CN109711399A - Shop recognition methods based on image, device, electronic equipment - Google Patents
Shop recognition methods based on image, device, electronic equipment Download PDFInfo
- Publication number
- CN109711399A CN109711399A CN201811309281.3A CN201811309281A CN109711399A CN 109711399 A CN109711399 A CN 109711399A CN 201811309281 A CN201811309281 A CN 201811309281A CN 109711399 A CN109711399 A CN 109711399A
- Authority
- CN
- China
- Prior art keywords
- image
- shop
- target board
- sub
- board
- 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
- 238000000034 method Methods 0.000 title claims abstract description 64
- 238000004364 calculation method Methods 0.000 claims description 14
- 230000004927 fusion Effects 0.000 claims description 14
- 238000003062 neural network model Methods 0.000 claims description 11
- 238000004590 computer program Methods 0.000 claims description 8
- 239000011159 matrix material Substances 0.000 claims description 4
- 230000004048 modification Effects 0.000 claims description 2
- 238000012986 modification Methods 0.000 claims description 2
- 238000005516 engineering process Methods 0.000 abstract description 3
- 238000012549 training Methods 0.000 description 18
- 230000000694 effects Effects 0.000 description 6
- 238000001514 detection method Methods 0.000 description 5
- 230000008569 process Effects 0.000 description 5
- 230000007812 deficiency Effects 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000004422 calculation algorithm Methods 0.000 description 2
- 238000005286 illumination Methods 0.000 description 2
- 238000009434 installation Methods 0.000 description 2
- 238000002372 labelling Methods 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 230000001174 ascending effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000001149 cognitive effect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 239000004744 fabric Substances 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000001373 regressive effect Effects 0.000 description 1
- 230000008439 repair process Effects 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Landscapes
- Image Analysis (AREA)
Abstract
The shop recognition methods based on image that this application discloses a kind of, belongs to field of computer technology, for solving the problems, such as that the accuracy rate for carrying out shop identification based on image is low.Shop recognition methods based on image disclosed in the embodiment of the present application includes: to obtain the global image feature of target board image, and obtain the subregion significant characteristics of the target board image；The characteristics of image of the target board image is determined according to the global image feature and subregion significant characteristics；According to the characteristics of image of the target board image and the determining shop with the target board images match of the shop characteristics of image in default shop.The characteristics of image that shop recognition methods as disclosed in the embodiment of the present application based on image uses, the global characteristics of image and the fine granularity region of part are taken into account, so that characteristics of image is both comprehensive, has and have discrimination, the accuracy rate for carrying out shop identification based on image can be improved.
Description
Technical field
This application involves field of computer technology, more particularly to a kind of shop recognition methods based on image, device, electricity
Sub- equipment.
Background technique
When in the prior art, by the shop in the picture search shop of shop board or identification image, it can pass through
Aspect indexing is established to entire signboard region, that is, considers character information, textural characteristics and board the content distribution letter of image
Breath, can have preferable robustness.For example, notification number is the Chinese patent application of CN104598885B, Scale invariant is used
Feature Conversion (SIFT, Scale-invariant feature transform), to detect or describe the spy of the locality in image
Sign carries out image recognition in conjunction with HS characteristic component.For neglect angular transformation and the lesser image of illumination scene changes have than
Preferable robustness.However, the characteristics of image extracted in above-mentioned shop recognition methods in the prior art for acquisition angles and
The identification robustness of the biggish image of illumination scene changes is smaller, also, inadequate for the feature representation of shop board image
It is clear and complete, so that the accuracy rate of shop identification is not high.
In conclusion the shop recognition methods based on image at least has that recognition accuracy is not high to ask in the prior art
Topic.
Summary of the invention
The application provides a kind of shop recognition methods based on image, helps to improve and carries out shop identification based on image
Accuracy rate.
To solve the above-mentioned problems, in a first aspect, the embodiment of the present application provides a kind of shop identification side based on image
Method, comprising:
The global image feature of target board image is obtained, and obtains the subregion conspicuousness of the target board image
Feature；
The global image feature and subregion significant characteristics determine the characteristics of image of the target board image；
According to the characteristics of image of the target board image and the determination of the shop characteristics of image in default shop and the target
The shop of board images match.
Second aspect, the embodiment of the present application provide a kind of shop identification device based on image, comprising:
Feature obtains module, for obtaining the global image feature of target board image, and the acquisition target board
The subregion significant characteristics of image；
Fusion Features module, for determining the target board according to the global image feature and subregion significant characteristics
The characteristics of image of plaque image；
Match cognization module, for special according to the characteristics of image of the target board image and the shop image in default shop
The determining shop with the target board images match of sign.
The third aspect, the embodiment of the present application also disclose a kind of electronic equipment, including memory, processor and are stored in institute
The computer program that can be run on memory and on a processor is stated, the processor realizes this when executing the computer program
Apply for the shop recognition methods based on image described in embodiment.
Fourth aspect, the embodiment of the present application provide a kind of computer readable storage medium, are stored thereon with computer journey
The step of sequence, the disclosed shop based on image of the embodiment of the present application identifies when which is executed by processor.
Shop recognition methods based on image disclosed in the embodiment of the present application is schemed by obtaining the global of target board image
As feature, and obtain the subregion significant characteristics of the target board image；According to the global image feature and subregion
Domain significant characteristics determine the characteristics of image of the target board image；According to the characteristics of image of the target board image and in advance
If the determining shop with the target board images match of the shop characteristics of image in shop helps to solve to carry out shop based on image
Spread the low problem of the accuracy rate of identification.The overall situation that shop recognition methods as disclosed in the embodiment of the present application based on image uses
Characteristics of image reflects that the appearance difference of board and overall effect are more comprehensive, but identifies for fine granularity, and can ignore some has area
Indexing, but tiny feature, and subregion significant characteristics are then more fine granularity local features, can overcome the disadvantages that global image feature
Deficiency, therefore, after above two Fusion Features, taken into account image global characteristics and part fine granularity region so that
Characteristics of image was both comprehensive, has and has discrimination, and the accuracy rate that shop identification is carried out based on image can be improved.
Detailed description of the invention
Technical solution in ord to more clearly illustrate embodiments of the present application, below will be in embodiment or description of the prior art
Required attached drawing is briefly described, it should be apparent that, the accompanying drawings in the following description is only some realities of the application
Example is applied, it for those of ordinary skill in the art, without any creative labor, can also be attached according to these
Figure obtains other attached drawings.
Fig. 1 is the shop recognition methods flow chart based on image of the embodiment of the present application one；
Fig. 2 is the shop recognition methods flow chart based on image of the embodiment of the present application two；
Fig. 3 is one of shop identification device structural schematic diagram based on image of the embodiment of the present application three；
Fig. 4 is the shop identification device second structural representation based on image of the embodiment of the present application three；
Fig. 5 is the shop identification device third structural representation based on image of the embodiment of the present application three.
Specific embodiment
Below in conjunction with the attached drawing in the embodiment of the present application, technical solutions in the embodiments of the present application carries out clear, complete
Site preparation description, it is clear that described embodiment is some embodiments of the present application, instead of all the embodiments.Based on this Shen
Please in embodiment, every other implementation obtained by those of ordinary skill in the art without making creative efforts
Example, shall fall in the protection scope of this application.
Embodiment one
A kind of shop recognition methods based on image disclosed in the present embodiment, as shown in Figure 1, this method comprises: step 110
To step 130.
Step 110, the global image feature of target board image is obtained, and obtains the subregion of the target board image
Domain significant characteristics.
In the application scenarios that the image based on shop board carries out shop identification, it is necessary first to training neural network mould
Type, for extracting the global image feature of target board image.In the embodiment of the present application, the global image is characterized in for table
The high dimensional feature of the appearance difference of target board in the target board image is levied, such as the content of the target board image point
Cloth, global color distributed intelligence, marginal information etc..In some embodiments of the present application, preparatory training convolutional mind can be passed through
The global image feature of target board image is extracted through network model.
Subregion significant characteristics described in the embodiment of the present application are the aobvious of each region divided in target board image
Characteristics.The significant characteristics are the Color Distribution Features extracted in marking area in image, and the marking area is
User interest can most be caused in image, can most show the region of picture material.Although the area-of-interest of people be it is very subjective,
And due to the difference of the knowledge background of people and specific business need, for same piece image, different users may be selected
Different regions are as area-of-interest.But due to the general character of human visual system and attention mechanism, and make some in image
Region total energy is significantly attracting note that these regions often information rich in.It therefore can be according to human vision system
The characteristics of system, is approximatively judged in image using the universal law during human cognitive by certain low-level image features of image
Marking area, significant characteristics are exactly the Color Distribution Features extracted in this marking area.When it is implemented, can lead to
The space length and color distance crossed between each image-region that image divides determine the subregion significant characteristics of image.
Step 120, the target board image is determined according to the global image feature and subregion significant characteristics
Characteristics of image.
In some embodiments of the present application, can by by the global image feature and subregion significant characteristics into
The fusion feature of higher-dimension global image feature and more fine-grained subregion significant characteristics has been merged in row splicing, and will
Obtained fusion feature is determined as the characteristics of image of the target board image.
Step 130, according to the characteristics of image of the target board image and the shop characteristics of image in default shop determine with
The shop of the target board images match.
Then, by matching the shop characteristics of image in described image feature and default shop.For example, passing through calculating
Similarity between the characteristics of image of the target board image and the shop characteristics of image in each default shop, then, according to similar
Spend size or preset similarity threshold, determine meet the corresponding shop of similarity of preset condition as with the target board
The shop of plaque images match.
Shop recognition methods based on image disclosed in the embodiment of the present application is schemed by obtaining the global of target board image
As feature, and obtain the subregion significant characteristics of the target board image；According to the global image feature and subregion
Domain significant characteristics determine the characteristics of image of the target board image；According to the characteristics of image of the target board image and in advance
If the determining shop with the target board images match of the shop characteristics of image in shop helps to solve to carry out shop based on image
Spread the low problem of the accuracy rate of identification.The overall situation that shop recognition methods as disclosed in the embodiment of the present application based on image uses
Characteristics of image reflects that the appearance difference of board and overall effect are more comprehensive, but identifies for fine granularity, and can ignore some has area
Indexing, but tiny feature, and subregion significant characteristics are then more fine granularity local features, can overcome the disadvantages that global image feature
Deficiency, therefore, after above two Fusion Features, taken into account image global characteristics and part fine granularity region so that
Characteristics of image was both comprehensive, has and has discrimination, and the accuracy rate that shop identification is carried out based on image can be improved.
Embodiment two
A kind of shop recognition methods based on image disclosed in the present embodiment, as shown in Figure 2, comprising: step 210 to step
270。
Step 210, training neural network model.
It, can be using Darknet53 network as basic network, using Coupled in some embodiments of the present application
Clusters Loss (coupled-cluster loss) compares training, to reduce feature inter- object distance, increases feature difference between class, with
The board characteristics of image with discrimination is extracted, finally, it is extended to the vector of designated length (such as 1000), training neural network mould
Type.The training method of neural network model is repeated no more in the present embodiment referring to the prior art.
In some embodiments of the present application, further includes: be based on YOLO object detection algorithms, training shop board detects mould
Type, to obtain the board image in input picture by the shop board detection model.
The core concept of YOLO (You Only Look Once) is exactly directly to exist using whole figure as the input of network
Classification belonging to the position of output layer regressive object frame and target frame.The implementation method of YOLO is: piece image is divided into SxS
Grid, if the center of some target is fallen in this grid, this grid is just responsible for predicting this target.Each grid is wanted
Predict B target frame, each target frame will also be attached to other than returning the position of itself and predict a confidence value.This
It is how quasi- double that a confidence value represents the confidence level containing target and having for this target frame prediction in predicted target frame
Information.
When it is implemented, first having to building shop board detection model.For example, main structure uses YOLO v3 algorithm.
The training process of shop board detection model is as follows.
Firstly, extracting the global image feature of shop board image.When it is implemented, according to shop board image labeling text
Part, cluster obtain 9 groups of anchor (anchor point) numerical value, image high dimensional feature are extracted using Darknet-53 basic network, as board
The feature foundation of plaque positioning classification.
Then, signboard target is predicted.Image-region is divided using the method for grid dividing, to different images regional prediction,
Fusion different dimensions feature is predicted that prospect, background, structure use 3 scales, is distributed according to anchor cluster centre, often
3 signboard targets of a scale prediction, mesoscale 1 are addition convolutional layer output prediction block coordinate after basic network；Scale 2 be from
The convolutional layer of layer second from the bottom in scale 1 up-samples, then is added with the last layer 16*16 characteristic pattern, again by multiple volumes
Prediction block information is exported after product, scale 2 becomes larger twice compared to scale 1；Scale 3 is similar with scale 2, uses the spy of 32*32 size
Sign figure.
Finally, returning signboard coordinates of targets.By the target area of anchor regression forecasting prospect, using logic, this spy is returned
Return mode, coordinate regression formula are as follows:
In formula, (bx,by,bw,bh) be final prediction block center point coordinate value and width
Height, (tx,ty,tw,th) be e-learning prediction block center point coordinate value and wide high, cx、cyCross, ordinate for grid
Offset, pw、phFor the side length of anchor combination；σ is the formula of this spy's recurrence of logic.
Step 220, it determines the shop characteristics of image in default shop, and constructs shop database.
Wherein, the default shop refers to include shop in the database of shop；The shop database includes at least pre-
If characteristics of image corresponding store information in shop described in the shop characteristics of image in shop and every group, the store information can be with
Including information such as shop title, geographical location, board images.The global image that the shop characteristics of image includes at least shop is special
It seeks peace subregion significant characteristics.
When it is implemented, can be extracted in the shop data by the neural network model of training in abovementioned steps
The global image feature in each default shop.
For each default shop, the subregion significant characteristics in the shop are obtained, comprising: by the board in the shop
Image is divided according at least one dividing method, determines several image-regions；For each described image region, according to face
Color distribution is split described image region, determines at least one sub-image area that described image region includes；For every
A described image region, the space length and color distance between sub-image area for including by described image region determine
The subregion significant characteristics in described image region；By the subregion significant characteristics of several image-regions, group is combined into institute
State the subregion significant characteristics in shop.
Further, the space length between the sub-image area for including by described image region and color away from
From determining the subregion significant characteristics in described image region, comprising: the sub-image area for including by described image region it
Between space length and color distance, determine the conspicuousness of each sub-image area；It is highest default according to the conspicuousness
The conspicuousness of the sub-image area of quantity determines that the subregion conspicuousness of each affiliated image-region in the sub-image area is special
Sign.
Space in some embodiments of the present application, between the sub-image area for including by described image region
Distance and color distance, the step of determining the conspicuousness of each sub-image area, comprising: pass through formulaEach subgraph that described image region includes is determined respectively
Region rkConspicuousness, wherein riIt indicates to be different from rkSub-image area, Ds(rk,ri) indicate sub-image area rkAnd riSky
Between distance, σsRepresentation space is apart from weight, ω (ri) indicate sub-image area riWeight, Dr(rk,ri) it is sub-image area
rkAnd riColor distance, Dr(rk, ri) calculation formula are as follows:
Wherein, f (cI, v) indicate v-th
Color ci,vIn i-th of sub-image area riIn all niThe probability occurred in kind color, f (cK, u) indicate u-th of color ck,u
In k-th of sub-image area rkIn all nkThe probability occurred in kind color, D (ck,u,ci,v) it is color ck,uWith color ci,v?
The color distance in the space L*a*b is measured.
It determines the technical solution of the subregion significant characteristics in each default shop, and determines mesh in the identification process of shop
The technical solution of the subregion significant characteristics of label plaque is identical, specifically describes referring to point for determining target board in subsequent step
The description of region significance feature, details are not described herein again.
In some embodiments of the present application, for each shop, by by the global image of the board image in the shop
Feature and subregion significant characteristics are spliced, and the shop characteristics of image in the shop can be obtained.The shop image in shop is special
The acquisition methods of sign are identical as the characteristics of image acquisition methods of target board image, referring specifically to being described below.
Step 230, the global image feature of target board image is obtained, and obtains the subregion of the target board image
Domain significant characteristics.
It further include other if including target board image in image to be identified in some embodiments of the present application
Background information, it may for example comprise multiple board information or comprising storefront information then pass through the board of training in signature step first
Detection model obtains the target board image in input picture, then, further obtains the global image of target board image
The subregion significant characteristics of feature and the target board image.
When it is implemented, the global image feature for obtaining target board image, comprising: pass through nerve trained in advance
Network model obtains the global image feature of the target board image.Pass through the neural network model of training in abovementioned steps
When obtaining the global image feature of target board image, the target board image is input to the neural network model, and
Using the output data of the neural network model specific characteristic presentation layer as the global image feature of the target board image.
Since the neural network model has fully considered the appearance difference of training sample image during training, such as target
The information such as content distribution, global color distributed intelligence, the marginal information of board image, therefore, the neural network model it is each
The output data of feature representation layer is the high dimensional feature for being able to reflect image Global Information.
In some embodiments of the present application, the subregion significant characteristics for obtaining the target board image, packet
It includes: the target board image being divided according at least one dividing method, determines several image-regions；For each institute
Image-region is stated, described image region is split according to distribution of color, determines at least one that described image region includes
Sub-image area；For each described image region, the space between sub-image area that includes by described image region away from
From and color distance, determine the subregion significant characteristics in described image region.
For example, target board image trisection in the horizontal direction is obtained 3 image-regions；By the target board
Plaque image trisection along the vertical direction, obtains 3 image-regions；By the target board image six equal part in the horizontal direction, obtain
To 6 image-regions；By the target board image six equal part along the vertical direction, 6 image-regions are obtained.According to above-mentioned side
Method can determine 3+3+6+6=18 image-region.
Then, for each of 18 image-regions image-region, further according to distribution of color to described image
Region is split, and determines at least one sub-image area that each image-region includes.It falls into a trap for example, calculating each image-region
Calculate the side dissmilarity degree of each pixel and its 8 neighborhood or 4 neighborhood territory pixels point, i.e. each pixel and its 8 neighborhood or 4 neighbours
The channel rgb color distance between the pixel of domain；Then, according to the ascending arrangement of side dissmilarity degree, edge sequence [e is obtained1, e2…
en]；Later, processing sequentially is merged to the side in line set；Determine that described image region is wrapped by the side after merging treatment
The sub-image area included.Specific merging treatment process can be with are as follows: to the side e currently selectedj, since its latter side, into
Row merges judgement, if the side e of selectionjThe vertex connected is (vi, vj), the two vertex are not belonging to same image-region, if
Side ejDissimilar degree be less than or equal to given threshold wTH, wherein wTH=min (Ci,Cj), i.e. two affiliated image-regions in vertex
Internal dissimilar degree minimum value, i.e. wI, j≤ min then continues to merge processing to rear a line, otherwise, updates the setting
Threshold value is wTH=wi,j+1/Ci+Cj, and by the Vertex Labeling vjIt is updated to vi, i.e., two sides are merged.If there are also do not merge place
The side of reason then continues to merge processing to rear a line, until all sides have all been handled.Finally, according to merging treatment
Each image-region is divided at least one sub-image area by side afterwards respectively.
In the other embodiments of the application, each image-region can also be divided using other methods in the prior art
It is not divided at least one sub-image area, is no longer enumerated in the application.
Later, for each described image region in above-mentioned 18 image-regions, include by described image region
Space length and color distance between sub-image area determine the subregion significant characteristics in described image region.
Space in some embodiments of the present application, between the sub-image area for including by described image region
Distance and color distance, determine the subregion significant characteristics in described image region, comprising: include by described image region
Space length and color distance between sub-image area determine the conspicuousness of each sub-image area；According to described significant
Property highest preset quantity the sub-image area conspicuousness, determine point of each affiliated image-region in the sub-image area
Region significance feature.
Space in some embodiments of the present application, between the sub-image area for including by described image region
Distance and color distance, the step of determining the conspicuousness of each sub-image area, comprising: pass through formulaEach subgraph that described image region includes is determined respectively
Region rkConspicuousness, wherein riIt indicates to be different from rkSub-image area, Ds(rk, ri) indicate sub-image area rkAnd riSky
Between distance, σsRepresentation space is apart from weight, ω (ri) indicate sub-image area riWeight, Dr(rk,ri) it is sub-image area
rkAnd riColor distance, Dr(rk, ri) calculation formula are as follows:Wherein, f (ci,v) indicate v-th of color ci,v
In i-th of sub-image area riIn all niThe probability occurred in kind color, f (ck,u) indicate u-th of color ck,uAt k-th
Sub-image area rkIn all nkThe probability occurred in kind color, D (cK, u,ci,v) it is color ck,uWith color ci,vIn L*a*b sky
Between color distance measurement.The circular of color distance measurement is the prior art, and this embodiment is not repeated.
Assuming that a certain image-region includes 5 sub-image areas, respectively indicate are as follows: r1,r2,r3,r4,r5, with rk=r1It lifts
The determination method of the subregion significant characteristics of example one image-region of explanation.Firstly, calculating separately sub-image area r1With it
The space length D of his sub-image areas(r1,ri) and color distance Dr(r1,ri), 2≤i≤4；Then, pass through formulaDetermining sub-image area r1Conspicuousness.According to above-mentioned side
Method can determine sub-image area r respectively1, r2, r3, r4,r5Conspicuousness.Finally, the whole subgraphs for including by the image-region
As region, such as the sub-image area r in this example1, r2, r3, r4, r5Conspicuousness be stitched together, obtain the subregion of the image-region
Domain significant characteristics；Alternatively, the sub-image area for including by the image-region, such as the sub-image area r in this example1, r2, r3,
r4,r5Preset quantity maximum conspicuousness (such as r1,r2,r3Conspicuousness) be stitched together, obtain the subregion of the image-region
Significant characteristics.
When it is implemented, in the conspicuousness calculation formula of sub-image area, σsWeight for controlling space length is strong
Degree, σsValue is bigger, and the influence of space length weight is smaller, causes the contrast compared with far region can be to current region conspicuousness shadow
Sound is larger,Value interval be (0,1].In embodiments herein, it is preferred thatValue is 0.4.ω(ri) it is subgraph
As region riWith sub-image area rkColor distance weight, value is usually the equal proportion number of distance between sub-image area
Value.Still to there is 5 sub-image area r in image-region figure1,r2, r3,r4,r5For, if r1With the Europe of other sub-image areas
Formula distance is respectively [20,30,10,40], then color distance weight ω (ri) it can be [0.2,0.3,0.1,0.4].
According to the method described above, the subregion significant characteristics of each image-region can be determined.
Step 240, the target board image is determined according to the global image feature and subregion significant characteristics
Characteristics of image.
Further, the target board image is determined according to the global image feature and subregion significant characteristics
Characteristics of image, comprising: being obtained after being spliced the global image feature and subregion significant characteristics as a result, as institute
State the characteristics of image of target board image.For example, can by by the global image feature and subregion significant characteristics into
The fusion feature of higher-dimension global image feature and more fine-grained subregion significant characteristics has been merged in row splicing, and will
Obtained fusion feature is determined as the characteristics of image of the target board image.
Step 250, by K-means++ Clustering Model trained in advance, the figure of the target board image is corrected
As feature, to retain representational feature in described image feature.
In some embodiments of the present application, the characteristics of image according to the target board image and default shop
Before the determining shop with the target board images match of shop characteristics of image, further includes: pass through K-means trained in advance
++ Clustering Model corrects the described image feature of the target board image, representational in described image feature to retain
Feature.
When it is implemented, the K-means++ Clustering Model is special according to the shop image in preset shop database
Sign training obtains.
To include M shop in the database of shop, the shop characteristics of image in each shop is the global figure that size is 1000
As the subregion significant characteristics composition that feature and size are 1000, then the shop characteristics of image in each shop is length 2000
Characteristic, using the data of the micro- M*2000 of the size in M shop as input；Output is M cluster centre and size is M*
1000 characteristic.K-means++ cluster training is carried out according to above-mentioned data, to obtain K-means++ Clustering Model.Its
In, the size of output is the revised shop characteristics of image of M*1000.Specific training process are as follows:
The first step randomly chooses at one o'clock as first cluster centre from the set of data points of input；
Second step calculates it at a distance from cluster centre for each of data set point；
Third step selects one to put apart from maximum as new cluster centre；
4th step repeats second and third step, until M cluster centre is selected；
Each data point is assigned to nearest mass center for M cluster centre of generation by the 5th step, forms M cluster；
6th step recalculates the mass center of each cluster；
7th step repeats the 5th step and the 6th step, until cluster does not change or reach maximum number of iterations, completes
The training of K-means++ Clustering Model.
After the training of K-means++ Clustering Model is completed, by the K-means++ Clustering Model to the target board
The described image feature of image, the feature if size is 2000 are modified, and the characteristics of image that size is 1000 can be obtained.
In some embodiments of the present application, the characteristics of image according to the target board image and default shop
Before the determining shop with the target board images match of shop characteristics of image, further includes: believed according to specified geographical location
Breath determines default shop.
When it is implemented, comparison efficiency is promoted in order to which drawdown ratio is to range, it can be first to the shop in the database of shop
Preliminary screening is carried out based on geographical location information, by the geographical location information of geographical location information and the target board image
The shop matched is as default shop.
Step 260, according to the characteristics of image of the target board image and the shop characteristics of image in default shop determine with
The shop of the target board images match.
By the processing of abovementioned steps, the shop in each shop in the characteristics of image and shop database of target board image
Characteristics of image has same size, then can be directly by calculating the characteristics of image of target board image and the shop in each shop
Similarity between characteristics of image is matched to carry out characteristics of image.When it is implemented, can according to similarity by greatly to
Small sequence arranges the shop, to export the shop with the target board images match.
Step 270, the characteristics of image to the target board image and with the shop of the target board images match
Shop characteristics of image is updated by the shop characteristics of image in respective similar shop respectively, and according to result weight after update
Newly matched.
In some embodiments of the present application, the characteristics of image according to the target board image and default shop
After the determining shop with the target board images match of shop characteristics of image, further includes: by with the target board figure
The shop characteristics of image average value in the shop as described in highest default first quantity of matching degree in matched shop, updates institute
State the described image feature of target board image；Matching degree highest in the determining shop with the target board images match respectively
Default second quantity described in shop, the similar shop in the default shop；For shop described in default second quantity, lead to
The shop characteristics of image average value for meeting the similar shop of preset condition in the respective similar shop in the shop is crossed,
Update the shop characteristics of image in the shop；According to the updated characteristics of image of the target board image, the default shop
The updated shop characteristics of image of paving determines the distance that reorders of the target board image Yu the default shop；According to institute
The distance that reorders is stated, the shop with the target board images match is redefined.
When it is implemented, the shop characteristics of image in the shop can also first be updated, then update the target board image
Described image feature, the application is to the described image feature for updating the target board image and updates the shop in the shop
The sequence of characteristics of image is without limitation.
In some embodiments of the present application, it is assumed that the described image feature of the target board image is que, shop number
According in library include 100 shops, the matching result of the target board image and default shop are as follows: [reg1,reg2…reg100],
Wherein, reg1,reg2…reg100Indicate the shop characteristics of image in shop in database.Then can by with the target board figure
As in matched shop matching degree it is highest before preset the first quantity shop the shop characteristics of image average value, update institute
State the described image feature of target board image.For example, the described image feature que of the target board image is updated to que
'=reg1+…+reg4/4。
It is similar, firstly, matching degree highest default the in the determining shop with the target board images match respectively
Two quantity (such as 3) described shop (such as shop 1,2,3), the similar shop in the shop database (except itself).It is assumed that
The similar shop in shop 1 include shop 7,8 ..., 85, the similar shop in shop 2 include shop 3,4 ..., 69, shop 3 it is similar
Shop include shop 4,6 ..., 67.Then, for above-mentioned 3 shops 1,2,3, respectively by meeting in respective similar shop
The shop characteristics of image average value in the similar shop (highest 4 shops of such as similarity) of preset condition, updates the shop
Shop characteristics of image.For the present embodiment, the shop characteristics of image in shop 1 can be updated to reg '1=reg7+…+
reg24/ 4, the shop characteristics of image in shop 2 can be updated to reg '2=reg3+…+reg24/ 4, the shop characteristics of image in shop 3
It can be updated to reg '3=reg4+…+reg24/4。
After updating, the shop characteristics of image in shop updates in the database of shop are as follows: [reg '1,reg‘2,reg‘3…
reg100]。
Finally, according to the updated characteristics of image of the target board image, the updated shop in the default shop
Paving characteristics of image determines the distance that reorders of the target board image Yu the default shop.
In some embodiments of the present application, the updated characteristics of image according to the target board image, institute
State default shop updated shop characteristics of image determine the target board image and the default shop reorder away from
From the step of, comprising:
Pass through formula d*(p,gi)=(1- λ) dJ(p,gi)+λ d (p, gi) determine the corresponding board p of the target board image
With the board g in the default shopiBetween the distance that reorders；Wherein, dJ(p,gi) indicate board p and board giBetween outstanding person
Card moral distance, calculation formula areIn formula,For board p characteristics of image and
Board gjThe neighbour of corresponding shop characteristics of image encodes distance vector；d(p,gi) indicate board p and board giBetween geneva
Distance, calculation formula are In formula, xpFor board p characteristics of image andFor board
Plaque gjCorresponding shop characteristics of image,For characteristics of image covariance matrix；λ is weight coefficient, according to specific
Business demand determines its value.
For example, taking xp=que ',Pass through above-mentioned formula, so that it may calculate target board image and shop
Spread the distance that reorders in shop 1 in database.And so on, shop in target board image and shop database can be calculated
2,3 ..., 100 distance that reorders.Later, according to reorder distance to shop in the database of shop be ranked up output to get
To the shop sequence with target board images match.
Shop recognition methods based on image disclosed in the embodiment of the present application is schemed by obtaining the global of target board image
As feature, and obtain the subregion significant characteristics of the target board image；According to the global image feature and subregion
Domain significant characteristics determine the characteristics of image of the target board image；By K-means++ Clustering Model trained in advance, repair
The described image feature of the just described target board image, to retain representational feature in described image feature, according to described
The characteristics of image of target board image and the determining shop with the target board images match of the shop characteristics of image in default shop
Paving, the accuracy rate for helping to solve the problems, such as to carry out shop identification based on image are low.It is based on as disclosed in the embodiment of the present application
The global image feature that the shop recognition methods of image uses, reflects that the appearance difference of board and overall effect are more comprehensive, still
Identified for fine granularity, can ignore it is some have a discrimination, but tiny feature, and subregion significant characteristics are then more particulates
Local feature is spent, the deficiency of global image feature is can overcome the disadvantages that, therefore, after above two Fusion Features, has taken into account the complete of image
The fine granularity region of office's feature and part has so that characteristics of image is both comprehensive and has discrimination, can be improved and is carried out based on image
The accuracy rate of shop identification.Meanwhile characteristics of image is modified by K-means++ Clustering Model, it can be into reservation image
Representative special war in feature, and filter out redundancy or do not have representative feature, to promote shop database
Storage efficiency and the efficiency that image recognition is promoted by the size of reduction characteristics of image.
Further, it only relies on characteristics of image and carries out the obtained recognition result of preliminary matches for indistinguishable sample searching
Effect is bad, for example, the result mismatched out may when target board image is the board image of " KFC " are as follows:
KFC, McDonald, Burger King ... shop recognition methods disclosed in the present application are the board of " KFC " to target board image
The characteristics of image of plaque image is updated by the shop image of KFC, McDonald, and, by " agreeing in the database of shop
The shop characteristics of image of other shops of De Ji " is updated the shop characteristics of image of " KFC ", then, then passes through update
Characteristics of image afterwards carries out match cognization, has drawn high KFC by the shop characteristics of image of other shops of " KFC " "
Sorting position can effectively promote the accuracy of images match.
Embodiment three
A kind of shop identification device based on image disclosed in the present embodiment, as shown in figure 3, described device includes:
Feature obtains module 310, for obtaining the global image feature of target board image, and the acquisition target board
The subregion significant characteristics of plaque image；
Fusion Features module 320, for determining the mesh according to the global image feature and subregion significant characteristics
The characteristics of image of label plaque image；
Match cognization module 330, for according to the characteristics of image of the target board image and the shop figure in default shop
As the determining shop with the target board images match of feature.
In some embodiments of the present application, as shown in figure 4, feature acquisition module 310 further comprises:
Fisrt feature acquisition submodule 3101, for obtaining the target board by neural network model trained in advance
The global image feature of plaque image.
The feature obtains module 310 and further comprises:
Second feature acquisition submodule 3102, for carrying out the target board image according at least one dividing method
It divides, determines several image-regions；
The second feature acquisition submodule 3102 is also used to for each described image region, according to distribution of color pair
Described image region is split, and determines at least one sub-image area that described image region includes；And
For each described image region, the space length between sub-image area that includes by described image region and
Color distance determines the subregion significant characteristics in described image region.
Space in some embodiments of the present application, between the sub-image area for including by described image region
Distance and color distance determine the subregion significant characteristics in described image region, comprising:
The space length and color distance between sub-image area for including by described image region, determine each son
The conspicuousness of image-region；
According to the conspicuousness of the sub-image area of the highest preset quantity of the conspicuousness, each subgraph is determined
The subregion significant characteristics of the affiliated image-region in region.
Optionally, the space length and color distance between the sub-image area for including by described image region,
Determine the conspicuousness of each sub-image area, comprising:
Pass through formulaDescribed image region is determined respectively
The each sub-image area r for includingkConspicuousness, wherein riIt indicates to be different from rkSub-image area, Ds(rk,ri) indicate son
Image-region rkAnd riSpace length, σsRepresentation space is apart from weight, ω (ri) indicate sub-image area riWeight, Dr
(rk, ri) it is sub-image area rkAnd riColor distance, Dr(rk,ri) calculation formula are as follows:
Wherein, f (ci,v) indicate v
A color ci,vIn i-th of sub-image area riIn all niThe probability occurred in kind color, f (ck,u) indicate u-th of color
ck,uIn k-th of sub-image area rkIn all nkThe probability occurred in kind color, D (ck,u,ci,v) it is color cK, uAnd color
ci,vColor distance in the space L*a*b is measured.
In some embodiments of the present application, as shown in figure 5, described device further include:
Characteristic modification module 340, for correcting the target board by K-means++ Clustering Model trained in advance
The described image feature of image, to retain representational feature in described image feature.
In some embodiments of the present application, as shown in figure 5, described device further include:
The module that reorders 350, described to reorder module for performing the following operations:
By with shop described in highest default first quantity of matching degree in the shop of the target board images match
The shop characteristics of image average value updates the described image feature of the target board image；
Shop described in highest default second quantity of matching degree in the determining shop with the target board images match respectively
Paving, the similar shop in the default shop；
For shop described in default second quantity, by meeting preset condition in the respective similar shop in the shop
The similar shop shop characteristics of image average value, update the shop characteristics of image in the shop；
According to the updated characteristics of image of the target board image, the updated shop image in the default shop
Feature determines the distance that reorders of the target board image Yu the default shop；
According to the distance that reorders, the shop with the target board images match is redefined.
Optionally, the update of the updated characteristics of image, the default shop according to the target board image
Shop characteristics of image afterwards determines the distance that reorders of the target board image Yu the default shop, comprising:
Pass through formula d*(p,gi)=(1- λ) dJ(p,gi)+λd(p,gi) determine the corresponding board p of the target board image
With the board g in the default shopiBetween the distance that reorders；Wherein, dJ(p,gi) indicate board p and board giBetween outstanding person
Card moral distance, calculation formula areIn formula,For board p characteristics of image and
Board gjThe neighbour of corresponding shop characteristics of image encodes distance vector；D (p, gi) indicate board p and board giBetween geneva
Distance, calculation formula are In formula, xpFor board p characteristics of image andFor board
Plaque gjCorresponding shop characteristics of image,For characteristics of image covariance matrix；λ is weight coefficient.
In some embodiments of the present application, as shown in figure 5, described device further include:
Primary dcreening operation module 360, for determining default shop according to specified geographical location information.
In some embodiments of the present application, the Fusion Features module 320 is further used for:
It is being obtained after the global image feature and subregion significant characteristics are spliced as a result, as the target
The characteristics of image of board image.
Shop identification device based on image disclosed in the embodiment of the present application for realizing the embodiment of the present application three and is implemented
Each step of shop recognition methods described in example four based on image, the specific embodiment of each module of device is referring to corresponding
Step, details are not described herein again.
Shop identification device based on image disclosed in the embodiment of the present application is schemed by obtaining the global of target board image
As feature, and obtain the subregion significant characteristics of the target board image；According to the global image feature and subregion
Domain significant characteristics determine the characteristics of image of the target board image；According to the characteristics of image of the target board image and in advance
If the determining shop with the target board images match of the shop characteristics of image in shop helps to solve to carry out shop based on image
Spread the low problem of the accuracy rate of identification.The overall situation that shop identification device as disclosed in the embodiment of the present application based on image uses
Characteristics of image reflects that the appearance difference of board and overall effect are more comprehensive, but identifies for fine granularity, and can ignore some has area
Indexing, but tiny feature, and subregion significant characteristics are then more fine granularity local features, can overcome the disadvantages that global image feature
Deficiency, therefore, after above two Fusion Features, taken into account image global characteristics and part fine granularity region so that
Characteristics of image was both comprehensive, has and has discrimination, and the accuracy rate that shop identification is carried out based on image can be improved.Meanwhile passing through K-
Means++ Clustering Model is modified characteristics of image, can fight into spy representative in characteristics of image is retained, and filters out
Redundancy or do not have representative feature, with promoted shop database storage efficiency and by reduce characteristics of image it is big
The small efficiency to promote image recognition.
Further, it only relies on characteristics of image and carries out the obtained recognition result of preliminary matches for indistinguishable sample searching
Effect is bad, for example, the result mismatched out may when target board image is the board image of " KFC " are as follows:
KFC, McDonald, Burger King ... shop recognition methods disclosed in the present application are the board of " KFC " to target board image
The characteristics of image of plaque image is updated by the shop image of KFC, McDonald, and, by " agreeing in the database of shop
The shop characteristics of image of other shops of De Ji " is updated the shop characteristics of image of " KFC ", then, then passes through update
Characteristics of image afterwards carries out match cognization, has drawn high KFC by the shop characteristics of image of other shops of " KFC " "
Sorting position can effectively promote the accuracy of images match.
Correspondingly, disclosed herein as well is a kind of electronic equipment, including memory, processor and it is stored in the memory
Computer program that is upper and can running on a processor, the processor are realized when executing the computer program as the application is real
Apply the shop recognition methods described in example one and embodiment two based on image.The electronic equipment can for PC machine, mobile terminal,
Personal digital assistant, tablet computer etc..
Disclosed herein as well is a kind of computer readable storage mediums, are stored thereon with computer program, which is located
Manage the step of realizing the shop recognition methods based on image as described in the embodiment of the present application one and embodiment two when device executes.
All the embodiments in this specification are described in a progressive manner, the highlights of each of the examples are with
The difference of other embodiments, the same or similar parts between the embodiments can be referred to each other.For Installation practice
For, since it is basically similar to the method embodiment, so being described relatively simple, referring to the portion of embodiment of the method in place of correlation
It defends oneself bright.
It above to a kind of shop recognition methods based on image provided by the present application, device, is described in detail, herein
In apply specific case the principle and implementation of this application are described, the explanation of above example is only intended to sides
Assistant solves the present processes and its core concept；At the same time, for those skilled in the art, the think of according to the application
Think, there will be changes in the specific implementation manner and application range, in conclusion the content of the present specification should not be construed as pair
The limitation of the application.
Through the above description of the embodiments, those skilled in the art can be understood that each embodiment can
It realizes by means of software and necessary general hardware platform, naturally it is also possible to pass through hardware realization.Based on such reason
Solution, substantially the part that contributes to existing technology can embody above-mentioned technical proposal in the form of software products in other words
Come, which may be stored in a computer readable storage medium, such as ROM/RAM, magnetic disk, CD, including
Some instructions are used so that a computer installation (can be personal computer, server or network equipment etc.) executes respectively
Method described in certain parts of a embodiment or embodiment.
Claims (22)
1. a kind of shop recognition methods based on image characterized by comprising
The global image feature of target board image is obtained, and obtains the subregion conspicuousness spy of the target board image
Sign；
The characteristics of image of the target board image is determined according to the global image feature and subregion significant characteristics；
According to the characteristics of image of the target board image and the determination of the shop characteristics of image in default shop and the target board
The shop of images match.
2. the method according to claim 1, wherein the global image feature for obtaining target board image
Step, comprising:
By neural network model trained in advance, the global image feature of the target board image is obtained.
3. the method according to claim 1, wherein the subregion for obtaining the target board image is significant
The step of property feature, comprising:
The target board image is divided according at least one dividing method, determines several image-regions；
For each described image region, described image region is split according to distribution of color, determines described image region
At least one sub-image area for including；
For each described image region, the space length and color between sub-image area that include by described image region
Distance determines the subregion significant characteristics in described image region.
4. according to the method described in claim 3, it is characterized in that, the sub-image area for including by described image region
Between space length and color distance, the step of determining the subregion significant characteristics in described image region, comprising:
The space length and color distance between sub-image area for including by described image region, determine each subgraph
The conspicuousness in region；
According to the conspicuousness of the sub-image area of the highest preset quantity of the conspicuousness, each sub-image area is determined
The subregion significant characteristics of affiliated image-region.
5. according to the method described in claim 4, it is characterized in that, the sub-image area for including by described image region
Between space length and color distance, the step of determining the conspicuousness of each sub-image area, comprising:
Pass through formulaDetermine that described image region includes respectively
Each sub-image area rkConspicuousness, wherein riIt indicates to be different from rkSub-image area, Ds(rk,ri) indicate sub-image regions
Domain rkAnd riSpace length, σsRepresentation space is apart from weight, ω (ri) indicate sub-image area riWeight, Dr(rk, ri) be
Sub-image area rkAnd riColor distance, Dr(rk,ri) calculation formula are as follows:
Wherein, f (ci,v) indicate v-th of color
ci,vIn i-th of sub-image area riIn all niThe probability occurred in kind color, f (ck,u) indicate u-th of color ck,uIn kth
A sub-image area rkIn all nkThe probability occurred in kind color, D (ck,u,ci,v) it is color ck,uWith color ci,vIn L*a*b
The color distance in space is measured.
6. the method according to claim 1, wherein the characteristics of image according to the target board image and
Before the step of shop characteristics of image determination in default shop and the shop of the target board images match, further includes:
By K-means++ Clustering Model trained in advance, the described image feature of the target board image is corrected, to retain
Representational feature in described image feature.
7. method according to any one of claims 1 to 6, which is characterized in that described according to the target board image
After the step of characteristics of image and determination of the shop characteristics of image in default shop and the shop of the target board images match, also
Include:
By with described in shop described in highest default first quantity of matching degree in the shop of the target board images match
Shop characteristics of image average value updates the described image feature of the target board image；
Shop described in highest default second quantity of matching degree in the determining shop with the target board images match respectively,
Similar shop in the default shop；
For shop described in default second quantity, by the institute for meeting preset condition in the respective similar shop in the shop
The shop characteristics of image average value for stating similar shop, updates the shop characteristics of image in the shop；
According to the updated characteristics of image of the target board image, the updated shop characteristics of image in the default shop
Determine the distance that reorders of the target board image Yu the default shop；
According to the distance that reorders, the shop with the target board images match is redefined.
8. the method according to the description of claim 7 is characterized in that the updated figure according to the target board image
As the updated shop characteristics of image of feature, the default shop determines the target board image and the default shop
Reorder apart from the step of, comprising:
Pass through formula d*(p,gi)=(1- λ) dJ(p,gi)+λ d (p, gi) determine the corresponding board p of the target board image and institute
State the board g in default shopiBetween the distance that reorders；Wherein, dJ(p,gi) indicate board p and board giBetween Jie Kade
Distance, calculation formula areIn formula,For the characteristics of image and board of board p
gjThe neighbour of corresponding shop characteristics of image encodes distance vector；D (p, gi) indicate board p and board giBetween mahalanobis distance,
Calculation formula is In formula, xpFor board p characteristics of image andFor board gj
Corresponding shop characteristics of image,For characteristics of image covariance matrix；λ is weight coefficient.
9. method according to any one of claims 1 to 6, which is characterized in that described according to the target board image
Before the step of characteristics of image and determination of the shop characteristics of image in default shop and the shop of the target board images match, also
Include:
Default shop is determined according to specified geographical location information.
10. method according to any one of claims 1 to 6, which is characterized in that it is described according to the global image feature and
Subregion significant characteristics determine the step of characteristics of image of the target board image, comprising:
It is being obtained after the global image feature and subregion significant characteristics are spliced as a result, as the target board
The characteristics of image of image.
11. a kind of shop identification device based on image characterized by comprising
Feature obtains module, for obtaining the global image feature of target board image, and the acquisition target board image
Subregion significant characteristics；
Fusion Features module, for determining the target board figure according to the global image feature and subregion significant characteristics
The characteristics of image of picture；
Match cognization module, for true according to the characteristics of image of the target board image and the shop characteristics of image in default shop
The fixed shop with the target board images match.
12. device according to claim 11, which is characterized in that the feature obtains module and further comprises:
Fisrt feature acquisition submodule, for obtaining the target board image by neural network model trained in advance
Global image feature.
13. device according to claim 11, which is characterized in that the feature obtains module and further comprises:
Second feature acquisition submodule, for dividing the target board image according at least one dividing method, really
Fixed several image-regions；
The second feature acquisition submodule is also used to for each described image region, according to distribution of color to described image
Region is split, and determines at least one sub-image area that described image region includes；And
For each described image region, the space length and color between sub-image area that include by described image region
Distance determines the subregion significant characteristics in described image region.
14. device according to claim 13, which is characterized in that the sub-image regions for including by described image region
Space length and color distance between domain determine the subregion significant characteristics in described image region, comprising:
The space length and color distance between sub-image area for including by described image region, determine each subgraph
The conspicuousness in region；
According to the conspicuousness of the sub-image area of the highest preset quantity of the conspicuousness, each sub-image area is determined
The subregion significant characteristics of affiliated image-region.
15. device according to claim 14, which is characterized in that the sub-image regions for including by described image region
Space length and color distance between domain determine the conspicuousness of each sub-image area, comprising:
Pass through formulaDetermine that described image region includes respectively
Each sub-image area rkConspicuousness, wherein riIt indicates to be different from rkSub-image area, Ds(rk,ri) indicate sub-image regions
Domain rkAnd riSpace length, σsRepresentation space is apart from weight, ω (ri) indicate sub-image area riWeight, Dr(rk,ri) be
Sub-image area rkAnd riColor distance, Dr(rk,ri) calculation formula are as follows:
Wherein, f (ci,v) indicate v-th of color
cI, vIn i-th of sub-image area riIn all niThe probability occurred in kind color, f (ck,u) indicate u-th of color ck,uIn kth
A sub-image area rkIn all nkThe probability occurred in kind color, D (ck,u,ci,v) it is color ck,uWith color ci,vIn L*a*b
The color distance in space is measured.
16. device according to claim 11, which is characterized in that described device further include:
Characteristic modification module, for correcting the institute of the target board image by K-means++ Clustering Model trained in advance
Characteristics of image is stated, to retain representational feature in described image feature.
17. 1 to 15 described in any item devices according to claim 1, which is characterized in that described device further include: reorder mould
Block, described to reorder module for performing the following operations:
By with described in shop described in highest default first quantity of matching degree in the shop of the target board images match
Shop characteristics of image average value updates the described image feature of the target board image；
Shop described in highest default second quantity of matching degree in the determining shop with the target board images match respectively,
Similar shop in the default shop；
For shop described in default second quantity, by the institute for meeting preset condition in the respective similar shop in the shop
The shop characteristics of image average value for stating similar shop, updates the shop characteristics of image in the shop；
According to the updated characteristics of image of the target board image, the updated shop characteristics of image in the default shop
Determine the distance that reorders of the target board image Yu the default shop；
According to the distance that reorders, the shop with the target board images match is redefined.
18. device according to claim 17, which is characterized in that described according to the updated of the target board image
Characteristics of image, the default shop updated shop characteristics of image determine the target board image and the default shop
The distance that reorders, comprising:
Pass through formula d*(p,gi)=(1- λ) dJ(p,gi)+λd(p,gi) determine the corresponding board p of the target board image and institute
State the board g in default shopiBetween the distance that reorders；Wherein, dJ(p,gi) indicate board p and board giBetween Jie Kade
Distance, calculation formula areIn formula,For the characteristics of image and board of board p
Plaque gjThe neighbour of corresponding shop characteristics of image encodes distance vector；d(p,gi) indicate board p and board giBetween geneva away from
From calculation formula is In formula, xpFor board p characteristics of image andFor board
gjCorresponding shop characteristics of image,For characteristics of image covariance matrix；λ is weight coefficient.
19. 1 to 15 described in any item devices according to claim 1, which is characterized in that described device further include:
Primary dcreening operation module, for determining default shop according to specified geographical location information.
20. 1 to 15 described in any item devices according to claim 1, which is characterized in that the Fusion Features module is further used
In:
It is being obtained after the global image feature and subregion significant characteristics are spliced as a result, as the target board
The characteristics of image of image.
21. a kind of electronic equipment, including memory, processor and it is stored on the memory and can runs on a processor
Computer program, which is characterized in that the processor realizes claims 1 to 10 any one when executing the computer program
The shop recognition methods based on image.
22. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that the program is by processor
The step of shop recognition methods described in claims 1 to 10 any one based on image is realized when execution.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811309281.3A CN109711399B (en) | 2018-11-05 | 2018-11-05 | Shop identification method and device based on image and electronic equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811309281.3A CN109711399B (en) | 2018-11-05 | 2018-11-05 | Shop identification method and device based on image and electronic equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109711399A true CN109711399A (en) | 2019-05-03 |
| CN109711399B CN109711399B (en) | 2021-04-27 |
Family
ID=66254284
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811309281.3A Active CN109711399B (en) | 2018-11-05 | 2018-11-05 | Shop identification method and device based on image and electronic equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109711399B (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110175655A (en) * | 2019-06-03 | 2019-08-27 | 中国科学技术大学 | Data identification method and device, storage medium and electronic equipment |
| CN110223050A (en) * | 2019-06-24 | 2019-09-10 | 广东工业大学 | A kind of verification method and relevant apparatus of merchant store fronts title |
| CN110335270A (en) * | 2019-07-09 | 2019-10-15 | 华北电力大学（保定） | Transmission line of electricity defect inspection method based on the study of hierarchical regions Fusion Features |
| CN110796664A (en) * | 2019-10-14 | 2020-02-14 | 北京字节跳动网络技术有限公司 | Image processing method, image processing device, electronic equipment and computer readable storage medium |
| CN112784086A (en) * | 2021-01-28 | 2021-05-11 | 北京有竹居网络技术有限公司 | Picture screening method and device, storage medium and electronic equipment |
| CN113065559A (en) * | 2021-06-03 | 2021-07-02 | 城云科技（中国）有限公司 | Image comparison method and device, electronic equipment and storage medium |
| CN113450174A (en) * | 2020-03-27 | 2021-09-28 | 北京沃东天骏信息技术有限公司 | Method and device for determining similar stores |
| CN114169930A (en) * | 2021-12-07 | 2022-03-11 | 钻技(上海)信息科技有限公司 | Online and offline cooperative store accurate marketing method and system |
| CN114693957A (en) * | 2020-12-30 | 2022-07-01 | 顺丰科技有限公司 | Package classification method and device, electronic equipment and storage medium |
| CN116486145A (en) * | 2023-03-30 | 2023-07-25 | 北京百炼智能科技有限公司 | Brand image store discriminating method and device based on target detection |
| TWI832642B (en) * | 2022-12-28 | 2024-02-11 | 國立中央大學 | Image processing method for robust signboard detection and recognition |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101369316A (en) * | 2008-07-09 | 2009-02-18 | 东华大学 | Image Feature Extraction Method Based on Global and Local Structure Fusion |
| CN102129693A (en) * | 2011-03-15 | 2011-07-20 | 清华大学 | Image vision significance calculation method based on color histogram and global contrast |
| CN106203746A (en) * | 2016-09-30 | 2016-12-07 | 携程计算机技术（上海）有限公司 | Hotel group divides and the method for requirement forecasting |
| CN107122701A (en) * | 2017-03-03 | 2017-09-01 | 华南理工大学 | A kind of traffic route sign based on saliency and deep learning |
| CN107451156A (en) * | 2016-05-31 | 2017-12-08 | 杭州华为企业通信技术有限公司 | A kind of image recognition methods and identification device again |
| US20180068198A1 (en) * | 2016-09-06 | 2018-03-08 | Carnegie Mellon University | Methods and Software for Detecting Objects in an Image Using Contextual Multiscale Fast Region-Based Convolutional Neural Network |
| US20180101964A1 (en) * | 2016-10-11 | 2018-04-12 | Electronics And Telecommunications Research Institute | Method and system for generating integral image marker |
| CN108280469A (en) * | 2018-01-16 | 2018-07-13 | 佛山市顺德区中山大学研究院 | A kind of supermarket's commodity image recognition methods based on rarefaction representation |
-
2018
- 2018-11-05 CN CN201811309281.3A patent/CN109711399B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101369316A (en) * | 2008-07-09 | 2009-02-18 | 东华大学 | Image Feature Extraction Method Based on Global and Local Structure Fusion |
| CN102129693A (en) * | 2011-03-15 | 2011-07-20 | 清华大学 | Image vision significance calculation method based on color histogram and global contrast |
| CN107451156A (en) * | 2016-05-31 | 2017-12-08 | 杭州华为企业通信技术有限公司 | A kind of image recognition methods and identification device again |
| US20180068198A1 (en) * | 2016-09-06 | 2018-03-08 | Carnegie Mellon University | Methods and Software for Detecting Objects in an Image Using Contextual Multiscale Fast Region-Based Convolutional Neural Network |
| CN106203746A (en) * | 2016-09-30 | 2016-12-07 | 携程计算机技术（上海）有限公司 | Hotel group divides and the method for requirement forecasting |
| US20180101964A1 (en) * | 2016-10-11 | 2018-04-12 | Electronics And Telecommunications Research Institute | Method and system for generating integral image marker |
| CN107122701A (en) * | 2017-03-03 | 2017-09-01 | 华南理工大学 | A kind of traffic route sign based on saliency and deep learning |
| CN108280469A (en) * | 2018-01-16 | 2018-07-13 | 佛山市顺德区中山大学研究院 | A kind of supermarket's commodity image recognition methods based on rarefaction representation |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110175655A (en) * | 2019-06-03 | 2019-08-27 | 中国科学技术大学 | Data identification method and device, storage medium and electronic equipment |
| CN110223050A (en) * | 2019-06-24 | 2019-09-10 | 广东工业大学 | A kind of verification method and relevant apparatus of merchant store fronts title |
| CN110335270B (en) * | 2019-07-09 | 2022-09-13 | 华北电力大学（保定） | Power transmission line defect detection method based on hierarchical regional feature fusion learning |
| CN110335270A (en) * | 2019-07-09 | 2019-10-15 | 华北电力大学（保定） | Transmission line of electricity defect inspection method based on the study of hierarchical regions Fusion Features |
| CN110796664A (en) * | 2019-10-14 | 2020-02-14 | 北京字节跳动网络技术有限公司 | Image processing method, image processing device, electronic equipment and computer readable storage medium |
| CN110796664B (en) * | 2019-10-14 | 2023-05-23 | 北京字节跳动网络技术有限公司 | Image processing method, device, electronic equipment and computer readable storage medium |
| CN113450174A (en) * | 2020-03-27 | 2021-09-28 | 北京沃东天骏信息技术有限公司 | Method and device for determining similar stores |
| CN114693957A (en) * | 2020-12-30 | 2022-07-01 | 顺丰科技有限公司 | Package classification method and device, electronic equipment and storage medium |
| CN112784086A (en) * | 2021-01-28 | 2021-05-11 | 北京有竹居网络技术有限公司 | Picture screening method and device, storage medium and electronic equipment |
| CN113065559A (en) * | 2021-06-03 | 2021-07-02 | 城云科技（中国）有限公司 | Image comparison method and device, electronic equipment and storage medium |
| CN114169930A (en) * | 2021-12-07 | 2022-03-11 | 钻技(上海)信息科技有限公司 | Online and offline cooperative store accurate marketing method and system |
| TWI832642B (en) * | 2022-12-28 | 2024-02-11 | 國立中央大學 | Image processing method for robust signboard detection and recognition |
| CN116486145A (en) * | 2023-03-30 | 2023-07-25 | 北京百炼智能科技有限公司 | Brand image store discriminating method and device based on target detection |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109711399B (en) | 2021-04-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109711399A (en) | Shop recognition methods based on image, device, electronic equipment | |
| CN110276366B (en) | Detecting objects using a weakly supervised model | |
| CN103838810B (en) | The method and apparatus for carrying out picture search using label information | |
| CN109255352A (en) | Object detection method, apparatus and system | |
| Chen et al. | Outdoor shadow estimating using multiclass geometric decomposition based on BLS | |
| CN109271888A (en) | Personal identification method, device, electronic equipment based on gait | |
| CN110163076A (en) | A kind of image processing method and relevant apparatus | |
| CN108875522A (en) | Face cluster methods, devices and systems and storage medium | |
| CN105068918B (en) | A kind of page method of testing and device | |
| CN106204118A (en) | A kind of method and apparatus found for commercial circle | |
| CN114758362A (en) | Clothing changing pedestrian re-identification method based on semantic perception attention and visual masking | |
| CN109241808A (en) | Two dimensional code auto-screening method in a kind of fashion images | |
| CN105303163B (en) | A kind of method and detection device of target detection | |
| CN110443179B (en) | Off-post detection method and device and storage medium | |
| CN109345553A (en) | A kind of palm and its critical point detection method, apparatus and terminal device | |
| CN111414803A (en) | Face recognition method and device and electronic equipment | |
| Zhou et al. | Deep learning-based instance segmentation for indoor fire load recognition | |
| CN109766822A (en) | Gesture recognition method and system based on neural network | |
| CN113628245A (en) | Multi-target tracking method, device, electronic equipment and storage medium | |
| CN113705461A (en) | Face definition detection method, device, equipment and storage medium | |
| CN112861678A (en) | Image identification method and device | |
| CN113643420A (en) | Three-dimensional reconstruction method and device | |
| CN106709489A (en) | Processing method and device of character identification | |
| CN110399812A (en) | Face characteristic intelligent extract method, device and computer readable storage medium | |
| CN112132130A (en) | A real-time license plate detection method and system for all scenarios |
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 |