;
3) according to the described image characteristic format information area related prior information instruction, carry out the solution of corresponding precoding information
Tune and acquisition of information, and corresponding error-correcting decoding and information decoding are carried out, obtain the original encoded information in image.
2. the read method of machine sensible information inside image according to claim 1, which is characterized in that in the step a, lead to
Cross machine learning, deep learning algorithm carries out image feature value extraction.
3. the read method of machine sensible information inside image according to claim 1, which is characterized in that in the step b, lead to
Information recognition mode is crossed to preset the meaning of the dimensional information and orientation information.
4. the read method of machine sensible information inside image according to claim 1, which is characterized in that the precoded form
Information includes error-correction level, information type and message length;What the Chinese letter figure code was identified using picture, text or Logo for people
Specific image, the specific image include icon, trade mark or pattern.
5. the read method of machine sensible information inside image according to claim 1, which is characterized in that pass through when in step b
When the image feature value of the image feature value and the middle acquisition of step a) that decode acquisition is inconsistent, then give up the image knot of this acquisition
Beam decoding continues back at step a) and resurveys image.
6. the read method of machine sensible information inside image according to claim 1, which is characterized in that local recognition system
In the Standard Eigenvalue library for comparison be timed by network system or not timing synchronized update.
7. the reading system of machine sensible information inside image characterized by comprising
Judgment module, for judging in visual field with the presence or absence of Chinese letter figure code to be processed;
Image capture module, for carrying out Image Acquisition according to preset time in visual field;
Characteristic extracting module, for carrying out image feature value extraction to the image of acquisition;
Comparison module, for comparing the Standard Eigenvalue library stored in the image feature value of extraction and local recognition system
It is right;
Coordinate construct module, for when in Standard Eigenvalue library exist extract image feature value when, according to extract image ruler
Degree information and orientation information establish the coordinate system of reading information;
Characteristic value parsing module, for carrying out image feature value parsing according to information recognition mode prepared in advance;
Arda sampling module carries out arda sampling to the image of acquisition for distinguishing mode according to information;
Characteristics of image format information read module, for searching characteristics of image format information area and reading the characteristics of image lattice of storage
Formula information;
Precoded form information extraction modules, for distinguishing characteristics of image format information area's information of acquisition hum pattern code and obtaining
Precoded form information;
Precoding information extraction module carries out phase for the instruction of the related prior information according to characteristics of image format information area
The demodulation for the precoding information answered and acquisition of information;
Raw information extraction module, for obtaining the original encoded information in image by error-correcting decoding and information decoding.
8. the reading system of machine sensible information inside image according to claim 7, which is characterized in that in characteristic extracting module
Image feature value extraction is carried out by machine learning, deep learning algorithm;Coordinate constructs in module and distinguishes mode pair by information
The meaning of dimensional information and orientation information is preset.
9. the reading system of machine sensible information inside image according to claim 7, which is characterized in that precoded form information
Precoded form information in extraction module includes error-correction level, information type and message length.
10. the reading system of machine sensible information inside image according to claim 7, which is characterized in that further include updating mould
Block, for being timed to the Standard Eigenvalue library for comparison in local recognition system by network system or not timing is synchronous
It updates.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910041344.XA CN109815764B (en) | 2019-01-16 | 2019-01-16 | Method and system for reading machine-readable information in image |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910041344.XA CN109815764B (en) | 2019-01-16 | 2019-01-16 | Method and system for reading machine-readable information in image |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109815764A true CN109815764A (en) | 2019-05-28 |
CN109815764B CN109815764B (en) | 2022-04-08 |
Family
ID=66604506
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910041344.XA Active CN109815764B (en) | 2019-01-16 | 2019-01-16 | Method and system for reading machine-readable information in image |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109815764B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2022211064A1 (en) * | 2021-03-31 | 2022-10-06 | 株式会社デンソーウェーブ | Information reading device |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101526997A (en) * | 2009-04-22 | 2009-09-09 | 无锡名鹰科技发展有限公司 | Embedded infrared face image identifying method and identifying device |
US20140217178A1 (en) * | 2013-02-06 | 2014-08-07 | Alibaba Group Holding Limited | Information processing method and system |
CN104992481A (en) * | 2015-07-13 | 2015-10-21 | 广东电网有限责任公司惠州供电局 | Metered inspection system and method based on image recognition technology |
CN109871723A (en) * | 2019-01-16 | 2019-06-11 | 王诗会 | The processing method and system of image co-registration information |
-
2019
- 2019-01-16 CN CN201910041344.XA patent/CN109815764B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101526997A (en) * | 2009-04-22 | 2009-09-09 | 无锡名鹰科技发展有限公司 | Embedded infrared face image identifying method and identifying device |
US20140217178A1 (en) * | 2013-02-06 | 2014-08-07 | Alibaba Group Holding Limited | Information processing method and system |
CN104992481A (en) * | 2015-07-13 | 2015-10-21 | 广东电网有限责任公司惠州供电局 | Metered inspection system and method based on image recognition technology |
CN109871723A (en) * | 2019-01-16 | 2019-06-11 | 王诗会 | The processing method and system of image co-registration information |
Non-Patent Citations (1)
Title |
---|
丁永军等: "一种新的二维码识读工具的构建及应用方案", 《包装工程》 * |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2022211064A1 (en) * | 2021-03-31 | 2022-10-06 | 株式会社デンソーウェーブ | Information reading device |
Also Published As
Publication number | Publication date |
---|---|
CN109815764B (en) | 2022-04-08 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Yuliang et al. | Detecting curve text in the wild: New dataset and new solution | |
CN107609497B (en) | Real-time video face recognition method and system based on visual tracking technology | |
CN102136058B (en) | Bar code image identification method | |
Lin et al. | Multiple instance ffeature for robust part-based object detection | |
US20210064871A1 (en) | Apparatus and method for recognition of text information | |
CN110110335A (en) | A kind of name entity recognition method based on Overlay model | |
US20110063468A1 (en) | Method and apparatus for retrieving label | |
CN107526967A (en) | A kind of risk Address Recognition method, apparatus and electronic equipment | |
Park et al. | Study on fashion image retrieval methods for efficient fashion visual search | |
CN100414562C (en) | Method for positioning feature points of human face in human face recognition system | |
CN105590097A (en) | Security system and method for recognizing face in real time with cooperation of double cameras on dark condition | |
CN111582033A (en) | Garbage classification identification method and system and computer readable storage medium | |
CN110647956A (en) | Invoice information extraction method combined with two-dimensional code recognition | |
CN107609611A (en) | A kind of multiclass formation gathering method based on large-scale radio-frequency identification system | |
CN109101561B (en) | Wine label identification method | |
CN109086772A (en) | A kind of recognition methods and system distorting adhesion character picture validation code | |
CN109255414A (en) | A kind of colour barcode made an inventory for books, books recognition methods, electronic equipment and storage medium | |
CN112926379A (en) | Method and device for constructing face recognition model | |
CN117690031B (en) | SAM model-based small sample learning remote sensing image detection method | |
CN109815764A (en) | The read method and system of machine sensible information inside image | |
CN109871723A (en) | The processing method and system of image co-registration information | |
WO2021169207A1 (en) | Object identification method and apparatus based on machine learning | |
CN109255290A (en) | Menu recognition methods, device, electronic equipment and storage medium | |
CN111159456B (en) | Multi-scale clothing retrieval method and system based on deep learning and traditional features | |
Murtaza et al. | A novel approach of boundary preservative apparel detection and classification of fashion images using deep learning |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |