CN108230308A - A kind of handbill proofreading method, device and equipment - Google Patents
A kind of handbill proofreading method, device and equipment Download PDFInfo
- Publication number
- CN108230308A CN108230308A CN201711483666.7A CN201711483666A CN108230308A CN 108230308 A CN108230308 A CN 108230308A CN 201711483666 A CN201711483666 A CN 201711483666A CN 108230308 A CN108230308 A CN 108230308A
- Authority
- CN
- China
- Prior art keywords
- information
- block
- image
- commodity
- image block
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
- G06T7/33—Determination of transform parameters for the alignment of images, i.e. image registration using feature-based methods
- G06T7/344—Determination of transform parameters for the alignment of images, i.e. image registration using feature-based methods involving models
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10024—Color image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30168—Image quality inspection
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention provides a kind of handbill proofreading method, device and equipment, the method includes：Advertisement single image is obtained, and is multiple commodity image blocks by the advertisement single image cutting；Each commodity image block is identified, obtains each commodity image merchandise classification information in the block and corresponding descriptive labelling information；According to making demand information corresponding with the advertisement single image, the corresponding goal description information of each merchandise classification information is obtained；Descriptive labelling information is proofreaded according to each merchandise classification information corresponding goal description information.The present invention extracts ad content, and ad content is proofreaded according to making demand, instead of artificial check and correction, improves correction efficiency by carrying out image cutting and content recognition to handbill.
Description
Technical field
The present invention relates to technical field of image processing, and in particular to a kind of handbill proofreading method, device and equipment.
Background technology
The check and correction of handbill content is mainly using the method for artificial contrast's check and correction at present；Especially by the need of comparison ad.'s list
The making statement of requirements that the side of asking provides first prints handbill specimen page by the producer of handbill, carries out self check and correction；Then will
Specimen page submits to the full-time press corrector of handbill making side, carries out third party's check and correction；Specimen page is finally transferred to the need of handbill
It is finally proofreaded the side of asking.The check and correction process usually carries out 2-3 wheels, after completing check and correction, handbill is issued printing house and is printed
System.
During above check and correction, press corrector is marked the content in specimen page using e-Pointer and punctuates；For difference
Mistake, punctuated using the e-Pointer of different colours；After completing check and correction, it is modified by producer.Due to using people
Work is proofreaded, therefore in check and correction process or makeover process, is easy to cause check and correction or is corrected inaccurate；For example, holiday
Need check and correction either modified content check and correction or modified partial transposition etc.；Also, entire check and correction process need to grow very much when
Between, and easily error.
Invention content
In view of the above defects of the prior art, the present invention provides a kind of handbill proofreading method, device and equipment.
An aspect of of the present present invention provides a kind of handbill proofreading method, including：Advertisement single image is obtained, and by the advertisement
Single image cutting is multiple commodity image blocks；Each commodity image block is identified, obtains each commodity image quotient in the block
Product classification information and corresponding descriptive labelling information；According to making demand information corresponding with the advertisement single image, obtain every
The corresponding goal description information of a merchandise classification information；According to the corresponding goal description information of each merchandise classification information to commodity
Description information is proofreaded.
Wherein, it is described that each commodity image block is identified, obtain each commodity image merchandise classification information in the block
And the step of corresponding descriptive labelling information, specifically includes：Each commodity image block is identified, obtains the commodity image
The corresponding merchandise classification information of block；And each commodity image description information image block in the block of extraction, and identify the description
Frame descriptive labelling information in the block.
Wherein, the advertisement single image cutting is specifically included for the step of multiple commodity image blocks：By the handbill
The image segmentation model that image input pre-establishes obtains multiple commodity image blocks of output；Correspondingly, by the advertisement free hand drawing
As being further included before the step of image segmentation model that input pre-establishes, according to merchandise classification information, descriptive labelling information and wide
Slide former draws multiple advertisement single images at random, and the position rectangle for recording each commodity image block in each advertisement single image is sat
Mark；The input value and output valve of advertisement single image and the position rectangular coordinates as the model training that need cutting are subjected to mould
Type training obtains the image segmentation model that training is completed.
Wherein, it is described that each commodity image block is identified, obtain the corresponding merchandise classification letter of the commodity image block
The step of breath, specifically includes：The merchandise classification identification model that the input of each commodity image block is pre-established, obtains the every of output
The corresponding merchandise classification information of a commodity image block；Correspondingly, the commodity that the input of each commodity image block is pre-established
It was further included before the step of classification identification model, multiple commodity image blocks is generated according to merchandise classification information, descriptive labelling information；And
Commodity image block is pre-processed, wherein, the pretreatment includes overturning, color replaces, distorts and add in noise
At least one；Using pretreated commodity image block and corresponding merchandise classification information as the input value of model training
Model training is carried out with output valve, obtains the merchandise classification identification model that training is completed.
Wherein, the step of each commodity image of extraction description information image block in the block specifically includes：By each quotient
Product image block inputs the description information image block extraction model pre-established respectively, and each commodity image for obtaining output is in the block
Description information image block；Correspondingly, the description information image block pre-established that each commodity image block is inputted respectively carries
It was further included before the step of modulus type, multiple commodity image blocks is generated according to merchandise classification information, descriptive labelling information, wherein, institute
It states commodity image block and includes description information image block, the description information image block is by descriptive labelling information and randomly selected word
Body generates；And the position rectangular coordinates of record description frame block；Commodity image block is pre-processed, wherein, it is described pre-
Processing includes overturning, at least one of noise is replaced, distorts and added to color；By pretreated commodity image block and
The position rectangular coordinates of cutting is needed to carry out model training respectively as the input value and output valve of model training, acquisition has been trained
Into description information image block extraction model.
Wherein, the step of identification description information image descriptive labelling information in the block specifically includes：It will description
In the description information identification model that the input of frame block pre-establishes, the descriptive labelling information of output is obtained；Correspondingly, it is described
It further includes before step in the description information identification model that the input of description information image block is pre-established, is believed according to descriptive labelling
Breath and font random combine generate multiple description information image blocks；Description information image block is pre-processed, wherein, it is described pre-
Processing includes overturning, at least one of noise is replaced, distorts and added to color；By pretreated descriptive labelling information
Image block and descriptive labelling information carry out model training respectively as the input value and output valve of model training, obtain training and complete
Description information identification model.
A kind of handbill verifying unit of another aspect of the present invention, including：Cutting module, for obtaining advertisement single image,
And by the advertisement single image cutting be multiple commodity image blocks；Identification module, for each commodity image block to be identified,
Obtain each commodity image merchandise classification information in the block and corresponding descriptive labelling information；Acquisition module, for basis and institute
The corresponding making demand information of advertisement single image is stated, obtains the corresponding goal description information of each merchandise classification information；Proofread mould
Block, for being proofreaded according to the corresponding goal description information of each merchandise classification information to descriptive labelling information.
Wherein, the identification module specifically includes：Classification recognition unit, for each commodity image block to be identified,
Obtain the corresponding merchandise classification information of the commodity image block；Information extraction unit, it is in the block for extracting each commodity image
Description information image block, and identify description information image descriptive labelling information in the block.
Another aspect of the present invention provides a kind of handbill check and correction equipment, including：At least one processor；And with it is described
At least one processor of processor communication connection, wherein：The memory is stored with the program that can be performed by the processor
Instruction, the processor call described program instruction to be able to carry out the handbill proofreading method that the above-mentioned aspect of the present invention provides, example
Such as include：Advertisement single image is obtained, and is multiple commodity image blocks by the advertisement single image cutting；To each commodity image block
It is identified, obtains each commodity image merchandise classification information in the block and corresponding descriptive labelling information；According to it is described wide
The corresponding making demand information of single image is accused, obtains the corresponding goal description information of each merchandise classification information；According to each quotient
The corresponding goal description information of product classification information proofreads descriptive labelling information.
Another aspect of the present invention provides a kind of non-transient computer readable storage medium storing program for executing, and the non-transient computer is readable
Storage medium stores computer instruction, and the computer instruction makes the computer perform the advertisement that the above-mentioned aspect of the present invention provides
Single proofreading method, such as including：Advertisement single image is obtained, and is multiple commodity image blocks by the advertisement single image cutting；It is right
Each commodity image block is identified, and obtains each commodity image merchandise classification information in the block and corresponding descriptive labelling letter
Breath；According to making demand information corresponding with the advertisement single image, the corresponding goal description of each merchandise classification information is obtained
Information；Descriptive labelling information is proofreaded according to each merchandise classification information corresponding goal description information.
Handbill proofreading method provided by the invention, device and equipment, by carrying out image cutting and content to handbill
Identification, extracts ad content, and ad content is proofreaded according to making demand, instead of artificial check and correction, improves check and correction
Efficiency.
Description of the drawings
In order to illustrate more clearly about the embodiment of the present invention or technical scheme of the prior art, to embodiment or will show below
There is attached drawing needed in technology description to be briefly described, it should be apparent that, the accompanying drawings in the following description is this hair
Some bright embodiments, for those of ordinary skill in the art, without creative efforts, can be with root
Other attached drawings are obtained according to these attached drawings.
Fig. 1 is the flow diagram of handbill proofreading method provided in an embodiment of the present invention；
Fig. 2 is the flow diagram of handbill proofreading method that another embodiment of the present invention provides；
Fig. 3 is the structure diagram of handbill verifying unit provided in an embodiment of the present invention；
Fig. 4 is the structure diagram that handbill provided in an embodiment of the present invention proofreads equipment.
Specific embodiment
Purpose, technical scheme and advantage to make the embodiment of the present invention are clearer, below in conjunction with the embodiment of the present invention
In attached drawing, the technical solution in the embodiment of the present invention is explicitly described, it is clear that described embodiment be the present invention
Part of the embodiment, instead of all the embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art are not having
All other embodiments obtained under the premise of creative work are made, shall fall within the protection scope of the present invention.
Fig. 1 is the flow diagram of handbill proofreading method provided in an embodiment of the present invention, as shown in Figure 1, including：Step
101, advertisement single image is obtained, and be multiple commodity image blocks by the advertisement single image cutting；Step 102, to each commodity
Image block is identified, and obtains each commodity image merchandise classification information in the block and corresponding descriptive labelling information；Step
103, according to making demand information corresponding with the advertisement single image, obtain the corresponding goal description of each merchandise classification information
Information；Step 104, descriptive labelling information is proofreaded according to each merchandise classification information corresponding goal description information.
In a step 101, advertisement single image is made of multiple commodity image blocks, each commodity image block be include to
A kind of few commodity and its descriptive labelling information；Image merchandise classification information in the block is usually the image of commodity, and descriptive labelling
Information is for describing the commodity, such as the text informations such as price or tag line；Get proofreaded it is wide
After accusing single image, need to carry out cutting processing to advertisement single image.
In a step 102, according to the commodity image block obtained in step 101, in order to be carried out to the content of each Commdity advertisement
Check and correction, thus on the one hand need to identify the merchandise classification information corresponding to each commodity image block, that is, be specifically which kind of commodity, quotient
Product classification information can be specifically the bar code or other identification codes of commodity；On the other hand it needs to extract each image quotient in the block
Product description information.
In step 103, the goal description got corresponding to each merchandise classification information by making demand information is believed
Breath；Goal description information is the standard of check and correction, is the commodity judged according to goal description information on handbill during check and correction
Whether description information is wrong；It should be noted that step 103 is not limited to perform after step 101 and step 102, step 103
It only needs to perform before following step 104.
At step 104, using the goal description information corresponding to each merchandise classification information obtained in step 103 as
Check and correction standard, whether the descriptive labelling information obtained in judgment step 102 is wrong, and correct and wrong part is opened up
Show.
Handbill proofreading method provided in an embodiment of the present invention, by carrying out image cutting and content recognition to handbill,
Ad content is extracted, and ad content is proofreaded according to making demand, instead of artificial check and correction, improves correction efficiency.
It is described that each commodity image block is identified on the basis of any of the above-described embodiment, obtain each commodity figure
As the step of merchandise classification information in the block and corresponding descriptive labelling information specifically includes：Each commodity image block is known
Not, the corresponding merchandise classification information of the commodity image block is obtained；And each commodity image description information figure in the block of extraction
As block, and identify description information image descriptive labelling information in the block.
Specifically, for merchandise classification information, can commodity figure wherein included directly be got according to commodity image block
Merchandise classification information corresponding to piece；And for the extraction of descriptive labelling information, since descriptive labelling information is the shape with image
Formula exists, and therefore, extracts frame block described in commodity image block (word for wherein including description information) first, so
Description information image word in the block is identified afterwards, so as to get descriptive labelling information.
On the basis of any of the above-described embodiment, the advertisement single image cutting is had for the step of multiple commodity image blocks
Body includes：The image segmentation model that advertisement single image input is pre-established obtains multiple commodity image blocks of output；Phase
It further included before the step of Ying Di, the image segmentation model that advertisement single image input is pre-established, is believed according to merchandise classification
Breath, descriptive labelling information and advertisement formwork, draw multiple advertisement single images, and record each commodity in each advertisement single image at random
The position rectangular coordinates of image block；Using advertisement single image and the position rectangular coordinates of cutting are needed as the defeated of model training
Enter value and output valve carries out model training, obtain the image segmentation model that training is completed.
Specifically, in order to carry out cutting to image, neural fusion may be used；Such as Faster R- may be used
CNN models, training process specifically include following steps：
Step 1, all commodity picture (commodity picture and merchandise classification information pair for being possibly used for preparing advertising are collected
Should)；
Step 2, all templates for being possibly used for preparing advertising are collected, which includes multiple empty images to be filled
Block；
Step 3, by programming development, at random by the information such as commodity picture and commodity price (i.e. descriptive labelling information),
It is plotted in the advertisement formwork randomly selected, and the position rectangular coordinates of drafting is preserved；
Step 4, the position of the commodity image block of cutting will be needed in advertisement single image and image that programming code is drawn
Put rectangular coordinates (image block can be rectangle, coordinate when which is cutting), respectively as the x data of training pattern
(input value) and y data (output valve), input to model and are trained；
Step 5, after the completion of training, preservation model weight parameter file；
Step 6, during cutting advertisement single image, initialization model simultaneously loads trained weight parameter, to advertisement single image
Cutting is carried out, extracts each commodity image block.
It is described that each commodity image block is identified on the basis of any of the above-described embodiment, obtain the commodity figure
The step of merchandise classification information corresponding as block, specifically includes：The merchandise classification that the input of each commodity image block pre-establishes is known
Other model obtains the corresponding merchandise classification information of each commodity image block of output；Correspondingly, it is described by each commodity image block
It further includes before the step of inputting the merchandise classification identification model pre-established, is given birth to according to merchandise classification information, descriptive labelling information
Into multiple commodity image blocks；And commodity image block is pre-processed, wherein, the pretreatment includes overturning, color is replaced, becomes
Shape distorts and adds at least one of noise；Pretreated commodity image block and corresponding merchandise classification information are made respectively
Input value and output valve for model training carry out model training, obtain the merchandise classification identification model that training is completed.
Specifically, in order to which the classification of commodity in the block to image is identified, neural fusion can be utilized；Such as it adopts
With Inception Resnet V2 models, training process specifically includes following steps：
Step 1, collecting all commodity pictures for being possibly used for preparing advertising, (each commodity picture corresponds to a commodity class
Other information)；
Step 2, by programming development, commodity picture is randomly selected, and commodity picture is overturn, color is replaced, and is become
Shape distorts, the processing such as addition noise；
Step 3, programming code treated image and commodity are corresponded into unique bar code (i.e. merchandise classification information), point
Not as the x data (input value) of training pattern and y data (output valve), input to model and be trained；
Step 4, after the completion of training, preservation model weight parameter file；
Step 5, when the commodity image block after cutting being identified, first initialization model simultaneously loads trained weight ginseng
Number, is identified the merchandise classification information of particular commodity image block.
On the basis of any of the above-described embodiment, the step of each commodity image description information image block in the block of extraction
Suddenly it specifically includes：Each commodity image block is inputted to the description information image block extraction model pre-established respectively, obtains output
Each commodity image description information image block in the block；Correspondingly, described input each commodity image block respectively is built in advance
It was further included before the step of vertical description information image block extraction model, it is more according to merchandise classification information, the generation of descriptive labelling information
A commodity image block, wherein, the commodity image block includes description information image block, and the description information image block is retouched by commodity
State information and the generation of randomly selected font；And the position rectangular coordinates of record description frame block；To commodity image block into
Row pretreatment, wherein, the pretreatment includes overturning, at least one of noise is replaced, distorts and added to color；It will be pre-
Treated commodity image block and need the position rectangular coordinates of cutting respectively as model training input value and output valve into
Row model training obtains the description information image block extraction model that training is completed.
Specifically, description information image block can be extracted by neural network；For example, by using Textboxes
Model models, the training of model specifically include following steps：
Step 1, all commodity pictures for being possibly used for preparing advertising are collected；
Step 2, collect it is all be possibly used for preparing advertising described in font；
Step 3, by programming development, commodity picture is randomly selected, and commodity picture is overturn, color is replaced, and is become
Shape distorts, the processing such as addition noise；
Step 4, by programming development, to treated, the addition of commodity image block uses what the font randomly selected was drawn
Character description information, and record all word uniline position rectangular coordinates drawn；
Step 5, by the position rectangular coordinates of programming code treated image and description information image block, respectively as
The x data (input value) of training pattern and y data (output valve), input to model and are trained；
Step 6, after the completion of training, preservation model weight parameter file；
Step 7, when carrying out Word Input to the commodity image block after cutting, first initialization model simultaneously loads trained power
Weight parameter, extracts image description information image block in the block.
On the basis of any of the above-described embodiment, the identification description information image descriptive labelling information in the block
Step specifically includes：In the description information identification model that the input of description information image block is pre-established, the commodity of output are obtained
Description information；Correspondingly, before the step in the description information identification model that the input of description information image block is pre-established
It further includes：Multiple description information image blocks are generated according to descriptive labelling information and font random combine；To description information image block
It is pre-processed, wherein, the pretreatment includes overturning, at least one of noise is replaced, distorts and added to color；It will
Pretreated descriptive labelling frame block and descriptive labelling information respectively as model training input value and output valve into
Row model training obtains the description information identification model that training is completed.
Specifically, identify that description information can pass through nerve net from the description information image block for including description information
Network is realized；Such as CRNN models may be used, the training of model includes the following steps：
Step 1, the font of all middle descriptive labelling information that are possibly used for preparing advertising is collected；
Step 2, collect it is all be possibly used for preparing advertising described in text information (i.e. descriptive labelling information)；
Step 3, by programming development, font and word description are randomly selected, generates the word description picture of uniline,
And the word description picture of generation is overturn, color is replaced, and distorts and add the processing such as noise；
Step 4, by programming code treated image (i.e. description information image block) and descriptive labelling information, respectively
As the x data of training pattern and y data, input to model and be trained；
Step 5, after the completion of training, preservation model weight parameter file；
Step 6, when description information image block being identified, first initialization model simultaneously loads trained weight parameter,
Identify descriptive text.
Fig. 3 is the structure diagram of handbill verifying unit provided in an embodiment of the present invention, as shown in figure 3, including：Cutting
The advertisement single image cutting for obtaining advertisement single image, and is multiple commodity image blocks by module 301；Identification module
302, for each commodity image block to be identified, obtain each commodity image merchandise classification information in the block and corresponding quotient
Product description information；Acquisition module 303, for according to making demand information corresponding with the advertisement single image, obtaining each quotient
The corresponding goal description information of product classification information；Checking module 304, for being retouched according to the corresponding target of each merchandise classification information
Information is stated to proofread descriptive labelling information.
Wherein, advertisement single image is made of multiple commodity image blocks, and each commodity image block is to include at least one
Commodity and its descriptive labelling information；Image merchandise classification information in the block is usually the image of commodity, and descriptive labelling information is
For describing the commodity, such as the text informations such as price or tag line；Cutting module 301 is proofreaded getting
Advertisement single image after, need to advertisement single image carry out cutting processing.
Wherein, identification module 302 is according to the commodity image block obtained in cutting module 301, in order to each Commdity advertisement
Content is proofreaded, therefore one side identification module 302 needs to identify the merchandise classification information corresponding to each commodity image block,
It is specifically which kind of commodity, merchandise classification information can be specifically the bar code or other identification codes of commodity；On the other hand identification mould
Block 302 needs to extract each image descriptive labelling information in the block.
Wherein, acquisition module 303 is retouched by making the target that demand information is got corresponding to each merchandise classification information
State information；Goal description information is the standard of check and correction, during check and correction is judged on handbill according to goal description information
Whether descriptive labelling information is wrong.
Wherein, goal description of the checking module 304 corresponding to by each merchandise classification information obtained in acquisition module 303
Information judges whether the descriptive labelling information obtained in identification module 302 is wrong, and will be correct and wrong as check and correction standard
Part is shown.
Handbill verifying unit provided in an embodiment of the present invention, by carrying out image cutting and content recognition to handbill,
Ad content is extracted, and ad content is proofreaded according to making demand, instead of artificial check and correction, improves correction efficiency.
On the basis of any of the above-described embodiment, the identification module 302 specifically includes：Classification recognition unit, for pair
Each commodity image block is identified, and obtains the corresponding merchandise classification information of the commodity image block；Information extraction unit is used for
Each commodity image description information image block in the block is extracted, and identifies description information image descriptive labelling letter in the block
Breath.
Specifically, for merchandise classification information, classification recognition unit can directly be got wherein according to commodity image block
Comprising commodity picture corresponding to merchandise classification information；And for the extraction of descriptive labelling information, due to descriptive labelling information
It is to exist in the form of images, therefore, information extraction unit extracts described in commodity image block frame block (wherein first
Include the word of description information), then description information image word in the block is identified, so as to get descriptive labelling
Information.
On the basis of any of the above-described embodiment, the cutting module 301 is specifically used for：The advertisement single image is inputted
The image segmentation model pre-established obtains multiple commodity image blocks of output；Correspondingly, described device further includes image cutting
Model training module, for according to merchandise classification information, descriptive labelling information and advertisement formwork, drawing multiple advertisement free hand drawings at random
Picture, and record the position rectangular coordinates of each commodity image block in each advertisement single image；By advertisement single image and need cutting
Position rectangular coordinates carry out model training respectively as the input value and output valve of model training, obtain the image that training is completed and cut
Sub-model.
On the basis of any of the above-described embodiment, the classification recognition unit is specifically used for：Each commodity image block is defeated
Enter the merchandise classification identification model pre-established, obtain the corresponding merchandise classification information of each commodity image block of output；Accordingly
Ground, described device further include merchandise classification identification model training module, for according to merchandise classification information, the life of descriptive labelling information
Into multiple commodity image blocks；And commodity image block is pre-processed, wherein, the pretreatment includes overturning, color is replaced, becomes
Shape distorts and adds at least one of noise；Pretreated commodity image block and corresponding merchandise classification information are made respectively
Input value and output valve for model training carry out model training, obtain the merchandise classification identification model that training is completed.
On the basis of any of the above-described embodiment, described information extraction unit is specifically used for：By each commodity image block point
The description information image block extraction model pre-established is not inputted, obtains each commodity image description information figure in the block of output
As block；Correspondingly, described device further includes description information image block extraction model training module, for being believed according to merchandise classification
Breath, descriptive labelling information generate multiple commodity image blocks, wherein, the commodity image block includes description information image block, described
Description information image block is generated by descriptive labelling information and randomly selected font；And the position square of record description frame block
Shape coordinate；Commodity image block is pre-processed, wherein, the pretreatment includes overturning, color replaces, distorts and add
At least one of noise；Using pretreated commodity image block and the position rectangular coordinates of cutting is needed to be instructed as model
Experienced input value and output valve carry out model training, obtain the description information image block extraction model that training is completed.
On the basis of any of the above-described embodiment, described information extraction unit is specifically used for：Description information image block is defeated
Enter in the description information identification model pre-established, obtain the descriptive labelling information of output；Correspondingly, described device, which further includes, retouches
Information identification model training module is stated, for generating multiple description information images according to descriptive labelling information and font random combine
Block；Description information image block is pre-processed, wherein, the pretreatment includes overturning, color replaces, distorts and add
At least one of noise；Using pretreated descriptive labelling frame block and descriptive labelling information as model training
Input value and output valve carry out model training, obtain training complete description information identification model.
Fig. 4 is the structure diagram that handbill provided in an embodiment of the present invention proofreads equipment, as shown in figure 4, the equipment packet
It includes：At least one processor 401；And at least one processor 402 with the processor 401 communication connection, wherein：It is described
Memory 402 is stored with the program instruction that can be performed by the processor 401, and the processor 301 calls described program instruction
It is able to carry out the handbill proofreading method that the various embodiments described above are provided, such as including：Advertisement single image is obtained, and will be described wide
It is multiple commodity image blocks to accuse single image cutting；Each commodity image block is identified, it is in the block to obtain each commodity image
Merchandise classification information and corresponding descriptive labelling information；According to making demand information corresponding with the advertisement single image, obtain
Each corresponding goal description information of merchandise classification information；According to the corresponding goal description information of each merchandise classification information to quotient
Product description information is proofreaded.
The embodiment of the present invention also provides a kind of non-transient computer readable storage medium storing program for executing, the non-transient computer readable storage
Medium storing computer instructs, which makes computer perform the handbill proofreading method that corresponding embodiment is provided,
Such as including：Advertisement single image is obtained, and is multiple commodity image blocks by the advertisement single image cutting；To each commodity image
Block is identified, and obtains each commodity image merchandise classification information in the block and corresponding descriptive labelling information；According to it is described
The corresponding making demand information of advertisement single image obtains the corresponding goal description information of each merchandise classification information；According to each
The corresponding goal description information of merchandise classification information proofreads descriptive labelling information.
The embodiments such as handbill check and correction equipment described above are only schematical, wherein illustrating as separating component
Unit may or may not be physically separate, the component shown as unit may or may not be
Physical unit, you can be located at a place or can also be distributed in multiple network element.It can be according to the actual needs
Some or all of module therein is selected to realize the purpose of this embodiment scheme.Those of ordinary skill in the art are not paying
In the case of performing creative labour, you can to understand and implement.
Through the above description of the embodiments, those skilled in the art can be understood that each embodiment can
It is realized by the mode of software plus required general hardware platform, naturally it is also possible to pass through hardware.Based on such understanding, on
Technical solution is stated substantially in other words to embody the part that the prior art contributes in the form of software product, it should
Computer software product can store in a computer-readable storage medium, such as ROM/RAM, magnetic disc, CD, including several fingers
It enables and (can be personal computer, server or the network equipment etc.) so that computer equipment is used to perform each implementation
Certain Part Methods of example or embodiment.
Handbill proofreading method provided in an embodiment of the present invention, device and equipment, as shown in Fig. 2, using deep learning skill
Art, the deep learning model handled by visual pattern, to the commodity in handbill, word description and commodity price carry out
Identification, the content that will be extracted are compared with the making statement of requirements that the party in request of handbill provides by software program, known
Do not go out meeting the part of design requirement and do not meet the part of design requirement, and pass through computer display in handbill,
Show the result of model check and correction；Advertisement single image is subjected to automatic Proofreading, hand labor is substituted, makes the proof-reading of handbill
30~90 minutes are needed by individual handbill proof time before, is reduced to no more than 5 minutes and completes, and can be effective
Discovery descriptive labelling and price, with the statement of requirements of handbill party in request it is inconsistent the problem of.
Finally it should be noted that：The above embodiments are merely illustrative of the technical solutions of the present invention, rather than its limitations；Although
The present invention is described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that：It still may be used
To modify to the technical solution recorded in foregoing embodiments or carry out equivalent replacement to which part technical characteristic；
And these modification or replace, various embodiments of the present invention technical solution that it does not separate the essence of the corresponding technical solution spirit and
Range.
Claims (10)
1. a kind of handbill proofreading method, which is characterized in that including：
Advertisement single image is obtained, and is multiple commodity image blocks by the advertisement single image cutting；
Each commodity image block is identified, each commodity image merchandise classification information in the block is obtained and corresponding commodity is retouched
State information；
According to making demand information corresponding with the advertisement single image, the corresponding goal description of each merchandise classification information is obtained
Information；
Descriptive labelling information is proofreaded according to each merchandise classification information corresponding goal description information.
2. according to the method described in claim 1, it is characterized in that, described be identified each commodity image block, acquisition is every
The step of a commodity image merchandise classification information in the block and corresponding descriptive labelling information, specifically includes：
Each commodity image block is identified, obtains the corresponding merchandise classification information of the commodity image block；And extraction is every
A commodity image description information image block in the block, and identify description information image descriptive labelling information in the block.
3. according to the method described in claim 1, it is characterized in that, it is multiple commodity image blocks by the advertisement single image cutting
The step of specifically include：
The image segmentation model that advertisement single image input is pre-established obtains multiple commodity image blocks of output；
Correspondingly, it was further included before the step of image segmentation model advertisement single image input pre-established,
According to merchandise classification information, descriptive labelling information and advertisement formwork, multiple advertisement single images are drawn at random, and are recorded each
The position rectangular coordinates of each commodity image block in advertisement single image；
The input value and output valve of advertisement single image and the position rectangular coordinates as the model training that need cutting are carried out
Model training obtains the image segmentation model that training is completed.
4. according to the method described in claim 2, it is characterized in that, described be identified each commodity image block, institute is obtained
The step of stating commodity image block corresponding merchandise classification information specifically includes：
The merchandise classification identification model that the input of each commodity image block is pre-established obtains each commodity image block pair of output
The merchandise classification information answered；
Correspondingly, it was further included before described the step of each commodity image block is inputted the merchandise classification identification model pre-established,
Multiple commodity image blocks are generated according to merchandise classification information, descriptive labelling information；And commodity image block is pre-processed,
Wherein, the pretreatment includes overturning, at least one of noise is replaced, distorts and added to color；
Input value and output using pretreated commodity image block and corresponding merchandise classification information as model training
Value carries out model training, obtains the merchandise classification identification model that training is completed.
5. the according to the method described in claim 2, it is characterized in that, each commodity image description information figure in the block of extraction
As the step of block specifically includes：
Each commodity image block is inputted to the description information image block extraction model pre-established respectively, obtains each quotient of output
Product image description information image block in the block；
Correspondingly, described the step of each commodity image block is inputted into the description information image block extraction model pre-established respectively
Before further include,
Multiple commodity image blocks are generated according to merchandise classification information, descriptive labelling information, wherein, the commodity image block includes retouching
Frame block is stated, the description information image block is generated by descriptive labelling information and randomly selected font；And record description
The position rectangular coordinates of frame block；
Commodity image block is pre-processed, wherein, the pretreatment includes overturning, color is replaced, distort and addition is made an uproar
At least one of sound；
Using pretreated commodity image block and need the position rectangular coordinates of cutting as the input value of model training and
Output valve carries out model training, obtains the description information image block extraction model that training is completed.
6. according to the method described in claim 2, it is characterized in that, the identification description information image commodity in the block are retouched
The step of stating information specifically includes：
In the description information identification model that the input of description information image block is pre-established, the descriptive labelling information of output is obtained；
Correspondingly, it is also wrapped before the step in the description information identification model that the input of description information image block is pre-established
It includes,
Multiple description information image blocks are generated according to descriptive labelling information and font random combine；
Description information image block is pre-processed, wherein, the pretreatment includes overturning, color replaces, distorts and add
At least one of plus noise；
Using pretreated descriptive labelling frame block and descriptive labelling information as the input value of model training and defeated
Go out value and carry out model training, obtain the description information identification model that training is completed.
7. a kind of handbill verifying unit, which is characterized in that including：
The advertisement single image cutting for obtaining advertisement single image, and is multiple commodity image blocks by cutting module；
Identification module for each commodity image block to be identified, obtains each commodity image merchandise classification information in the block
And corresponding descriptive labelling information；
Acquisition module, for according to making demand information corresponding with the advertisement single image, obtaining each merchandise classification information
Corresponding goal description information；
Checking module, for carrying out school to descriptive labelling information according to the corresponding goal description information of each merchandise classification information
It is right.
8. device according to claim 7, which is characterized in that the identification module specifically includes：
Classification recognition unit for each commodity image block to be identified, obtains the corresponding commodity class of the commodity image block
Other information；
Information extraction unit for extracting each commodity image description information image block in the block, and identifies the description information
Image descriptive labelling information in the block.
9. a kind of handbill proofreads equipment, which is characterized in that including：
At least one processor；
And at least one processor being connect with the processor communication, wherein：The memory is stored with can be by the place
The program instruction that device performs is managed, the processor calls described program instruction to be able to carry out as described in claim 1 to 6 is any
Method.
10. a kind of non-transient computer readable storage medium storing program for executing, which is characterized in that the non-transient computer readable storage medium storing program for executing is deposited
Computer instruction is stored up, the computer instruction makes the computer perform the method as described in claim 1 to 6 is any.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711483666.7A CN108230308B (en) | 2017-12-29 | 2017-12-29 | Method, device and equipment for checking advertisement bill |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711483666.7A CN108230308B (en) | 2017-12-29 | 2017-12-29 | Method, device and equipment for checking advertisement bill |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108230308A true CN108230308A (en) | 2018-06-29 |
| CN108230308B CN108230308B (en) | 2021-02-09 |
Family
ID=62646185
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711483666.7A Active CN108230308B (en) | 2017-12-29 | 2017-12-29 | Method, device and equipment for checking advertisement bill |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108230308B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109543057A (en) * | 2018-11-20 | 2019-03-29 | 广州逗号智能零售有限公司 | Commodity recognition method, device, equipment and storage medium based on intelligent cashier platform |
| CN111144438A (en) * | 2019-11-26 | 2020-05-12 | 苏州方正璞华信息技术有限公司 | Method and device for detecting commodities in advertisement bill |
| CN112232320A (en) * | 2020-12-14 | 2021-01-15 | 北京沃东天骏信息技术有限公司 | Printed matter text proofreading method and related equipment |
| CN113610872A (en) * | 2021-08-24 | 2021-11-05 | 广东车前传媒有限公司 | Barrier advertisement management method, system, equipment and storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103257954A (en) * | 2013-06-05 | 2013-08-21 | 北京百度网讯科技有限公司 | Proofreading method, system and proofreading server of characters in ancient book |
| CN105590101A (en) * | 2015-12-28 | 2016-05-18 | 杭州淳敏软件技术有限公司 | Hand-written answer sheet automatic processing and marking method and system based on mobile phone photographing |
| CN106447366A (en) * | 2015-08-07 | 2017-02-22 | 百度在线网络技术（北京）有限公司 | Checking method of multimedia advertisement, and training method and apparatus of advertisement checking model |
| CN106778703A (en) * | 2017-01-25 | 2017-05-31 | 华中师范大学 | The method and apparatus of electric marking |
| CN107067399A (en) * | 2017-02-13 | 2017-08-18 | 杭州施强教育科技有限公司 | A kind of paper image segmentation processing method |
-
2017
- 2017-12-29 CN CN201711483666.7A patent/CN108230308B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103257954A (en) * | 2013-06-05 | 2013-08-21 | 北京百度网讯科技有限公司 | Proofreading method, system and proofreading server of characters in ancient book |
| CN106447366A (en) * | 2015-08-07 | 2017-02-22 | 百度在线网络技术（北京）有限公司 | Checking method of multimedia advertisement, and training method and apparatus of advertisement checking model |
| CN105590101A (en) * | 2015-12-28 | 2016-05-18 | 杭州淳敏软件技术有限公司 | Hand-written answer sheet automatic processing and marking method and system based on mobile phone photographing |
| CN106778703A (en) * | 2017-01-25 | 2017-05-31 | 华中师范大学 | The method and apparatus of electric marking |
| CN107067399A (en) * | 2017-02-13 | 2017-08-18 | 杭州施强教育科技有限公司 | A kind of paper image segmentation processing method |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109543057A (en) * | 2018-11-20 | 2019-03-29 | 广州逗号智能零售有限公司 | Commodity recognition method, device, equipment and storage medium based on intelligent cashier platform |
| CN111144438A (en) * | 2019-11-26 | 2020-05-12 | 苏州方正璞华信息技术有限公司 | Method and device for detecting commodities in advertisement bill |
| CN112232320A (en) * | 2020-12-14 | 2021-01-15 | 北京沃东天骏信息技术有限公司 | Printed matter text proofreading method and related equipment |
| CN112232320B (en) * | 2020-12-14 | 2021-05-25 | 北京沃东天骏信息技术有限公司 | Printed matter text proofreading method and related equipment |
| CN113610872A (en) * | 2021-08-24 | 2021-11-05 | 广东车前传媒有限公司 | Barrier advertisement management method, system, equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108230308B (en) | 2021-02-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108230308A (en) | A kind of handbill proofreading method, device and equipment | |
| Liu et al. | Recurrent multimodal interaction for referring image segmentation | |
| CN110390269A (en) | PDF document table extracting method, device, equipment and computer readable storage medium | |
| CN111753767A (en) | Method and device for automatically correcting operation, electronic equipment and storage medium | |
| CN107274345A (en) | A kind of Chinese printable character image combining method and device | |
| CN107545391A (en) | A kind of logistics document intellectual analysis and automatic storage method based on image recognition | |
| CN107516096A (en) | A kind of character identifying method and device | |
| CN110114776A (en) | Use the system and method for the character recognition of full convolutional neural networks | |
| CN110969681A (en) | Method for generating handwriting characters based on GAN network | |
| CN109615671A (en) | A kind of character library sample automatic generation method, computer installation and readable storage medium storing program for executing | |
| CN111753120A (en) | Method and device for searching questions, electronic equipment and storage medium | |
| CN109063720A (en) | Handwritten word training sample acquisition methods, device, computer equipment and storage medium | |
| CN104408403B (en) | A kind of referee method that secondary typing is inconsistent and device | |
| CN110287911A (en) | A kind of content identification method of invoice, device, equipment and storage medium | |
| CN114429636B (en) | Image scanning identification method and device and electronic equipment | |
| CN110399760A (en) | A kind of batch two dimensional code localization method, device, electronic equipment and storage medium | |
| CN113628113B (en) | Image stitching method and related equipment thereof | |
| CN111524503A (en) | Audio data processing method and device, audio recognition equipment and storage medium | |
| CN113342829A (en) | Script processing method and device, electronic equipment and computer storage medium | |
| CN116168275B (en) | Lightweight dual-attention mechanism identification method based on feature grouping and channel replacement | |
| CN115830599B (en) | Industrial character recognition method, model training method, device, equipment and medium | |
| CN116778032A (en) | Answer sheet generation method, device, equipment and storage medium | |
| CN113657279B (en) | Bill image layout analysis method and device | |
| CN108446701A (en) | A kind of best bounding volume method of writing brush word | |
| US9443139B1 (en) | Methods and apparatus for identifying labels and/or information associated with a label and/or using identified information |
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 |