CN106529609A - Image recognition method and device based on neural network structure - Google Patents
Image recognition method and device based on neural network structure Download PDFInfo
- Publication number
- CN106529609A CN106529609A CN201611122462.6A CN201611122462A CN106529609A CN 106529609 A CN106529609 A CN 106529609A CN 201611122462 A CN201611122462 A CN 201611122462A CN 106529609 A CN106529609 A CN 106529609A
- Authority
- CN
- China
- Prior art keywords
- look
- list item
- image
- network structure
- neural network
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
- G06V30/19—Recognition using electronic means
- G06V30/192—Recognition using electronic means using simultaneous comparisons or correlations of the image signals with a plurality of references
- G06V30/194—References adjustable by an adaptive method, e.g. learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Databases & Information Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses an image recognition method based on a neural network structure. The method comprises the following steps: determining a target image to be recognized according to an image recognition instruction; aiming at each lookup table in a pre-acquired lookup table set, matching each pixel of the target image with each entry of the lookup table, wherein the lookup table set comprises a plurality of lookup tables having different contents and corresponding to reference images, each lookup table is established on the basis of the neural network structure, and the entries in each lookup table are neurons of the neural network structure; and determining a reference image corresponding to the target image according to the matching result. By adopting the technical solution, the image recognition method based on a neural network structure simulates the learning competence of human brain, so that the image recognition is more accurate. The invention further discloses an image recognition device based on a neural network structure, having the corresponding advantages.
Description
Technical field
The present invention relates to Computer Applied Technology field, more particularly to a kind of image recognition based on neural network structure
Method and device.
Background technology
With the fast development of Computer Applied Technology, the application of image recognition technology is more and more extensive, to image recognition
Demand it is also more and more.Such as in video monitoring, or in signature is compared, it is required for carrying out image recognition.
With the eyes to human brain structure and neutral net, technical staff has understood and has gradually explored human brain to information
Process and working method.But how accurately identifying for image is carried out based on neural network structure, is current people in the art
The technical problem of member's urgent need to resolve.
The content of the invention
It is an object of the invention to provide a kind of image-recognizing method and device based on neural network structure, with based on nerve
Network structure is accurately identified to image.
For solving above-mentioned technical problem, the present invention provides following technical scheme：
A kind of image-recognizing method based on neural network structure, including：
Instructed according to image recognition, determine target image to be identified；
For each look-up table in the look-up table set that is obtained ahead of time, by each pixel of the target image and the lookup
Each list item of table is matched, and the look-up table set includes the corresponding lookup of multiple reference pictures with different content
Table, each look-up table are set up based on neural network structure, and the list item in each look-up table is the nerve of the neural network structure
Unit；
According to matching result, the corresponding reference picture of the target image is determined.
It is in a kind of specific embodiment of the present invention, described for each lookup in the look-up table set that is obtained ahead of time
Table, each pixel of the target image is matched with each list item of the look-up table, including：
For each look-up table in the look-up table set that is obtained ahead of time, by each pixel of the target image and the lookup
In table, the list item of respective pixel location is compared；
For the list item of each location of pixels in the look-up table, if the list item of the location of pixels had learning records,
Postpone to the next list item in location of pixels rear of the look-up table and do Iterative matching, until it reaches at predetermined depth.
In a kind of specific embodiment of the present invention, also include：
For each look-up table in the look-up table set, if setting none of figure to be identified in duration
As being identified as the corresponding reference picture of the look-up table, then the look-up table is deleted in the look-up table set.
In a kind of specific embodiment of the present invention, for any one reference picture, obtained by following steps in advance
Obtain the corresponding look-up table of the reference picture：
The image data stream to be learned for the reference picture is received, each image pattern has in described image data flow
Have and the reference picture identical content；
For each pixel of each image pattern in described image data flow, using the corresponding picture in a look-up table
The list item of plain position learns to the pixel；
Look-up table after study is defined as into the corresponding look-up table of the reference picture.
In a kind of specific embodiment of the present invention, each image pattern in the data flow for described image
Each pixel, is learnt to the pixel using the list item of the respective pixel location in a look-up table, including：
If in the list item Reception learning first of the location of pixels, directly learnt at the list item of the location of pixels
Record；
If in the non-Reception learning first of the list item of the location of pixels, being iterated to the location of pixels rear list item
Practise, until the list item of iteration to Reception learning first.
A kind of pattern recognition device based on neural network structure, including：
Target image determining module, for instructing according to image recognition, determines target image to be identified；
List item matching module, for each look-up table in the look-up table set for being obtained ahead of time, by the target image
Each pixel matched with each list item of the look-up table, the look-up table set include multiple ginsengs with different content
The corresponding look-up table of image is examined, each look-up table is set up based on neural network structure, the list item in each look-up table is the god
The neuron of Jing network structures；
Reference picture determining module, for according to matching result, determining the corresponding reference picture of the target image.
In a kind of specific embodiment of the present invention, the list item matching module, specifically for：
For each look-up table in the look-up table set that is obtained ahead of time, by each pixel of the target image and the lookup
In table, the list item of respective pixel location is compared；
For the list item of each location of pixels in the look-up table, if the list item of the location of pixels had learning records,
Postpone to the next list item in location of pixels rear of the look-up table and do Iterative matching, until it reaches at predetermined depth.
In a kind of specific embodiment of the present invention, also including look-up table removing module, it is used for：
For each look-up table in the look-up table set, if setting none of figure to be identified in duration
As being identified as the corresponding reference picture of the look-up table, then the look-up table is deleted in the look-up table set.
In a kind of specific embodiment of the present invention, also module is obtained including look-up table, for for any one ginseng
Image is examined, the corresponding look-up table of the reference picture is obtained ahead of time by following steps：
The image data stream to be learned for the reference picture is received, each image pattern has in described image data flow
Have and the reference picture identical content；
For each pixel of each image pattern in described image data flow, using the corresponding picture in a look-up table
The list item of plain position learns to the pixel；
Look-up table after study is defined as into the corresponding look-up table of the reference picture.
In a kind of specific embodiment of the present invention, the look-up table obtains module, specifically for：
If in the list item Reception learning first of the location of pixels, directly learnt at the list item of the location of pixels
Record；
If in the non-Reception learning first of the list item of the location of pixels, being iterated to the location of pixels rear list item
Practise, until the list item of iteration to Reception learning first.
The technical scheme provided using the embodiment of the present invention, instructs according to image recognition, it may be determined that mesh to be identified
Logo image, for each look-up table in the look-up table set that is obtained ahead of time, by each pixel of target image and the look-up table
Each list item is matched, according to matching result, it may be determined that the corresponding reference picture of target image, i.e., target image is more like which
One reference picture.Each look-up table in look-up table set is set up based on neural network structure, is obtained by image study, should
The corresponding look-up table of multiple reference pictures with different content is included in set, and the list item of each look-up table is tied for neutral net
The neuron of structure.Image recognition is carried out based on neural network structure, the learning capacity of human brain is simulated so that image recognition is more
Accurately.
Description of the drawings
In order to be illustrated more clearly that the embodiment of the present invention or technical scheme of the prior art, below will be to embodiment or existing
Accompanying drawing to be used needed for having technology description is briefly described, it should be apparent that, drawings in the following description are only this
Some embodiments of invention, for those of ordinary skill in the art, on the premise of not paying creative work, can be with
Other accompanying drawings are obtained according to these accompanying drawings.
Fig. 1 is a kind of implementing procedure figure of the image-recognizing method based on neural network structure in the embodiment of the present invention；
Fig. 2 is a kind of structural representation of image identification system in the embodiment of the present invention；
Fig. 3 is a kind of schematic diagram of neural network structure in the embodiment of the present invention；
Fig. 4 is the schematic diagram of an image pattern in the embodiment of the present invention；
Fig. 5 is sample learning process schematic in the embodiment of the present invention；
Fig. 6 is image recognition processes schematic diagram in the embodiment of the present invention；
Fig. 7 is look-up table read-write schematic diagram in the embodiment of the present invention；
Fig. 8 is a kind of structural representation of the pattern recognition device based on neural network structure in the embodiment of the present invention.
Specific embodiment
In order that those skilled in the art more fully understand the present invention program, with reference to the accompanying drawings and detailed description
The present invention is described in further detail.Obviously, described embodiment is only a part of embodiment of the invention, rather than
Whole embodiments.Based on the embodiment in the present invention, those of ordinary skill in the art are not making creative work premise
Lower obtained every other embodiment, belongs to the scope of protection of the invention.
A kind of image-recognizing method based on neural network structure shown in Figure 1, being provided by the embodiment of the present invention
Implementing procedure figure, the method may comprise steps of：
S110：Instructed according to image recognition, determine target image to be identified.
In embodiments of the present invention, control unit can receive image recognition instruction by front end interface unit, such as Fig. 2 institutes
Show.Target image to be identified can be carried in image recognition instruction.Instructed according to image recognition, it may be determined that mesh to be identified
Logo image.
Front end interface can parse upper strata instruction, and that what is interacted can be the PCI-E (PCI under general x86 frameworks
Express, EBI of new generation), can also be AXI under ARM frameworks (Advancedextensible Interface, one
Kind of bus protocol) etc. interface.
When image recognition instruction is received, first the target image in image recognition instruction can be entered by computing unit
Row pretreatment, to reach the purpose for accelerating identification, as shown in Figure 2.Such as, it is possible to use convolution feature extraction algorithm, by high picture
Sketch map picture is matched to the specification of the look-up table that rear end is set up based on neural network structure, and make image profile become apparent from it is bright
It is aobvious, or, the rotation recognition mode to recognizing image can be set, for example, what is be obtained ahead of time is set up based on neural network structure
Look-up table learning cross a upright character picture " M ", then can be identified as the reversion for learning when " W " is received
180 ° of " M ", or, can learn and recognize the color of content with labelling, the different colours of Division identification image are represented not
Same meaning.
Neural network structure that the embodiment of the present invention is based on as shown in figure 3, be a cube structure, cubical three
Dimension coordinate system is respectively XwYhZd, and each of which node is a neuron.In figure 3, if target image is 64*64 pictures
Plain size, then each of which pixel can correspond to [X0, Y0] to [X63, Y63] respective pixel location neuron.
After determining target image to be identified, the operation of step S120 can be continued executing with.
S120：For each look-up table in the look-up table set that is obtained ahead of time, each pixel and this of target image are looked into
Each list item of table is looked for be matched.
Look-up table set includes the corresponding look-up table of multiple reference pictures with different content, and each look-up table is based on god
Jing network structures are set up, neuron of the list item in each look-up table for neural network structure.
In embodiments of the present invention, the corresponding lookup of reference picture with different content can be obtained by image study
Table, look-up table are set up based on neural network structure.That is, each has the look-up table of neural network structure corresponding to not
With the reference picture of content.Multiple look-up tables constitute look-up table set.
In a kind of specific embodiment of the present invention, for any one reference picture, can be pre- by following steps
The corresponding look-up table of the reference picture is obtained first：
First step：Receive the image data stream to be learned for the reference picture, each figure in image data stream
Decent with the reference picture identical content；
Second step：For each pixel of each image pattern in image data stream, using in a look-up table
The list item of respective pixel location the pixel is learnt；
3rd step：Look-up table after study is defined as into the corresponding look-up table of the reference picture.
For ease of description, above three step is combined and is illustrated.
In actual applications, the study instruction for reference picture can be received by front end interface.Can in study instruction
To carry the image data stream to be learned for the reference picture.In image data stream each image pattern with the reference
Image identical content, concrete manifestation form may be different.Can the content information comprising reference picture in image data stream.
Look-up table based on neural network structure learns to each image pattern in image data stream, and corresponding
Record in list item.For each pixel of each image pattern in image data stream, using the corresponding picture in a look-up table
The list item of plain position learns to the pixel, obtains the information such as image implication, characteristics of image, can configure in learning process
Study weighted value.
Specifically, if in the list item Reception learning first of the location of pixels, directly at the list item of the location of pixels
Carry out learning records；If in the non-Reception learning first of the list item of the location of pixels, carried out to the location of pixels rear list item
Iterative learning, until the list item of iteration to Reception learning first.
Such as, Fig. 4 is the schematic diagram of an image pattern, and the image pattern is learnt, shown in Fig. 3 based on god
Learning records are carried out at the list item of the respective pixel location of the look-up table that Jing network structures are set up, the study note shown in Fig. 3 is obtained
Record result.If X2Y3Z1The list item at place Reception learning first, then directly can carry out learning records at the list item, if should
The non-Reception learning first of list item, then can be to Z-direction list item, i.e. X2Y3Z2The list item at place is iterated study, until iteration is extremely
The list item of Reception learning first.
Concrete learning process is may be referred to shown in Fig. 5, after configuring corresponding learning parameter, carries out sample learning, until study
Complete.During sample learning, first according to search index look-up table, according to configurable write list item, determine whether list item has overwriting, such as
Fruit is then to change message to prepare to look into latter list item, if it is not, then study is completed.
By learning records, the look-up table after study can be defined as the corresponding look-up table of the reference picture.
When image pattern to be learnt is few, the purpose of Fast Learning can be reached with the self-defined dispersion of distribution and depth.
In a kind of specific embodiment of the present invention, before above-mentioned second step, can also be to image data stream
In each image carry out acceleration pretreatment.
Specifically, the preprocessing process of identification process is may refer to, the embodiment of the present invention will not be described here.
For each look-up table in the look-up table set that is obtained ahead of time, by each pixel of target image and the look-up table
Each list item is matched.
In a kind of specific embodiment of the present invention, step S120 may comprise steps of：
Step one：For each look-up table in the look-up table set that is obtained ahead of time, by each pixel of target image with should
In look-up table, the list item of respective pixel location is compared；
Step 2：For the list item of each location of pixels in the look-up table, if the list item of the location of pixels had study
Record, then postpone to the next list item in location of pixels rear of the look-up table and do Iterative matching, until it reaches at predetermined depth.
For ease of description, above-mentioned two step is combined and is illustrated.
For being obtained ahead of time each look-up table in look-up table set, the look-up table corresponds to a reference picture, and which can be with
Obtained by learning various deformation patterns with identical content of the reference picture.
By the list item of each pixel correspondence to respective pixel location in the look-up table of target image, and with respective pixel position
The list item put is compared.For the list item of each location of pixels in the look-up table, if the list item of the location of pixels had
Record is practised, then can be postponed to the next list item in location of pixels rear of the look-up table and be done Iterative matching, that is, postpone to this and look into
Look for and Iterative matching is done at the next list item in Z axis rear of table, until it reaches at predetermined depth.
For each pixel of target image, by by the pixel with the look-up table list item of respective pixel location
Match somebody with somebody, the matching degree of the learning records of pixel list item corresponding to each can be obtained, quantifiable matching can be finally obtained
Depth results.
Predetermined depth can be set according to practical situation and be adjusted, and the embodiment of the present invention is without limitation.
S130：According to matching result, the corresponding reference picture of target image is determined.
In step S120, for each look-up table in the look-up table set that is obtained ahead of time, by each pixel of target image
Matched with each list item of the look-up table, each pixel of target image and the matching result of the look-up table can be obtained,
According to matching result, it may be determined that the corresponding reference picture of target image.
Specifically, the maximum corresponding reference picture of look-up table of matching degree can be defined as the corresponding ginseng of target image
Examine image.
Specific identification process may be referred to Fig. 6, after input identification image, be recognized by look-up table, and identification completes to return
As a result.When being recognized by look-up table, first according to search index look-up table, list item is analyzed, determines whether list item has overwriting, such as
Fruit is then to change message to prepare to look into latter list item, if it is not, then identification is completed.
Read-write look-up table schematic diagram is as shown in fig. 7, table request module Tbl_req sends one after a Pixel Information input
The request of individual meter reading item is sent to ddr interface Ddr_if, and ddr interface Ddr_if can read data from DDR, returns to table confirmation
Module Tbl_act is matched or is learnt, and the data after iteration is written back in the list item of DDR when returning
Face.
Identification process of the embodiment of the present invention to image, is the learning capacity for simulating human brain, realizes the image to learning
Obtain target image more " as " which, rather than "Yes" which, realize that an image is matched in the information to learning
Identification obtains a study depth that can quantify, and reaches the purpose of identification image.
The method provided using the embodiment of the present invention, instructs according to image recognition, it may be determined that target figure to be identified
Picture, for each look-up table in the look-up table set that is obtained ahead of time, by each pixel of target image and the look-up table each
List item is matched, according to matching result, it may be determined that the corresponding reference picture of target image, i.e., target image is more like which
Reference picture.Each look-up table in look-up table set is set up based on neural network structure, is obtained by image study, the set
In include the corresponding look-up table of multiple reference pictures with different content, the list item of each look-up table is neural network structure
Neuron.Image recognition is carried out based on neural network structure, the learning capacity of human brain is simulated so that image recognition is more accurate
Really.
In one embodiment of the invention, the method can also be comprised the following steps：
For each look-up table in look-up table set, if be identified none of target image in duration is set
For the corresponding reference picture of the look-up table, then the look-up table is deleted in look-up table set.
Specifically, the forgetting feature to learning image can be simulated, by look-up table by configuring forgotten memory rate parameter
In set, long-term untapped look-up table carries out deletion action, when being identified to image, to save match time, improves and knows
Other efficiency.
Corresponding to above method embodiment, the embodiment of the present invention additionally provides a kind of image based on neural network structure
Identifying device, a kind of pattern recognition device based on neural network structure described below are a kind of based on nerve with above-described
The image-recognizing method of network structure can be mutually to should refer to.
Shown in Figure 4, the device is included with lower module：
Target image determining module 210, for instructing according to image recognition, determines target image to be identified；
List item matching module 220, for each look-up table in the look-up table set for being obtained ahead of time, by target image
Each pixel is matched with each list item of the look-up table, and look-up table set includes multiple reference pictures with different content
Corresponding look-up table, each look-up table are set up based on neural network structure, and the list item in each look-up table is neural network structure
Neuron；
Reference picture determining module 230, for according to matching result, determining the corresponding reference picture of target image.
The device provided using the embodiment of the present invention, instructs according to image recognition, it may be determined that target figure to be identified
Picture, for each look-up table in the look-up table set that is obtained ahead of time, by each pixel of target image and the look-up table each
List item is matched, according to matching result, it may be determined that the corresponding reference picture of target image, i.e., target image is more like which
Reference picture.Each look-up table in look-up table set is set up based on neural network structure, is obtained by image study, the set
In include the corresponding look-up table of multiple reference pictures with different content, the list item of each look-up table is neural network structure
Neuron.Image recognition is carried out based on neural network structure, the learning capacity of human brain is simulated so that image recognition is more accurate
Really.
In a kind of specific embodiment of the present invention, list item matching module 220, specifically for：
For each look-up table in the look-up table set that is obtained ahead of time, by each pixel of target image and the look-up table
The list item of respective pixel location is compared；
For the list item of each location of pixels in the look-up table, if the list item of the location of pixels had learning records,
Postpone to the next list item in location of pixels rear of the look-up table and do Iterative matching, until it reaches at predetermined depth.
In a kind of specific embodiment of the present invention, also including look-up table removing module, it is used for：
For each look-up table in look-up table set, if setting none of image quilt to be identified in duration
The corresponding reference picture of the look-up table is identified as, then deletes the look-up table in look-up table set.
In a kind of specific embodiment of the present invention, also module is obtained including look-up table, for for any one ginseng
Image is examined, the corresponding look-up table of the reference picture is obtained ahead of time by following steps：
Receive for the reference picture image data stream to be learned, in image data stream each image pattern with
The reference picture identical content；
For each pixel of each image pattern in image data stream, using the respective pixel position in a look-up table
The list item put learns to the pixel；
Look-up table after study is defined as into the corresponding look-up table of the reference picture.
In a kind of specific embodiment of the present invention, look-up table obtains module, specifically for：
If in the list item Reception learning first of the location of pixels, directly learnt at the list item of the location of pixels
Record；
If in the non-Reception learning first of the list item of the location of pixels, being iterated to the location of pixels rear list item
Practise, until the list item of iteration to Reception learning first.
In this specification, each embodiment is described by the way of progressive, and what each embodiment was stressed is and other
The difference of embodiment, between each embodiment same or similar part mutually referring to.For dress disclosed in embodiment
For putting, as which corresponds to the method disclosed in Example, so description is fairly simple, related part is referring to method part
Illustrate.
Professional further appreciates that, with reference to the unit of each example of the embodiments described herein description
And algorithm steps, can with electronic hardware, computer software or the two be implemented in combination in, in order to clearly demonstrate hardware and
The interchangeability of software, generally describes composition and the step of each example in the above description according to function.These
Function actually with hardware or software mode performing, the application-specific and design constraint depending on technical scheme.Specialty
Technical staff can use different methods to realize described function to each specific application, but this realization should
Think beyond the scope of this invention.
The step of method described with reference to the embodiments described herein or algorithm, directly can be held with hardware, processor
Capable software module, or the combination of the two is implementing.Software module can be placed in random access memory (RAM), internal memory, read-only deposit
Reservoir (ROM), electrically programmable ROM, electrically erasable ROM, depositor, hard disk, moveable magnetic disc, CD-ROM or technology
In any other form of storage medium well known in field.
Specific case used herein is set forth to the principle of the present invention and embodiment, and above example is said
It is bright to be only intended to help and understand technical scheme and its core concept.It should be pointed out that common for the art
For technical staff, under the premise without departing from the principles of the invention, some improvement and modification can also be carried out to the present invention, these
Improve and modification is also fallen in the protection domain of the claims in the present invention.
Claims (10)
1. a kind of image-recognizing method based on neural network structure, it is characterised in that include：
Instructed according to image recognition, determine target image to be identified；
For each look-up table in the look-up table set that is obtained ahead of time, by each pixel of the target image and the look-up table
Each list item is matched, and the look-up table set includes the corresponding look-up table of multiple reference pictures with different content, often
Individual look-up table is set up based on neural network structure, and the list item in each look-up table is the neuron of the neural network structure；
According to matching result, the corresponding reference picture of the target image is determined.
2. the image-recognizing method based on neural network structure according to claim 1, it is characterised in that described for pre-
Each look-up table in the look-up table set for first obtaining, each list item of each pixel of the target image with the look-up table is entered
Row matching, including：
For each look-up table in the look-up table set that is obtained ahead of time, by each pixel of the target image and the look-up table
The list item of respective pixel location is compared；
For the list item of each location of pixels in the look-up table, if the list item of the location of pixels there are learning records, postpone
Iterative matching is done to the next list item in location of pixels rear of the look-up table, until it reaches at predetermined depth.
3. the image-recognizing method based on neural network structure according to claim 1, it is characterised in that also include：
For each look-up table in the look-up table set, if setting none of image quilt to be identified in duration
The corresponding reference picture of the look-up table is identified as, then the look-up table is deleted in the look-up table set.
4. the image-recognizing method based on neural network structure according to any one of claims 1 to 3, it is characterised in that
For any one reference picture, the corresponding look-up table of the reference picture is obtained ahead of time by following steps：
Receive for the reference picture image data stream to be learned, in described image data flow each image pattern with
The reference picture identical content；
For each pixel of each image pattern in described image data flow, using the respective pixel position in a look-up table
The list item put learns to the pixel；
Look-up table after study is defined as into the corresponding look-up table of the reference picture.
5. the image-recognizing method based on neural network structure according to claim 4, it is characterised in that described for institute
Each pixel of each image pattern in image data stream is stated, using the list item pair of the respective pixel location in a look-up table
The pixel is learnt, including：
If in the list item Reception learning first of the location of pixels, directly carrying out study note at the list item of the location of pixels
Record；
If in the non-Reception learning first of the list item of the location of pixels, being iterated study to the location of pixels rear list item,
Until the list item of iteration to Reception learning first.
6. a kind of pattern recognition device based on neural network structure, it is characterised in that include：
Target image determining module, for instructing according to image recognition, determines target image to be identified；
List item matching module, for each look-up table in the look-up table set for being obtained ahead of time, by the every of the target image
Individual pixel is matched with each list item of the look-up table, the look-up table set comprising it is multiple with different content with reference to figure
As corresponding look-up table, each look-up table is set up based on neural network structure, and the list item in each look-up table is the nerve net
The neuron of network structure；
Reference picture determining module, for according to matching result, determining the corresponding reference picture of the target image.
7. the pattern recognition device based on neural network structure according to claim 6, it is characterised in that the list item
With module, specifically for：
For each look-up table in the look-up table set that is obtained ahead of time, by each pixel of the target image and the look-up table
The list item of respective pixel location is compared；
For the list item of each location of pixels in the look-up table, if the list item of the location of pixels there are learning records, postpone
Iterative matching is done to the next list item in location of pixels rear of the look-up table, until it reaches at predetermined depth.
8. the pattern recognition device based on neural network structure according to claim 6, it is characterised in that also including lookup
Table removing module, is used for：
For each look-up table in the look-up table set, if setting none of image quilt to be identified in duration
The corresponding reference picture of the look-up table is identified as, then the look-up table is deleted in the look-up table set.
9. the pattern recognition device based on neural network structure according to any one of claim 6 to 8, it is characterised in that
Also include that look-up table obtains module, for for any one reference picture, the reference picture being obtained ahead of time by following steps
Corresponding look-up table：
Receive for the reference picture image data stream to be learned, in described image data flow each image pattern with
The reference picture identical content；
For each pixel of each image pattern in described image data flow, using the respective pixel position in a look-up table
The list item put learns to the pixel；
Look-up table after study is defined as into the corresponding look-up table of the reference picture.
10. the pattern recognition device based on neural network structure according to claim 9, it is characterised in that the lookup
Table obtains module, specifically for：
If in the list item Reception learning first of the location of pixels, directly carrying out study note at the list item of the location of pixels
Record；
If in the non-Reception learning first of the list item of the location of pixels, being iterated study to the location of pixels rear list item,
Until the list item of iteration to Reception learning first.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611122462.6A CN106529609B (en) | 2016-12-08 | 2016-12-08 | A kind of image-recognizing method and device based on neural network structure |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611122462.6A CN106529609B (en) | 2016-12-08 | 2016-12-08 | A kind of image-recognizing method and device based on neural network structure |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106529609A true CN106529609A (en) | 2017-03-22 |
| CN106529609B CN106529609B (en) | 2019-11-01 |
Family
ID=58342338
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611122462.6A Active CN106529609B (en) | 2016-12-08 | 2016-12-08 | A kind of image-recognizing method and device based on neural network structure |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106529609B (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108965687A (en) * | 2017-05-22 | 2018-12-07 | 阿里巴巴集团控股有限公司 | Shooting direction recognition methods, server and monitoring method, system and picture pick-up device |
| CN109285539A (en) * | 2018-11-28 | 2019-01-29 | 中国电子科技集团公司第四十七研究所 | A kind of sound identification method neural network based |
| WO2019076095A1 (en) * | 2017-10-20 | 2019-04-25 | 上海寒武纪信息科技有限公司 | Processing method and apparatus |
| CN109697507A (en) * | 2017-10-24 | 2019-04-30 | 上海寒武纪信息科技有限公司 | Processing method and processing device |
| CN110162757A (en) * | 2019-04-29 | 2019-08-23 | 北京百度网讯科技有限公司 | A kind of tableau format extracting method and system |
| KR20190104406A (en) * | 2017-10-20 | 2019-09-09 | 상하이 캠브리콘 인포메이션 테크놀로지 컴퍼니 리미티드 | Treatment method and device |
| CN112149676A (en) * | 2020-09-11 | 2020-12-29 | 中国铁道科学研究院集团有限公司 | Small target detection processing method for railway goods loading state image |
| CN114022357A (en) * | 2021-10-29 | 2022-02-08 | 北京百度网讯科技有限公司 | Image reconstruction method, training method, device and equipment of image reconstruction model |
| CN114885094A (en) * | 2022-03-25 | 2022-08-09 | 北京旷视科技有限公司 | Image processing method, image processor, image processing module and equipment |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| AU3394193A (en) * | 1992-03-13 | 1993-09-16 | Pilkington Micro-Electronics Limited | Improved artificial digital neuron, neural network and network training algorithm |
| BE1011273A4 (en) * | 1997-07-11 | 1999-07-06 | Euresys Sa | Process and device for on-line recognition of handwritten characters |
| CN1313975A (en) * | 1998-06-23 | 2001-09-19 | 英泰利克斯公司 | N-tuple or ram based neural network classification system and method |
| CN1625755A (en) * | 2002-04-12 | 2005-06-08 | 科学、技术与研究机构 | Robust face registration via multi-face prototypes synthesis |
| US20080204832A1 (en) * | 2007-02-27 | 2008-08-28 | Canon Kabushiki Kaisha | Constructing a color transform using a neural network for colors outside the spectrum locus |
| CN101727472A (en) * | 2008-10-21 | 2010-06-09 | 联发科技股份有限公司 | Image recognizing system and image recognizing method |
| CN101916393A (en) * | 2010-07-14 | 2010-12-15 | 中国科学院半导体研究所 | Realized circuit of pulse coupled neural network with function of image segmentation |
| CN102262728A (en) * | 2011-07-28 | 2011-11-30 | 电子科技大学 | Road traffic sign identification method |
| JP5113810B2 (en) * | 2009-08-07 | 2013-01-09 | 日本電信電話株式会社 | Image processing method, image processing apparatus, and crack detection system |
| US20160148078A1 (en) * | 2014-11-20 | 2016-05-26 | Adobe Systems Incorporated | Convolutional Neural Network Using a Binarized Convolution Layer |
-
2016
- 2016-12-08 CN CN201611122462.6A patent/CN106529609B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| AU3394193A (en) * | 1992-03-13 | 1993-09-16 | Pilkington Micro-Electronics Limited | Improved artificial digital neuron, neural network and network training algorithm |
| BE1011273A4 (en) * | 1997-07-11 | 1999-07-06 | Euresys Sa | Process and device for on-line recognition of handwritten characters |
| CN1313975A (en) * | 1998-06-23 | 2001-09-19 | 英泰利克斯公司 | N-tuple or ram based neural network classification system and method |
| CN1625755A (en) * | 2002-04-12 | 2005-06-08 | 科学、技术与研究机构 | Robust face registration via multi-face prototypes synthesis |
| US20080204832A1 (en) * | 2007-02-27 | 2008-08-28 | Canon Kabushiki Kaisha | Constructing a color transform using a neural network for colors outside the spectrum locus |
| CN101727472A (en) * | 2008-10-21 | 2010-06-09 | 联发科技股份有限公司 | Image recognizing system and image recognizing method |
| JP5113810B2 (en) * | 2009-08-07 | 2013-01-09 | 日本電信電話株式会社 | Image processing method, image processing apparatus, and crack detection system |
| CN101916393A (en) * | 2010-07-14 | 2010-12-15 | 中国科学院半导体研究所 | Realized circuit of pulse coupled neural network with function of image segmentation |
| CN102262728A (en) * | 2011-07-28 | 2011-11-30 | 电子科技大学 | Road traffic sign identification method |
| US20160148078A1 (en) * | 2014-11-20 | 2016-05-26 | Adobe Systems Incorporated | Convolutional Neural Network Using a Binarized Convolution Layer |
Non-Patent Citations (3)
| Title |
|---|
| LUCIANE Y 等: "dynamic zoning selection for handwritten character recognition", 《PROGRESS IN PATTERN RECOGNITION, IMAGE ANALYSIS, COMPUTER VISION, AND APPLICATIONS》 * |
| 孔月萍 等: "查找表与神经网络相结合的图像逆半调算法", 《计算机工程与科学》 * |
| 王巍 等: "基于ＣＮＮ的红外图像边缘检测算法的FPGA实现", 《光子学报》 * |
Cited By (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108965687A (en) * | 2017-05-22 | 2018-12-07 | 阿里巴巴集团控股有限公司 | Shooting direction recognition methods, server and monitoring method, system and picture pick-up device |
| CN108965687B (en) * | 2017-05-22 | 2021-01-29 | 阿里巴巴集团控股有限公司 | Shooting direction identification method, server, monitoring method, monitoring system and camera equipment |
| US10949995B2 (en) | 2017-05-22 | 2021-03-16 | Alibaba Group Holding Limited | Image capture direction recognition method and server, surveillance method and system and image capture device |
| WO2019076095A1 (en) * | 2017-10-20 | 2019-04-25 | 上海寒武纪信息科技有限公司 | Processing method and apparatus |
| KR20190104406A (en) * | 2017-10-20 | 2019-09-09 | 상하이 캠브리콘 인포메이션 테크놀로지 컴퍼니 리미티드 | Treatment method and device |
| KR102434726B1 (en) | 2017-10-20 | 2022-08-19 | 상하이 캠브리콘 인포메이션 테크놀로지 컴퍼니 리미티드 | Treatment method and device |
| CN109697507A (en) * | 2017-10-24 | 2019-04-30 | 上海寒武纪信息科技有限公司 | Processing method and processing device |
| CN109697507B (en) * | 2017-10-24 | 2020-12-25 | 安徽寒武纪信息科技有限公司 | Processing method and device |
| CN109285539B (en) * | 2018-11-28 | 2022-07-05 | 中国电子科技集团公司第四十七研究所 | Sound recognition method based on neural network |
| CN109285539A (en) * | 2018-11-28 | 2019-01-29 | 中国电子科技集团公司第四十七研究所 | A kind of sound identification method neural network based |
| CN110162757A (en) * | 2019-04-29 | 2019-08-23 | 北京百度网讯科技有限公司 | A kind of tableau format extracting method and system |
| CN110162757B (en) * | 2019-04-29 | 2023-08-18 | 北京百度网讯科技有限公司 | Table structure extraction method and system |
| CN112149676A (en) * | 2020-09-11 | 2020-12-29 | 中国铁道科学研究院集团有限公司 | Small target detection processing method for railway goods loading state image |
| CN112149676B (en) * | 2020-09-11 | 2024-04-30 | 中国铁道科学研究院集团有限公司 | Small target detection processing method for railway cargo loading state image |
| CN114022357A (en) * | 2021-10-29 | 2022-02-08 | 北京百度网讯科技有限公司 | Image reconstruction method, training method, device and equipment of image reconstruction model |
| CN114885094A (en) * | 2022-03-25 | 2022-08-09 | 北京旷视科技有限公司 | Image processing method, image processor, image processing module and equipment |
| CN114885094B (en) * | 2022-03-25 | 2024-03-29 | 北京旷视科技有限公司 | Image processing method, image processor, image processing module and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106529609B (en) | 2019-11-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106529609A (en) | Image recognition method and device based on neural network structure | |
| CN109359539B (en) | Attention assessment method and device, terminal equipment and computer readable storage medium | |
| CN110009027B (en) | Image comparison method and device, storage medium and electronic device | |
| US11080553B2 (en) | Image search method and apparatus | |
| CN110111248A (en) | A kind of image split-joint method based on characteristic point, virtual reality system, camera | |
| CN106200960A (en) | The content display method of electronic interactive product and device | |
| CN106202529A (en) | A kind of page rendering method and device | |
| CN111309222B (en) | Sliding block notch positioning and dragging track generation method for sliding block verification code | |
| CN110133443B (en) | Power transmission line component detection method, system and device based on parallel vision | |
| CN112257665A (en) | Image content recognition method, image recognition model training method, and medium | |
| CN111310156B (en) | Automatic identification method and system for slider verification code | |
| CN103679792B (en) | The rendering intent and system of a kind of threedimensional model | |
| CN111310155B (en) | System architecture for automatic identification of slider verification code and implementation method | |
| CN110135318A (en) | Cross determination method, apparatus, equipment and the storage medium of vehicle record | |
| CN113505854A (en) | Method, device, equipment and medium for constructing facial image quality evaluation model | |
| KR20200136723A (en) | Method and apparatus for generating learning data for object recognition using virtual city model | |
| CN114972646B (en) | Method and system for extracting and modifying independent ground objects of live-action three-dimensional model | |
| CN109816001A (en) | A kind of more attribute recognition approaches of vehicle based on deep learning, device and equipment | |
| CN104900094A (en) | Method and device for realizing writing simulation in terminal | |
| CN110472600A (en) | The identification of eyeground figure and its training method, device, equipment and storage medium | |
| WO2022160406A1 (en) | Implementation method and system for internet of things practical training system based on augmented reality technology | |
| CN108961268A (en) | A kind of notable figure calculation method and relevant apparatus | |
| CN115953330B (en) | Texture optimization method, device, equipment and storage medium for virtual scene image | |
| US20230206573A1 (en) | Method of learning a target object by detecting an edge from a digital model of the target object and setting sample points, and method of augmenting a virtual model on a real object implementing the target object using the learning method | |
| CN110349108A (en) | Handle method, apparatus, electronic equipment and the storage medium of image |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | 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 |