CN109815873A - Merchandise display method, apparatus, equipment and medium based on image recognition - Google Patents
Merchandise display method, apparatus, equipment and medium based on image recognition Download PDFInfo
- Publication number
- CN109815873A CN109815873A CN201910042541.3A CN201910042541A CN109815873A CN 109815873 A CN109815873 A CN 109815873A CN 201910042541 A CN201910042541 A CN 201910042541A CN 109815873 A CN109815873 A CN 109815873A
- Authority
- CN
- China
- Prior art keywords
- image
- recognized
- images
- commodity
- target
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 50
- 230000036651 mood Effects 0.000 claims abstract description 150
- 230000008451 emotion Effects 0.000 claims abstract description 49
- 230000001815 facial effect Effects 0.000 claims description 65
- 238000004422 calculation algorithm Methods 0.000 claims description 46
- 238000004590 computer program Methods 0.000 claims description 18
- 238000000605 extraction Methods 0.000 claims description 15
- 238000012549 training Methods 0.000 claims description 13
- 230000006870 function Effects 0.000 claims description 9
- 230000001747 exhibiting effect Effects 0.000 claims description 3
- 238000013527 convolutional neural network Methods 0.000 description 10
- 210000004709 eyebrow Anatomy 0.000 description 10
- 238000013528 artificial neural network Methods 0.000 description 7
- 239000000284 extract Substances 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 238000012163 sequencing technique Methods 0.000 description 4
- 238000001514 detection method Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 239000000203 mixture Substances 0.000 description 3
- 230000001537 neural effect Effects 0.000 description 3
- 238000012216 screening Methods 0.000 description 3
- 206010021703 Indifference Diseases 0.000 description 2
- 230000009471 action Effects 0.000 description 2
- 239000002537 cosmetic Substances 0.000 description 2
- 230000008878 coupling Effects 0.000 description 2
- 238000010168 coupling process Methods 0.000 description 2
- 238000005859 coupling reaction Methods 0.000 description 2
- 210000000744 eyelid Anatomy 0.000 description 2
- 230000033001 locomotion Effects 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 230000004913 activation Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 210000004027 cell Anatomy 0.000 description 1
- 230000010485 coping Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 210000004209 hair Anatomy 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000004445 quantitative analysis Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 230000001360 synchronised effect Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/06—Buying, selling or leasing transactions
Landscapes
- Business, Economics & Management (AREA)
- Accounting & Taxation (AREA)
- Finance (AREA)
- Development Economics (AREA)
- Economics (AREA)
- Marketing (AREA)
- Strategic Management (AREA)
- Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The present invention discloses a kind of merchandise display method, apparatus, equipment and medium based on image recognition, this method includes carrying out recognition of face and cluster at least two frame images to be recognized in the target video data of commodity to be presented using Face datection model, obtains the corresponding customer quantity of target video data and the corresponding image clustering set of each client；If customer quantity is greater than preset quantity, the images to be recognized in each image clustering set is identified using preparatory trained micro- Expression Recognition model, obtains the single frames mood of each images to be recognized；Single frames mood based on images to be recognized obtains the target emotion of client corresponding with image clustering set；According to the target emotion of customer quantity and the corresponding client of each image clustering set, final mood is obtained；According to the corresponding final mood of commodity to be presented, intended display commodity are obtained, to solve the problems, such as to show that commodity attraction is inadequate.
Description
Technical field
The present invention relates to intelligent decision field more particularly to a kind of merchandise display method, apparatus based on image recognition, set
Standby and medium.
Background technique
Currently, numerous businessmans can release new commodity during new product release or season as the time alternates, need
The commodity of release are promoted, be bought with attracting clients.Usual businessman is that the commodity that will need to release pick out one
A little commodity are shown, so that client's browsing is checked, but the hobby due to not knowing client, and chosen from commodity with personal eye
It selects the commodity that the outstanding commodity of comparison are shown or intra-company personnel show needs to determine, to make
The commodity that must be shown do not meet client's eye or demand, reduce the attraction for showing commodity, can not attract clients and buy into shop.
Summary of the invention
The embodiment of the present invention provides a kind of merchandise display method, apparatus, equipment and medium based on image recognition, to solve
Show the inadequate problem of commodity attraction.
A kind of merchandise display method based on image recognition, comprising:
The target video data of commodity to be presented is obtained, the target video data includes at least two frame images to be recognized；
Recognition of face and cluster are carried out using Face datection model images to be recognized described at least two frames, obtain the mesh
The corresponding customer quantity of video data and the corresponding image clustering set of each client are marked, each described image cluster set includes
An at least frame images to be recognized；
If the customer quantity is greater than preset quantity, using preparatory trained micro- Expression Recognition model to each described
Images to be recognized in image clustering set is identified, the single frames mood of each images to be recognized is obtained；
Based on the single frames mood of images to be recognized described in an at least frame, client corresponding with described image cluster set is obtained
Target emotion；
The target emotion for gathering corresponding client according to the customer quantity and each described image cluster, obtains final feelings
Thread；
According to the corresponding final mood of the commodity to be presented, intended display commodity are obtained.
A kind of commodity-exhibiting device based on image recognition, comprising:
Data acquisition module, for obtaining the target video data of commodity to be presented, the target video data includes extremely
Few two frame images to be recognized；
Image clustering set obtains module, for being carried out using Face datection model images to be recognized described at least two frames
Recognition of face and cluster obtain the corresponding customer quantity of the target video data and the corresponding image clustering collection of each client
It closes, each described image cluster set includes an at least frame images to be recognized；
Single frames mood determining module, if being greater than preset quantity for the customer quantity, using trained micro- in advance
Expression Recognition model identifies the images to be recognized in each described image cluster set, obtains each figure to be identified
The single frames mood of picture；
Target emotion obtains module, for the single frames mood based on images to be recognized described in an at least frame, obtain with it is described
The target emotion of the corresponding client of image clustering set；
Final mood obtains module, for gathering corresponding client according to the customer quantity and each described image cluster
Target emotion, obtain final mood；
Intended display commodity obtain module, for obtaining target exhibition according to the corresponding final mood of the commodity to be presented
Show commodity.
A kind of computer equipment, including memory, processor and storage are in the memory and can be in the processing
The computer program run on device, the processor realize the above-mentioned commodity based on image recognition when executing the computer program
The step of methods of exhibiting.
A kind of computer readable storage medium, the computer-readable recording medium storage have computer program, the meter
Calculation machine program realizes the step of above-mentioned merchandise display method based on image recognition when being executed by processor.
Above-mentioned merchandise display method, apparatus, equipment and medium based on image recognition, by the mesh for obtaining commodity to be presented
Video data is marked, recognition of face and cluster are carried out to images to be recognized in target video data using Face datection model, obtained
The corresponding customer quantity of target video data and the corresponding image clustering set of each client, so as to subsequent according to a certain number of
The target emotion of client determines intended display commodity, improves the accuracy rate of intended display commodity.If customer quantity is greater than present count
Amount then uses preparatory trained micro- Expression Recognition model to identify the images to be recognized in each image clustering set,
The single frames mood of each images to be recognized is obtained, to realize the mood of identification client.List based on an at least frame images to be recognized
Frame mood obtains the target emotion of client corresponding with image clustering set, to determine whether the commodity to be presented are client institute
Interested commodity.It is certain to get according to the target emotion of customer quantity and the corresponding client of each image clustering set
The client of quantity obtains intended display according to the corresponding final mood of commodity to be presented to the final mood of the commodity to be presented
Commodity improve the accuracy rate of intended display commodity, so that end article is the more interested commodity of most of clients.
Detailed description of the invention
In order to illustrate the technical solution of the embodiments of the present invention more clearly, below by institute in the description to the embodiment of the present invention
Attached drawing to be used is needed to be briefly described, it should be apparent that, the accompanying drawings in the following description is only some implementations of the invention
Example, for those of ordinary skill in the art, without any creative labor, can also be according to these attached drawings
Obtain other attached drawings.
Fig. 1 is the application environment schematic diagram of the merchandise display method in one embodiment of the invention based on image recognition；
Fig. 2 is the flow chart of the merchandise display method in one embodiment of the invention based on image recognition；
Fig. 3 is the flow chart of the merchandise display method in one embodiment of the invention based on image recognition；
Fig. 4 is the flow chart of the merchandise display method in one embodiment of the invention based on image recognition；
Fig. 5 is the flow chart of the merchandise display method in one embodiment of the invention based on image recognition；
Fig. 6 is the flow chart of the merchandise display method in one embodiment of the invention based on image recognition；
Fig. 7 is the flow chart of the merchandise display method in one embodiment of the invention based on image recognition；
Fig. 8 is the functional block diagram of the commodity-exhibiting device in one embodiment of the invention based on image recognition；
Fig. 9 is a schematic diagram of computer equipment in one embodiment of the invention.
Specific embodiment
Below by the attached drawing in knot and the embodiment of the present invention, technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are some of the embodiments of the present invention, instead of all the embodiments.Based on this hair
Embodiment in bright, every other implementation obtained by those of ordinary skill in the art without making creative efforts
Example, shall fall within the protection scope of the present invention.
Merchandise display method provided in an embodiment of the present invention based on image recognition, can be applicable to the application environment such as Fig. 1
In, wherein client is communicated by network with server-side.The merchandise display method based on image recognition, which is applied, to be serviced
On end, by carrying out analysis identification to the corresponding target video data of commodity to be presented, each visitor in target video data is obtained
The corresponding mood in family, and client is determined to the final mood of the commodity to be presented, according to every according to the corresponding mood of each client
The corresponding final mood of one commodity to be presented determines intended display commodity, so that intended display commodity are the quotient that client more pays close attention to
Product improve the attraction of intended display commodity, are bought with attracting clients to displaying commodity.Wherein, client can with but not
It is limited to various personal computers, laptop, smart phone, tablet computer and portable wearable device.Server-side can be with
It is realized with the server-side cluster of the either multiple server-side compositions of independent server-side.
In one embodiment, it as shown in Fig. 2, providing a kind of merchandise display method based on image recognition, answers in this way
It is illustrated, specifically comprises the following steps: for the server-side in Fig. 1
S10: obtaining the target video data of commodity to be presented, and target video data includes at least two frame images to be recognized.
Wherein, target video data, which refers to, screens to get initial video data corresponding with commodity to be presented
Video data, specifically can be the video data for meeting some conditions.For example, meet the attribute information of commodity to be presented
Video data.Attribute information may include Suitable Age and suitable gender.It is to be appreciated that passing through Suitable Age and suitable gender pair
Collected initial video data corresponding with commodity to be presented is screened, the target video data got.Figure to be identified
It seem the image referred to after being screened according to Suitable Age and suitable gender.Wherein, initial video data is video acquisition tool
Collected video data corresponding with each commodity to be presented.
Specifically, a video acquisition tool, the video acquisition tool are configured for each commodity region to be presented in advance
For acquiring image or video data, when video acquisition tool detect client appear in sampling instrument acquisition range it
When interior, video acquisition tool automatic trigger, and image or video data acquiring are carried out to client.The video acquisition tool is specific
For camera, the initial video data in the corresponding acquisition range of each commodity to be presented can be acquired by camera in real time.By
Correspond to a commodity to be presented in each sampling instrument, each commodity region to be presented is acquired by each camera
Initial video data, screened by the initial video data to commodity region to be presented, with get it is each with
The corresponding target video data of commodity to be presented.Wherein, it is corresponding that commodity to be presented are carried in collected initial video data
Commodity sign, to determine corresponding target video data subsequently through commodity sign.For example, video acquisition tool collects
Initial video data include commodity sign A, then the initial video data be the corresponding initial video data of commodity sign A,
Initial video data is screened, to get the corresponding target video data of commodity sign A.Wherein, commodity sign refers to
For distinguishing the unique identification of different commodity to be presented.Optionally, commodity sign can be by number, letter, text or symbol
Number at least one of composition.For example, commodity sign can be number either sequence number of commodity to be presented etc..
S20: recognition of face and cluster are carried out at least two frame images to be recognized using Face datection model, obtain target view
For frequency according to corresponding customer quantity and the corresponding image clustering set of each client, each image clustering set includes an at least frame
Images to be recognized.
Wherein, Face datection model refers to trained in advance for detecting whether each frame images to be recognized includes people's
The model of facial area.Customer quantity refers in target video data, according to quantity determined by client's difference.Image clustering
Set, which refers to, clusters the corresponding images to be recognized of same client, with the set of formation.
Specifically, server-side is connect with data bank network, and Face datection model, target video data are stored in database
In include the corresponding images to be recognized of different clients, images to be recognized is identified using Face datection model, to get
The facial image for including in images to be recognized, wherein facial image refers to the image of the human face region of client.Server-side will acquire
To target video data in each images to be recognized be input to Face datection model, by Face datection model to target regard
Frequency each images to be recognized in carries out recognition of face, determines whether each images to be recognized is facial image, if to be identified
Image is facial image, then clusters identical facial image, i.e., to the corresponding images to be recognized of identical facial image
It is clustered, to get the corresponding image clustering set of each client, according to image clustering set in target video data
Quantity determines customer quantity in target video data.
More specifically, the face characteristic of the corresponding facial image of each images to be recognized is extracted using feature extraction algorithm,
The corresponding face characteristic of each images to be recognized is subjected to characteristic similarity calculating, if characteristic similarity is greater than preset threshold,
Illustrating that characteristic similarity is greater than the corresponding facial image of preset threshold is the corresponding facial image of same client, by same client couple
The corresponding images to be recognized of the facial image answered is clustered, to obtain the corresponding image clustering set of each client, and according to
The quantity of image clustering set determines the customer quantity in target video data.Wherein, preset when preset threshold to be used for
Whether assessment similarity reaches the value for being determined as the face of same client.Feature extraction algorithm includes but is not limited to CNN
(Convolutional Neural Network, convolutional Neural net) algorithm, for example, it is to be identified to extract people by CNN algorithm
The face characteristic of the corresponding facial image of image.
S30: if customer quantity is greater than preset quantity, using preparatory trained micro- Expression Recognition model to each image
Images to be recognized in cluster set is identified, the single frames mood of each images to be recognized is obtained.
Wherein, micro- Expression Recognition model refers to the local feature by capturing the client face in images to be recognized, and root
According to each target face motor unit of face in local feature recognition images to be recognized, further according to the target face identified
Motor unit determines the model of its mood.Single frames mood, which refers to, identifies images to be recognized by micro- Expression Recognition model,
The mood determined according to the target face motor unit identified.
Wherein, micro- Expression Recognition model can be the neural network recognization model based on deep learning, can be and is based on dividing
The local identification model of class can also be the local mood based on local binary patterns (Local Binary Pattern, LBP)
Identification model.In the present embodiment, micro- Expression Recognition model is the local identification model based on classification, and micro- Expression Recognition model is preparatory
It include each Facial action unit in training image data by collecting a large amount of training image data in advance when being trained
Positive sample and Facial action unit negative sample, training image data are trained by sorting algorithm, obtain micro- expression
Identification model.In the present embodiment, it can be and a large amount of training image data are trained by svm classifier algorithm, to obtain
To SVM classifier corresponding with N number of Facial action unit.For example, it may be 39 Facial action units are 39 SVM points corresponding
Class device is also possible to corresponding 54 SVM classifiers of 54 Facial action units, includes in the training image data being trained
Different Facial action units positive sample and negative sample it is more, then the SVM classifier quantity got is more.It is understood that
Ground forms micro- Expression Recognition model by N number of SVM classifier, and the SVM classifier got is more, then the micro- table formed
The mood that feelings identification model is identified is more accurate.
Wherein, preset quantity is preset value, and the corresponding preset quantity of each commodity to be presented is identical, and server-side is sentenced
Disconnected customer quantity out is greater than preset quantity, then using preparatory trained micro- Expression Recognition model in each image clustering set
Images to be recognized identified, obtain the corresponding single frames mood of each images to be recognized.Using preparatory trained micro- expression
Identification model carries out identification to the images to be recognized in each image clustering set and specifically comprises the following steps that server-side is first treated
Identify that image carries out face critical point detection and feature extraction, to obtain corresponding local feature from images to be recognized, then will
The corresponding local feature of images to be recognized is input in preparatory trained micro- Expression Recognition model.It is wrapped in micro- Expression Recognition model
N number of SVM classifier is included, each SVM classifier identifies the corresponding local feature of an images to be recognized, passes through N number of SVM
All local features of the input of classifier pair identify, obtain N number of SVM classifier output with the Facial action unit pair
The Facial action unit is then determined as and the images to be recognized by the probability value answered when probability value is greater than predetermined probabilities threshold value
Corresponding target face motor unit.Wherein, preset value when predetermined probabilities threshold value.Wherein, target face motor unit
Refer to and images to be recognized is identified according to micro- Expression Recognition model, the Facial action unit that gets (Action Unit,
AU)。
Include 54 SVM classifiers in micro- Expression Recognition model in the present embodiment, establishes Facial action unit number mapping
Table, each Facial action unit are indicated with a prespecified number.For example, AU1 is that interior eyebrow raises up, AU2 is that outer eyebrow raises up,
AU5 be upper eyelid raise up with AU26 be lower jaw open etc..Each Facial action unit trains corresponding SVM classifier.Example
Such as, the probability value that interior eyebrow raises up is belonged to by the interior eyebrow local feature that corresponding SVM classifier may recognize that interior eyebrow raises up that raises up,
Belong to the probability value etc. that outer eyebrow raises up by the outer eyebrow local feature that corresponding SVM classifier may recognize that outer eyebrow raises up that raises up.
Probability value specifically can be the value between 0-1, if the probability value of output is 0.6, predetermined probabilities threshold value is 0.5, then probability value
0.6 is greater than predetermined probabilities threshold value 0.5, then by 0.6 corresponding Facial action unit, the target face as images to be recognized is acted
Unit.
Specifically, server-side can determine the figure to be identified according to the corresponding target face motor unit of each images to be recognized
The single frames mood of picture, i.e., according to the corresponding target face motor unit query assessment table of each images to be recognized, obtain with should be to
Identify the corresponding single frames mood of image.In embodiment as above, local feature is identified by 54 SVM classifiers, really
Fixed all target face motor units corresponding with images to be recognized, and according to all target face motor units, search assessment
Table determines single frames mood corresponding with images to be recognized, to improve the accuracy for obtaining client's single frames mood.Wherein, table is assessed
It is preconfigured table.One or more Facial action units form different moods, and it is corresponding to get each mood in advance
The combination of each Facial action unit and corresponding mood are associated storage, to form assessment table by Facial action unit combination.
S40: the single frames mood based on an at least frame images to be recognized obtains the mesh of client corresponding with image clustering set
Mark mood.
Wherein, target emotion refers to the feelings determined according to the corresponding single frames mood of each images to be recognized of image clustering set
Thread.It is to be appreciated that being the corresponding images to be recognized of same client in image clustering set, pass through each to be identified of the client
The corresponding single frames mood of image, determines the corresponding target emotion of the client.
Specifically, server-side obtains the corresponding single frames mood of each images to be recognized of image clustering set, to image clustering
Gather the corresponding single frames mood of each images to be recognized to be analyzed, to obtain the target feelings of the corresponding client of image clustering set
Thread.It is to be appreciated that judging whether the corresponding single frames mood of each images to be recognized of image clustering set is identical, if image clustering
All single frames moods are identical in set, then using the single frames mood as target emotion.If at least two single frames moods are not identical,
Then determine which kind of the corresponding quantity of single frames mood is most in image clustering set, using the at most corresponding single frames mood of quantity as mesh
Mark mood.Specifically, the corresponding quantity of each single frames mood in statistical picture cluster set, by the corresponding single frames feelings of maximum quantity
Target emotion of the thread as client corresponding with image clustering set.In the present embodiment, successively get and commodity pair to be presented
The target emotion of the corresponding client of each image clustering set, that is, get each client and wait for this in the target video data answered
Show the mood of commodity.For example, including 100 image clustering set in target video data corresponding with A commodity, obtain each
The target emotion of the corresponding client of image clustering set gets 100 clients to the target emotion of A commodity.
S50: according to the target emotion of customer quantity and the corresponding client of each image clustering set, final mood is obtained.
Wherein, final mood, which refers to, carries out quantitative analysis by target emotion of each client to the commodity to be presented, obtains
The mood corresponding with commodity to be presented got.
Specifically, server-side judges that all images are poly- according to the target emotion of the corresponding client of each image clustering set
Whether the target emotion of the corresponding client of class set is identical, if the target emotion of the corresponding client of image clustering set is different,
According to the target emotion of customer quantity and the corresponding client of each image clustering set, the corresponding number of each target emotion is counted
Amount, using the maximum target emotion of quantity as final mood.For example, the corresponding customer quantity of the target video data of A commodity is
100, i.e. image clustering set is also 100, if the corresponding target emotion of 100 image clustering set, 50 are happiness, and 30 are
Calmness, 20 are indifference, then the final mood by the maximum corresponding target emotion (i.e. happy) of quantity as A commodity.If number
The maximum corresponding target emotion of amount is at least two, then determines final mood based on the quantity of the mood classification of target emotion.It is excellent
Target emotion is the final mood of conduct of active mood by selection of land.For example, the corresponding client's number of the target video data of A commodity
Amount be 100, i.e. image clustering set is also 100, if the corresponding target emotion of 100 image clustering set, 50 for happiness, 50
A is indifference, then is the happy final mood as the commodity to be presented by target emotion.If quantity maximum arranged side by side is corresponding
Target emotion is at least two, and target emotion is negative emotions, since the intended display commodity finally shown are corresponding most
Whole mood should be active mood, then any one target emotion may be selected as final mood.
S60: according to the corresponding final mood of commodity to be presented, intended display commodity are obtained.
Wherein, intended display commodity refer to according to the corresponding final mood of each commodity to be presented, from commodity to be presented
Get the commodity that can be shown.
Specifically, server-side gets mesh from commodity to be presented according to the corresponding final mood of each commodity to be presented
Mark shows that commodity specifically comprise the following steps:
(1) first determine whether the corresponding final mood of each commodity to be presented is the commodity that can be shown.Specifically, in advance
The default mood that can be shown first is set, the corresponding final mood of each commodity to be presented is matched with default mood,
It is that can be shown by the corresponding commodity to be presented of the final mood of successful match if final mood and default mood successful match
Commodity.By the step, avoid target emotion and the unmatched commodity to be presented of default mood as intended display commodity.One
As for, preset mood be active mood, then avoid using final mood be the corresponding commodity to be presented of negative emotions as mesh
Mark shows commodity.For example, the corresponding final mood of A commodity to be presented is happiness, then most of client is to the commodity to be presented
It is more interested, happiness is matched with default mood, if happy and default mood successful match, A commodity are determined as
The commodity that can be shown.For another example the corresponding final mood of B commodity to be presented is to detest, is angry or disappointed, then it is most of
Client does not like the commodity to be presented, and by the corresponding final mood of B commodity to be presented, it fails to match with default mood, then
B not can be used as the commodity of displaying.
(2) quantity for the commodity that determination can be shown, if the quantity for the commodity that can be shown be greater than preset value, according to it is each can
The corresponding final mood of the commodity of displaying inquires mood sequencing table, and the corresponding commodity to be presented of final mood are ranked up, are obtained
Take the commodity to be presented of preset value as intended display commodity.Wherein, mood sequencing table is preset table, more positive feelings
Before thread comes more, for example, being ranked up according to happy, happy, surprised, tranquil, detest, angry and disappointment etc..For example, can carry out
The quantity of the commodity to be presented shown is 4, and preset value 3 then obtains the corresponding final feelings of commodity to be presented that can be shown
Thread, the corresponding final mood of A commodity to be presented are happiness, and the corresponding final mood of B commodity to be presented is joy, C quotient to be presented
The corresponding final mood of product be it is surprised, the corresponding final mood of D commodity to be presented be calmness, then being treated according to mood sequencing table
Show that commodity are ranked up, as A, B, C and D, before the row of acquisition the commodity to be presented of preset value are arranged as intended display commodity
The commodity to be presented of preceding 3 A, B and C are as intended display commodity.
Further, if according to mood sequencing table, occur simultaneously when the corresponding commodity to be presented of final mood are ranked up
The case where column, first judges whether to get intended display commodity in preset value；If target exhibition cannot be got in preset value
Show commodity, it is determined that quantity (the i.e. target emotion to match in the target video data of commodity to be presented with final mood arranged side by side
Maximum quantity), the limited grade of final mood side by side is determined according to the quantity, obtains the intended display commodity of preset value.For example,
The quantity for the commodity to be presented that can be shown is 4, preset value 3, then it is corresponding to obtain the commodity to be presented that can be shown
Final mood, the corresponding final mood of A commodity to be presented are happiness, and the corresponding final mood of B commodity to be presented is joy, and C is waited for
Show that the corresponding final mood of commodity is calmness, the corresponding final mood of D commodity to be presented is calmness, then sorting according to mood
Table is ranked up commodity to be presented, as A, B, C and D, and C is arranged side by side with D, and intended display quotient cannot be got in preset value
Product then obtain the quantity that final mood corresponding with C commodity to be presented matches in the target video data of C commodity to be presented,
If the quantity is 50；In the target video data for determining D commodity to be presented again, final mood phase corresponding with D commodity to be presented
Matched quantity, if the quantity is 60；It is then 60 corresponding D commodity to be presented by quantity prior to quantity is 50 corresponding C
Commodity to be presented obtain the commodity to be presented of preset value as intended display commodity, i.e., the commodity to be presented of A, B and D are as mesh
Mark shows commodity.By the step, final mood is avoided situation arranged side by side occur, to improve intended display commodity constant speed really
Degree.
In step S10-S60, the target video data of commodity to be presented is obtained, using Face datection model to target video
Images to be recognized carries out recognition of face and cluster in data, obtains the corresponding customer quantity of target video data and each client couple
The image clustering set answered improves mesh so that the subsequent target emotion according to a certain number of clients determines intended display commodity
Mark shows the accuracy rate of commodity.If customer quantity is greater than preset quantity, using preparatory trained micro- Expression Recognition model pair
Images to be recognized in each image clustering set is identified, the single frames mood of each images to be recognized is obtained, and is known with realizing
The mood of other client.Based on the single frames mood of an at least frame images to be recognized, obtain client's corresponding with image clustering set
Target emotion, to determine whether the commodity to be presented are commodity interested to client.It is poly- according to customer quantity and each image
The target emotion of the corresponding client of class set, to get a certain number of clients to the final mood of the commodity to be presented, according to
According to the corresponding final mood of commodity to be presented, intended display commodity are obtained, the accuracy rate of intended display commodity are improved, so that target
Commodity are the more interested commodity of most of clients.
In one embodiment, as shown in figure 3, in step S10, that is, the target video data of commodity to be presented, target are obtained
Video data includes at least two frame images to be recognized, is specifically comprised the following steps:
S11: obtaining the initial video data of commodity to be presented, and initial video data includes at least two frame initial video figures
Picture.
Wherein, initial video data refers to the video counts corresponding with each commodity to be presented of video acquisition tool acquisition
According to.
Specifically, the corresponding initial video data of each commodity to be presented, initial video are acquired using video acquisition tool
Include at least two frame initial video images in data, include the corresponding commodity sign of commodity to be presented in initial video data,
It can determine initial video data corresponding with each commodity to be presented according to the commodity sign.By acquiring each commodity to be presented
Corresponding initial video data analyzes initial video data so as to subsequent, and it is corresponding to get each commodity to be presented
Final mood.
S12: obtaining the attribute information of commodity to be presented, and attribute information includes Suitable Age and suitable gender.
Wherein, Suitable Age referred to suitable for the commodity to be presented corresponding age.Suitable gender refers to that be suitable for should be to
Show the corresponding gender of commodity.
Specifically, the corresponding attribute information of each commodity to be presented is stored in database.Server-side is according to quotient to be presented
Product search database, obtain attribute information corresponding with each commodity to be presented.Wherein, in attribute information comprising Suitable Age and
It is suitable for gender.For example, a certain commodity to be presented are dress ornament, Suitable Age is 20-24 years old in the corresponding attribute information of the dress ornament, is fitted
The women that suitable gender is.For another example commodity to be presented are cosmetics, Suitable Age is in the corresponding attribute information of the cosmetics
It is male etc. that 25-30 years old, which is suitable for gender,.Commodity to be presented are not done in the present embodiment specifically defined.
S13: screening initial video image using Suitable Age and suitable gender, obtains images to be recognized, is based on
At least two frame images to be recognized form target video data.
Specifically, using preparatory trained classifier at least two frame initial video images in initial video data into
Row identification obtains target age corresponding with each initial video image and target gender, and server-side is by target age and is suitable for
Age is matched, and target gender is matched with suitable gender, successful match and and suitability will be carried out with Suitable Age
Not carry out the initial video image of successful match be determined as images to be recognized, unsuccessful initial video image will be matched and deleted
It removes, forms target video data based at least two frame images to be recognized.Wherein, target age refers to by trained point in advance
Class device identifies initial video image, the accessed age.Target gender refers to through preparatory trained classifier pair
Initial video image identified, accessed gender.
Step S11-S13 carries out initial video image according to the corresponding Suitable Age of commodity to be presented and suitable gender
Screening, obtain images to be recognized, based at least two frame images to be recognized formed target video data so that target video data with
Commodity to be presented more match, and improve the accuracy rate of intended display commodity.
In one embodiment, before step S3, i.e., using Suitable Age and suitable gender to initial video image into
Before the step of row screening, the merchandise display method based on image recognition further include:
(1) at least two frame initial video images are handled using super-resolution technique, is obtained initial at least two frames
The corresponding high-definition picture of video image, using high-definition picture as image to be determined.
Wherein, super-resolution technique (Super-Resolution, SR) refers to rebuilds from the low-resolution image got
Corresponding high-definition picture out.In general, initial video image is low-resolution image in initial video data.Image to be determined
Refer to and initial video image is converted into high-resolution image.
Specifically, server-side obtains initial video data, and initial video data includes at least two frame initial video images, just
Beginning video image extracts the characteristic pattern of low-resolution spatial by ESPCN algorithm, by effective in the space low resolution (LR)
Initial video image is amplified high-resolution from low resolution, final low resolution characteristic pattern is upgraded by sub-pix convolutional layer
To high-resolution features figure, high-definition picture corresponding with each initial video image is obtained according to high-resolution features figure,
Using high-definition picture as image to be determined.Wherein, the key concept of ESPCN algorithm is sub-pix convolutional layer (sub-pixel
Convolutional layer), input is low-resolution image (i.e. initial video image), passes through two sub-pix convolutional layers
After, obtained characteristic image size is as input picture, but feature channel is that (r is the target times magnification of image to r^2
Number).R^2 channel of each pixel is rearranged into the region of a r x r, corresponding to one in high-definition picture
The sub-block of r x r size, so that size is that the characteristic image of r^2x H x W is rearranged into the height of 1x rH x rW size
Resolution characteristics figure obtains high-definition picture corresponding with each frame initial video image according to high-resolution features figure, will
High-definition picture is as image to be determined.
As shown in figure 4, being screened, being obtained to initial video image using Suitable Age and suitable gender in step S13
Images to be recognized is taken, is specifically comprised the following steps:
S131: identifying at least two frames image to be determined using preparatory trained classifier, obtain with it is each to
Determine the corresponding target age of image and target gender.
Wherein, trained classifier includes gender sorter and character classification by age device in advance, passes through gender sorter and year
Age classifier respectively identifies image to be determined, to get target age corresponding with image to be determined and Objective
Not.Wherein, target gender, which refers to, identifies image to be determined by gender sorter, accessed gender.Target year
Age, which refers to, identifies image to be determined by character classification by age device, the accessed age.
Wherein, when gender sorter and character classification by age device are trained, a large amount of training image data, training figure are first obtained
Facial image as in data including all ages and classes and different sexes, carries out year for each facial image in training image data
Training image data after mark are input to deep neural network by age and gender mark, and by deep neural network to mark
Training image data after note are trained, since deep neural network includes at least two convolutional layers, by age predicted value and
The age marked is compared, and with the weight and deviation of each layer in percentage regulation neural network, until model is restrained, obtains year
Age classifier.Gender prediction's value and the gender marked are compared, with the weight of each layer in percentage regulation neural network and
Deviation obtains gender sorter until model is restrained.
Specifically, image to be determined is identified using preparatory trained gender sorter, image to be determined is packet
Image to be determined comprising client face is carried out face critical point detection and feature extraction, to obtain by the image of the face containing client
Take face feature.Finally, the face feature of extraction is input in preparatory trained gender sorter, pass through gender sorter
Face feature is identified, to obtain target gender corresponding with image to be determined.And the face feature that will be extracted is inputted
Into preparatory trained character classification by age device, classified by character classification by age device to face feature, to obtain and figure to be determined
As corresponding target age.Visitor is carried out to people's image to be determined by preparatory trained gender sorter and character classification by age device
Family gender and age are estimated, and obtain accuracy to promote target gender and target age.
S132: target age is matched with Suitable Age, and target gender is matched with suitable gender.
Specifically, Suitable Age can be an age bracket, for example, 20-24 years old.Server-side is by target age and suitable year
Whether age is matched, mainly judge the target age within the scope of Suitable Age.Be suitable for gender be women and two kinds of male,
The target gender that will identify that is matched with suitable gender.
S133: if target age and Suitable Age successful match, and target gender and suitable gender successful match then will be with
Target age and the corresponding image to be determined of target gender are as images to be recognized.
Specifically, server-side judge target age in Suitable Age segment limit, and target gender and suitable gender
With success, then will image to be determined corresponding with target age and target gender as images to be recognized.
Step S131-S132 identifies at least two frames image to be determined using preparatory trained classifier, obtains
Target age corresponding with each image to be determined and target gender, to realize through classifier to target age and target gender
Determination, improve intended display commodity acquisition speed.If target age and Suitable Age successful match, and target gender and suitable
Suitable gender successful match, then will image to be determined corresponding with target age and target gender as images to be recognized so that obtaining
The images to be recognized got is corresponding with the attribute information of commodity to be presented, to improve the crowd's to match with product to be presented
Attraction, so that the intended display commodity got are more accurate, improves intended display by analyzing images to be recognized
The acquisition accuracy rate of commodity.
In one embodiment, as shown in figure 5, step S20, that is, use Face datection model at least two frame images to be recognized
Recognition of face and cluster are carried out, the corresponding customer quantity of target video data and the corresponding image clustering collection of each client are obtained
It closes, specifically comprises the following steps:
S21: recognition of face is carried out at least two frame images to be recognized using Face datection model, obtains each figure to be identified
As corresponding facial image.
Specifically, server-side obtains target video data, using Face datection model to frame each in target video data
Images to be recognized carries out recognition of face, gets the corresponding facial image of each images to be recognized in target video data.Wherein,
Recognition of face refers to the image given for any one frame, use certain strategy to scan for it with determine in image whether
Contain face.Further, Face datection model be in advance it is trained for detect each frame images to be recognized whether include
The model of the facial area of people.Specifically, each frame images to be recognized is input in Face datection model by server-side, and detection is every
Whether in one frame images to be recognized includes face, if in images to be recognized including face, is obtained every in target video data
The corresponding facial image of one images to be recognized.
S22: clustering the corresponding facial image of images to be recognized, obtains at least two image clustering set, each
Image clustering set includes an at least frame images to be recognized.
Specifically, the corresponding facial image of images to be recognized that server-side will acquire is clustered, and will include identical visitor
The facial image at family is clustered, and at least two image clustering set are obtained, wherein includes at least in each image clustering set
One frame images to be recognized.More specifically, extracting the people of the corresponding facial image of each images to be recognized using feature extraction algorithm
The corresponding face characteristic of facial image is carried out characteristic similarity calculating by face feature, if characteristic similarity is greater than preset threshold,
Illustrate the facial image for same client, and the corresponding images to be recognized of the facial image of same client is clustered, to obtain
Take the corresponding image clustering set of each client.That is, the corresponding image clustering set of a client, each image clustering set
In include an at least frame images to be recognized.
S23: according to the quantity of image clustering set, the corresponding customer quantity of target video data is obtained.
Specifically, the quantity for counting the corresponding image clustering set of each commodity to be presented, by the number of image clustering set
Amount is as customer quantity corresponding with target video data.
Step S21-S23 carries out recognition of face at least two frame images to be recognized using Face datection model, obtains each
The corresponding facial image of images to be recognized is avoided with realizing determine whether images to be recognized is facial image not comprising face
Images to be recognized is clustered, and the acquisition speed of subsequent image cluster set is improved.Facial image corresponding to images to be recognized
It is clustered, obtains at least two image clustering set, according to the quantity of image clustering set, it is corresponding to obtain target video data
Customer quantity guarantee the accuracy that customer quantity obtains to realize the determination of customer quantity.
In one embodiment, the corresponding time identifier of each images to be recognized, time identifier refer to that collect this to be identified
The image corresponding time.
As shown in fig. 6, clustering to the corresponding facial image of images to be recognized in step S22, at least two are obtained
Image clustering set, specifically comprises the following steps:
S221: according to time identifier, using the facial image recognized for the first time at least two frame images to be recognized as benchmark
Image.
Wherein, benchmark image refers to the facial image recognized for the first time from images to be recognized.
Specifically, server-side obtains the corresponding time identifier of at least two frame images to be recognized and is first determined according to time identifier
The facial image recognized for the first time at least two frame images to be recognized, using the facial image as benchmark image.By determining base
Quasi- image, to improve the acquisition speed of image clustering set.
S222: according to time identifier, successively using the spy of similarity algorithm calculating benchmark image and remaining images to be recognized
Levy similarity.
Specifically, it according to time identifier, will be removed in benchmark image and at least two frame images to be recognized using similarity algorithm
Remaining images to be recognized except benchmark image carries out characteristic similarity calculating, to obtain characteristic similarity.Wherein, similarity operator
Method can be Euclidean distance algorithm, manhatton distance algorithm, Minkowski distance algorithm or cosine similarity algorithm
Deng.It in the present embodiment, can using the characteristic similarity of cosine similarity algorithm calculating benchmark image and remaining images to be recognized
Accelerate the acquisition speed of image clustering set, improves the acquisition efficiency of intended display commodity.
S223: if characteristic similarity be greater than preset threshold, by characteristic similarity be greater than preset threshold images to be recognized and
Benchmark image belongs to same image clustering set.
Specifically, preset threshold is preset value, if server-side judges benchmark image and remaining images to be recognized
Characteristic similarity be greater than preset threshold, then it is assumed that benchmark image and remaining images to be recognized successful match, benchmark image and surplus
Remaining images to be recognized is the image of same client, then characteristic similarity is greater than to the images to be recognized and benchmark image of preset threshold
Belong to same image clustering set.For example, the characteristic similarity corresponding with remaining images to be recognized 2 of benchmark image 1 is 80%,
The characteristic similarity corresponding with remaining images to be recognized 3 of benchmark image 1 is 99%, preset threshold 90%, then benchmark image 1
Characteristic similarity corresponding with remaining images to be recognized 3 is greater than preset threshold, then by benchmark image 1 and remaining images to be recognized 3
Belong to same image clustering set.
S224: if characteristic similarity is not more than preset threshold, according to time identifier, characteristic similarity is not more than default threshold
First in the remaining images to be recognized of value is updated to new benchmark image, repeats according to time identifier, successively uses
The step of similarity algorithm calculating benchmark image and the characteristic similarity of remaining images to be recognized, until at least two frames figure to be identified
As cluster set completion, at least two image clustering set are formed.
Specifically, server-side judges that the characteristic similarity of benchmark image and remaining images to be recognized is not more than default threshold
Value, then it is assumed that it fails to match for benchmark image and remaining images to be recognized, the corresponding client of benchmark image and remaining images to be recognized
Corresponding client is not same client, then according to time identifier, characteristic similarity is to be identified no more than the residue of preset threshold
First in image is updated to new benchmark image.For example, the feature phase corresponding with remaining images to be recognized 2 of benchmark image 1
It is 80% like degree, preset threshold 90%, then the characteristic similarity corresponding with remaining images to be recognized 2 of benchmark image 1 is little
In preset threshold, then according to time identifier, remaining images to be recognized 2 is updated to new benchmark image.It repeats according to the time
The step of identifying, successively using similarity algorithm calculating benchmark image and the characteristic similarity of remaining images to be recognized, until extremely
All images to be recognized cluster set are completed in few two frame images to be recognized, form at least two image clustering set.
Step S221-S224, according to time identifier, the facial image that will be recognized for the first time at least two frame images to be recognized
As benchmark image, the characteristic similarity of similarity algorithm calculating benchmark image and remaining images to be recognized is successively used, so as to
Determine whether benchmark image and remaining images to be recognized are same client.If characteristic similarity is greater than preset threshold, by feature phase
It is greater than the images to be recognized of preset threshold like degree and benchmark image belongs to same image clustering set, realizes same client
Images to be recognized clustered.If characteristic similarity is not more than preset threshold, according to time identifier, characteristic similarity is little
First in the remaining images to be recognized of preset threshold is updated to new benchmark image, repeats according to time identifier,
Successively using similarity algorithm calculating benchmark image and the characteristic similarity of remaining images to be recognized the step of, until at least two frames
Images to be recognized cluster set is completed, and at least two image clustering set are formed, to realize the images to be recognized of same client
It is clustered, so that each client of subsequent determination is to the target emotion of the commodity to be presented.
In implementing one, as shown in fig. 7, in step S30, i.e., using preparatory trained micro- Expression Recognition model to each
Images to be recognized in image clustering set is identified, is obtained the single frames mood of each images to be recognized, is specifically included as follows
Step:
S31: recognition of face is carried out to the images to be recognized in each image clustering set using face key point algorithm, is obtained
Take face key point corresponding with each images to be recognized.
Wherein, face key point algorithm can be but not limited to Ensemble ofRegression Tress (abbreviation ERT)
Algorithm, SIFT (scale-invariant feature transform) algorithm, SURF (Speeded Up Robust
Features) algorithm, LBP (Local Binary Patterns) algorithm and HOG (Histogram ofOriented
Gridients) algorithm.In the present embodiment, using ERT algorithm, people is carried out to the images to be recognized in each image clustering set
Face identification, to get the corresponding face key point of each images to be recognized.Wherein, ERT algorithm is a kind of side based on recurrence
Method, ERT algorithm are formulated as follows:Wherein,It is obtained for the t+1 times iteration
The shape or coordinate of the characteristic point of images to be recognized, t indicate cascade serial number,For predict image characteristic point shape or
Coordinate, I are the images to be recognized for returning device input, rtIndicate t grades of recurrence device, each recurrence device is by many regression trees
(tree) it forms, by the available regression tree of training, it is crucial that the corresponding face of each images to be recognized is got by regression tree
Point.
S32: feature extraction is carried out to the corresponding face key point of each images to be recognized using feature extraction algorithm, is obtained
The corresponding local feature of face key point.
Wherein, feature extraction algorithm can be calculated with CNN (Convolutional Neural Network, convolutional Neural net)
Method extracts the local feature of images to be recognized corresponding face key point position by CNN algorithm, specifically according to face action
Local feature is extracted in the corresponding position of unit.Wherein, CNN algorithm is a kind of feedforward neural network, its artificial neuron can be with
The surrounding cells in a part of coverage area are responded, image procossing can be rapidly and efficiently carried out.In the present embodiment, using preparatory
Trained convolutional neural networks rapidly extract the corresponding local feature of face key point.
Specifically, by the corresponding face key point of each images to be recognized, convolution algorithm is carried out by several convolution kernels,
Result after convolution is the corresponding local feature of face test point.Particular by formula
(0≤m < M, 0≤n < N) carries out convolution algorithm, to get local feature.Wherein, y is the local feature of output, and x is one
Size is the two dimension input variable of (M, N), i.e., is formed by the coordinate of L face key point, wijBe size be I*J convolution kernel, b
For biasing, size M*N, activation primitive is indicated with f, and the face of input images to be recognized of each convolution kernel with upper one layer closes
Key point carries out convolution operation, and each convolution kernel can have a corresponding local feature, and weight is shared in convolution kernel, number of parameters
Reduce significantly, with greatly improving network training speed.
Further, face key point is input to after being identified in preset convolutional neural networks, available face
The corresponding local feature of portion's motor unit.For example, AU1, AU2, AU5 and AU26, i.e., interior eyebrow raises up, outer eyebrow raises up, on upper eyelid
It raises and opens corresponding local feature with lower jaw.It is crucial to face in images to be recognized using convolutional neural networks in the present embodiment
The local feature of point extracts, determine target face motor unit according to local feature so as to subsequent, and according to recognizing
Target face motor unit determines the mood of client.In this motion, relative to LBP-TOP operator, convolutional Neural net is used
Faster, and accuracy of identification is higher for the arithmetic speed that network is identified.
S33: local feature is identified using preparatory trained classifier, obtains the corresponding mesh of each local feature
Mark Facial action unit.
Specifically, local feature is known by each SVM classifier in preparatory trained micro- Expression Recognition model
Not, wherein SVM classifier is identical as the quantity of identifiable Facial action unit, i.e., identifiable Facial action unit is 54
It is a, then trained SVM classifier is 54 in advance, by the way that local feature to be input in corresponding SVM classifier, obtain
To probability value, probability value and the predetermined probabilities threshold value that will acquire are compared, and will be greater than the probability value pair of predetermined probabilities threshold value
The Facial action unit answered is got corresponding with local feature as target face motor unit corresponding with local feature
All target face motor units.
S34: being based on the corresponding target face motor unit of each local feature, searches assessment table, obtains each to be identified
The single frames mood of image.
Wherein, assessment table is preconfigured table, and it is corresponding with mood that Facial action unit combination is stored in assessment table
Relationship, for example, AU12, AU6 and AU7 are combined, corresponding mood is happy, AU9, AU10, AU17 and AU24, and corresponding mood is
Detest.Server-side searches assessment table by the corresponding target face motor unit of each local feature, obtains dynamic with target face
The combination to match as unit, using the corresponding mood of the combination as single frames mood corresponding with images to be recognized.
In step S31-S34, the images to be recognized in each image clustering set is carried out using face key point algorithm
Recognition of face obtains face key point corresponding with each images to be recognized, provides technical assistance for subsequent extracted local feature,
To improve the accuracy rate of local shape factor；Feature extraction is carried out to face key point using feature extraction algorithm, quickly to obtain
The corresponding local feature of face key point is got, so that target face motor unit that subsequent extracted arrives is more accurate；Using pre-
First trained classifier identifies local feature, single with the corresponding target face movement of each local feature of quick obtaining
Member realizes the determination of target face motor unit.Based on the corresponding target face motor unit of each local feature, assessment is searched
Table obtains the single frames mood of each images to be recognized, to determine the corresponding mood of each client according to single frames mood, according to every
The corresponding mood of one client determines intended display commodity, to improve the accuracy rate of intended display commodity.
It should be understood that the size of the serial number of each step is not meant that the order of the execution order in above-described embodiment, each process
Execution sequence should be determined by its function and internal logic, the implementation process without coping with the embodiment of the present invention constitutes any limit
It is fixed.
In one embodiment, a kind of commodity-exhibiting device based on image recognition is provided, it should the commodity based on image recognition
Show that the merchandise display method in device and above-described embodiment based on image recognition corresponds.As shown in figure 8, image should be based on
The commodity-exhibiting device bag data of identification obtains module 10, image clustering set obtains module 20, single frames mood determining module 30,
Target emotion obtains module 40, final mood obtains module 50 and intended display commodity obtain module 60.Each functional module is detailed
It is described as follows:
Data acquisition module 10, for obtaining the target video data of commodity to be presented, target video data includes at least
Two frame images to be recognized.
Image clustering set obtains module 20, for carrying out people at least two frame images to be recognized using Face datection model
Face identification and cluster obtain the corresponding customer quantity of target video data and the corresponding image clustering set of each client, each
Image clustering set includes an at least frame images to be recognized.
Single frames mood determining module 30, if being greater than preset quantity for customer quantity, using preparatory trained micro- table
Feelings identification model identifies the images to be recognized in each image clustering set, obtains the single frames feelings of each images to be recognized
Thread.
Target emotion obtains module 40, for the single frames mood based on an at least frame images to be recognized, obtains poly- with image
The target emotion of the corresponding client of class set.
Final mood obtains module 50, for the target according to customer quantity and the corresponding client of each image clustering set
Mood obtains final mood.
Intended display commodity obtain module 60, for obtaining intended display according to the corresponding final mood of commodity to be presented
Commodity.
In one embodiment, data acquisition module 10, including initial video data acquiring unit 11, attribute information determine list
Member 12 and target video data form unit 13.
Initial video data acquiring unit 11, for obtaining the initial video data of commodity to be presented, initial video data
Including at least two frame initial video images.
Attribute information determination unit 12, for obtaining the attribute information of commodity to be presented, attribute information includes Suitable Age
With suitable gender.
Target video data forms unit 13, for being sieved using Suitable Age and suitable gender to initial video image
Choosing obtains images to be recognized, forms target video data based at least two frame images to be recognized.
In one embodiment, before target video data forms unit, the commodity-exhibiting device based on image recognition is also
Including image resolution ratio converting unit.
Image resolution ratio converting unit, for using super-resolution technique at least two frame initial video images at
Reason obtains high-definition picture corresponding at least two frame initial video images, using high-definition picture as image to be determined.
Target video data formed unit, including target age and target gender determine subelement, coupling subelement and to
Identification image determines subelement.
Target age and target gender determine subelement, for being waited for really using preparatory trained classifier at least two frames
Determine image to be identified, obtains target age corresponding with each image to be determined and target gender.
Coupling subelement, for target age to be matched with Suitable Age, and by target gender and suitable gender into
Row matching.
Images to be recognized determines subelement, if for target age and Suitable Age successful match, and target gender and suitable
Suitable gender successful match, then will image to be determined corresponding with target age and target gender as images to be recognized.
In one embodiment, image clustering set obtains module 20, including facial image acquiring unit, image clustering set
Acquiring unit and customer quantity determination unit.
Facial image acquiring unit, for carrying out face knowledge at least two frame images to be recognized using Face datection model
Not, the corresponding facial image of each images to be recognized is obtained.
Image clustering set acquiring unit obtains at least for clustering to the corresponding facial image of images to be recognized
Two image clustering set, each image clustering set include an at least frame images to be recognized.
Customer quantity determination unit obtains the corresponding visitor of target video data for the quantity according to image clustering set
Amount amount.
In one embodiment, the corresponding time identifier of each images to be recognized.
Image clustering set acquiring unit includes that benchmark image determines subelement, characteristic similarity computing unit, the first figure
Gather as cluster and determines that subelement and the second image clustering set determine subelement.
Benchmark image determines subelement, for that will recognize for the first time at least two frame images to be recognized according to time identifier
Facial image as benchmark image.
Characteristic similarity computing unit, for according to time identifier, successively using similarity algorithm calculating benchmark image with
The characteristic similarity of remaining images to be recognized.
First image clustering set determines subelement, if being greater than preset threshold for characteristic similarity, by characteristic similarity
Same image clustering set is belonged to greater than the images to be recognized of preset threshold and benchmark image.
Second image clustering set determines subelement, if being not more than preset threshold for characteristic similarity, is marked according to the time
Know, characteristic similarity is updated to new benchmark image no more than first in the remaining images to be recognized of preset threshold, weight
It is multiple to execute according to time identifier, successively using the characteristic similarity of similarity algorithm calculating benchmark image and remaining images to be recognized
The step of, until at least two frame images to be recognized cluster set is completed, form at least two image clustering set.
In one embodiment, single frames mood determining module 30, including face key point acquiring unit, local shape factor list
Member, target face motor unit acquiring unit, single frames mood acquiring unit.
Face key point acquiring unit, for using face key point algorithm to be identified in each image clustering set
Image carries out recognition of face, obtains face key point corresponding with each images to be recognized.
Local shape factor unit, for using feature extraction algorithm to the corresponding face key point of each images to be recognized
Feature extraction is carried out, the corresponding local feature of face key point is obtained.
Target face motor unit acquiring unit, for being known using preparatory trained classifier to local feature
Not, the corresponding target face motor unit of each local feature is obtained.
Single frames mood acquiring unit searches assessment for being based on the corresponding target face motor unit of each local feature
Table obtains the single frames mood of each images to be recognized.
Specific restriction about the commodity-exhibiting device based on image recognition may refer to know above for based on image
The restriction of other merchandise display method, details are not described herein.Each mould in the above-mentioned commodity-exhibiting device based on image recognition
Block can be fully or partially through software, hardware and its group and to realize.Above-mentioned each module can be embedded in the form of hardware or independence
In processor in computer equipment, it can also be stored in a software form in the memory in computer equipment, in order to
Processor, which calls, executes the corresponding operation of the above modules.
In one embodiment, a kind of computer equipment is provided, which can be server-side, internal junction
Composition can be as shown in Figure 9.The computer equipment include by system bus connect processor, memory, network interface and
Database.Wherein, the processor of the computer equipment is for providing calculating and control ability.The memory packet of the computer equipment
Include non-volatile memory medium, built-in storage.The non-volatile memory medium is stored with operating system, computer program and data
Library.The built-in storage provides environment for the operation of operating system and computer program in non-volatile memory medium.The calculating
The database of machine equipment is for storing Face datection model and the attribute information of commodity to be presented etc..The network of the computer equipment
Interface is used to communicate with external terminal by network connection.To realize that one kind is based on when the computer program is executed by processor
The merchandise display method of image recognition.
In one embodiment, a kind of computer equipment is provided, including memory, processor and storage are on a memory simultaneously
The computer program that can be run on a processor, processor is realized when executing computer program in above-described embodiment to be known based on image
The step of other merchandise display method, for example, step S10 shown in Fig. 2 to step S60, alternatively, Fig. 3 is to step shown in Fig. 7
Suddenly.Processor realizes each mould in the commodity-exhibiting device in above-described embodiment based on image recognition when executing computer program
Block/unit function, for example, function of the module 10 shown in Fig. 8 to module 50.To avoid repeating, details are not described herein again.
In one embodiment, a kind of computer readable storage medium is provided, computer program, computer are stored thereon with
The merchandise display method based on image recognition in above method embodiment is realized when program is executed by processor, for example, shown in Fig. 2
Step S10 to step S60, alternatively, Fig. 3 is to step shown in Fig. 7.The computer program is realized above-mentioned when being executed by processor
The function of each module/unit in commodity-exhibiting device in embodiment based on image recognition, for example, module 10 shown in Fig. 8 is to mould
The function of block 60.To avoid repeating, details are not described herein again.
Those of ordinary skill in the art will appreciate that realizing all or part of the process in above-described embodiment method, being can be with
Relevant hardware is instructed to complete by computer program, computer program can be stored in a non-volatile computer and can be read
In storage medium, the computer program is when being executed, it may include such as the process of the embodiment of above-mentioned each method.Wherein, the application
To any reference of memory, storage, database or other media used in provided each embodiment, may each comprise non-
Volatibility and/or volatile memory.Nonvolatile memory may include read-only memory (ROM), programming ROM (PROM),
Electrically programmable ROM (EPROM), electrically erasable ROM (EEPROM) or flash memory.Volatile memory may include arbitrary access
Memory (RAM) or external cache.By way of illustration and not limitation, RAM is available in many forms, such as static
RAM (SRAM), dynamic ram (DRAM), synchronous dram (SDRAM), double data rate sdram (DDRSDRAM), enhanced SDRAM
(ESDRAM), synchronization link (Synchlink) DRAM (SLDRAM), memory bus (RambuS) directly RAM (RDRAM), straight
Connect memory bus dynamic ram (DRDRAM) and memory bus dynamic ram (RDRAM) etc..
It is apparent to those skilled in the art that for convenience of description and succinctly, only with above-mentioned each function
Can unit, module division progress for example, in practical application, can according to need and by above-mentioned function distribution by different
Functional unit, module are completed, i.e., the internal structure of device are divided into different functional unit or module, to complete above description
All or part of function.
The above embodiments are merely illustrative of the technical solutions of the present invention, rather than its limitations；Although with reference to the foregoing embodiments
Invention is explained in detail, those skilled in the art should understand that: it still can be to aforementioned each implementation
Technical solution documented by example is modified or equivalent replacement of some of the technical features；And these modification or
Replacement, the spirit and scope for technical solution of various embodiments of the present invention that it does not separate the essence of the corresponding technical solution should all include
Within protection scope of the present invention.
Claims (10)
1. a kind of merchandise display method based on image recognition characterized by comprising
The target video data of commodity to be presented is obtained, the target video data includes at least two frame images to be recognized；
Recognition of face and cluster are carried out using Face datection model images to be recognized described at least two frames, obtain the target view
For frequency according to corresponding customer quantity and the corresponding image clustering set of each client, each described image cluster set includes at least
One frame images to be recognized；
If the customer quantity is greater than preset quantity, using preparatory trained micro- Expression Recognition model to each described image
Images to be recognized in cluster set is identified, the single frames mood of each images to be recognized is obtained；
Based on the single frames mood of images to be recognized described in an at least frame, the mesh of client corresponding with described image cluster set is obtained
Mark mood；
The target emotion for gathering corresponding client according to the customer quantity and each described image cluster, obtains final mood；
According to the corresponding final mood of the commodity to be presented, intended display commodity are obtained.
2. the merchandise display method based on image recognition as described in claim 1, which is characterized in that described to obtain quotient to be presented
The target video data of product, the target video data include at least two frame images to be recognized, comprising:
The initial video data of commodity to be presented is obtained, the initial video data includes at least two frame initial video images；
The attribute information of the commodity to be presented is obtained, the attribute information includes Suitable Age and suitable gender；
The initial video image is screened using the Suitable Age and the suitable gender, obtains images to be recognized,
Target video data is formed based on images to be recognized described at least two frames.
3. the merchandise display method based on image recognition as described in claim 1, which is characterized in that described using described suitable
Before the step of suitable age and the suitable gender screen the initial video image, the quotient based on image recognition
Product methods of exhibiting further include:
It is handled using super-resolution technique initial video image described at least two frames, it is initial described in acquisition and at least two frames
The corresponding high-definition picture of video image, using the high-definition picture as image to be determined；
It is described that the initial video image is screened using the Suitable Age and the suitable gender, obtain figure to be identified
Picture, comprising:
Identified using preparatory trained classifier image to be determined described at least two frames, obtain with it is each described to true
Determine the corresponding target age of image and target gender；
The target age is matched with the Suitable Age, and the target gender and the suitable gender are carried out
Match；
If the target age and the Suitable Age successful match, and the target gender is matched into the suitable gender
Function, then will image to be determined corresponding with the target age and the target gender as images to be recognized.
4. the merchandise display method based on image recognition as described in claim 1, which is characterized in that described to use Face datection
Model images to be recognized described at least two frames carries out recognition of face and cluster, obtains the corresponding client of the target video data
Quantity and the corresponding image clustering set of each client, comprising:
Recognition of face is carried out using Face datection model images to be recognized described at least two frames, obtains each figure to be identified
As corresponding facial image；
The corresponding facial image of the images to be recognized is clustered, at least two image clustering set, each image are obtained
Cluster set includes an at least frame images to be recognized；
According to the quantity of described image cluster set, the corresponding customer quantity of the target video data is obtained.
5. the merchandise display method based on image recognition as claimed in claim 4, which is characterized in that each figure to be identified
As a corresponding time identifier；
It is described that the corresponding facial image of the images to be recognized is clustered, obtain at least two image clustering set, comprising:
According to time identifier, using the facial image recognized for the first time in images to be recognized described at least two frames as benchmark image；
According to time identifier, it is similar to the remaining feature of images to be recognized that the benchmark image is successively calculated using similarity algorithm
Degree；
If the characteristic similarity is greater than preset threshold, the characteristic similarity is greater than to images to be recognized and the institute of preset threshold
It states benchmark image and belongs to same image clustering set；
If the characteristic similarity is not more than preset threshold, according to time identifier, the characteristic similarity is not more than default threshold
First in the remaining images to be recognized of value is updated to new benchmark image, repeat it is described according to time identifier, successively
The step of benchmark image and the characteristic similarity of remaining images to be recognized are calculated using similarity algorithm, until at least two frames
The images to be recognized cluster set is completed, and at least two image clustering set are formed.
6. the merchandise display method based on image recognition as described in claim 1, which is characterized in that described using training in advance
Good micro- Expression Recognition model identifies the images to be recognized in each described image cluster set, obtain it is each it is described to
Identify the single frames mood of image, comprising:
Using face key point algorithm to each described image cluster set in images to be recognized carry out recognition of face, obtain with
The corresponding face key point of each images to be recognized；
Feature extraction is carried out to the corresponding face key point of each images to be recognized using feature extraction algorithm, described in acquisition
The corresponding local feature of face key point；
The local feature is identified using preparatory trained classifier, obtains the corresponding mesh of each local feature
Mark Facial action unit；
Based on the corresponding target face motor unit of each local feature, assessment table is searched, is obtained each described to be identified
The single frames mood of image.
7. a kind of commodity-exhibiting device based on image recognition characterized by comprising
Data acquisition module, for obtaining the target video data of commodity to be presented, the target video data includes at least two
Frame images to be recognized；
Image clustering set obtains module, for carrying out face using Face datection model images to be recognized described at least two frames
Identification and cluster, obtain the corresponding customer quantity of the target video data and the corresponding image clustering set of each client, often
One described image cluster set includes an at least frame images to be recognized；
Single frames mood determining module, if being greater than preset quantity for the customer quantity, using preparatory trained micro- expression
Identification model identifies the images to be recognized in each described image cluster set, obtains each images to be recognized
Single frames mood；
Target emotion obtains module, for the single frames mood based on images to be recognized described in an at least frame, acquisition and described image
Cluster gathers the target emotion of corresponding client；
Final mood obtains module, for gathering the mesh of corresponding client according to the customer quantity and each described image cluster
Mood is marked, final mood is obtained；
Intended display commodity obtain module, for obtaining intended display quotient according to the corresponding final mood of the commodity to be presented
Product.
8. as claimed in claim 7 based on the commodity-exhibiting device of image recognition, which is characterized in that the data acquisition mould
Block, comprising:
Initial video data acquiring unit, for obtaining the initial video data of commodity to be presented, the initial video data packet
Include at least two frame initial video images；
Attribute information determination unit, for obtaining the attribute information of the commodity to be presented, the attribute information includes suitable year
Age and suitable gender；
Target video data forms unit, for using the Suitable Age and the suitable gender to the initial video image
It is screened, obtains images to be recognized, target video data is formed based on images to be recognized described at least two frames.
9. a kind of computer equipment, including memory, processor and storage are in the memory and can be in the processor
The computer program of upper operation, which is characterized in that the processor realized when executing the computer program as claim 1 to
The step of merchandise display method described in 6 any one based on image recognition.
10. a kind of computer readable storage medium, the computer-readable recording medium storage has computer program, and feature exists
In realization is as described in any one of claim 1 to 6 based on the commodity of image recognition when the computer program is executed by processor
The step of methods of exhibiting.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910042541.3A CN109815873A (en) | 2019-01-17 | 2019-01-17 | Merchandise display method, apparatus, equipment and medium based on image recognition |
| PCT/CN2019/120988 WO2020147430A1 (en) | 2019-01-17 | 2019-11-26 | Image identification-based product display method, device, apparatus, and medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910042541.3A CN109815873A (en) | 2019-01-17 | 2019-01-17 | Merchandise display method, apparatus, equipment and medium based on image recognition |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109815873A true CN109815873A (en) | 2019-05-28 |
Family
ID=66604505
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910042541.3A Pending CN109815873A (en) | 2019-01-17 | 2019-01-17 | Merchandise display method, apparatus, equipment and medium based on image recognition |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN109815873A (en) |
| WO (1) | WO2020147430A1 (en) |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110598790A (en) * | 2019-09-12 | 2019-12-20 | 北京达佳互联信息技术有限公司 | Image identification method and device, electronic equipment and storage medium |
| CN110650306A (en) * | 2019-09-03 | 2020-01-03 | 平安科技（深圳）有限公司 | Method and device for adding expression in video chat, computer equipment and storage medium |
| CN111062786A (en) * | 2019-12-25 | 2020-04-24 | 创新奇智(青岛)科技有限公司 | Model updating method based on establishment of commodity appearance characteristic mapping table |
| CN111144241A (en) * | 2019-12-13 | 2020-05-12 | 深圳奇迹智慧网络有限公司 | Target identification method and device based on image verification and computer equipment |
| CN111291623A (en) * | 2020-01-15 | 2020-06-16 | 浙江连信科技有限公司 | Heart physiological characteristic prediction method and device based on face information |
| CN111310602A (en) * | 2020-01-20 | 2020-06-19 | 北京正和恒基滨水生态环境治理股份有限公司 | System and method for analyzing attention of exhibit based on emotion recognition |
| WO2020147430A1 (en) * | 2019-01-17 | 2020-07-23 | 深圳壹账通智能科技有限公司 | Image identification-based product display method, device, apparatus, and medium |
| CN111563503A (en) * | 2020-05-09 | 2020-08-21 | 南宁市第三中学 | Minority culture symbol identification method |
| CN112363624A (en) * | 2020-11-16 | 2021-02-12 | 新之航传媒科技集团有限公司 | Interactive exhibition hall system based on emotion analysis |
| CN113177603A (en) * | 2021-05-12 | 2021-07-27 | 中移智行网络科技有限公司 | Training method of classification model, video classification method and related equipment |
| CN113762156A (en) * | 2021-09-08 | 2021-12-07 | 北京优酷科技有限公司 | Viewing data processing method, device and storage medium |
| CN114153342A (en) * | 2020-08-18 | 2022-03-08 | 深圳市万普拉斯科技有限公司 | Visual information display method and device, computer equipment and storage medium |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112215653A (en) * | 2020-10-12 | 2021-01-12 | 珠海格力电器股份有限公司 | Commodity recommendation method, commodity recommendation device, server and storage medium |
| CN112434711B (en) * | 2020-11-27 | 2023-10-13 | 杭州海康威视数字技术股份有限公司 | Data management method and device and electronic equipment |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106843028A (en) * | 2016-12-04 | 2017-06-13 | 徐州乐泰机电科技有限公司 | A kind of Intelligent clothes cabinet control method comprising advertisement and its intelligent screening |
| WO2018143630A1 (en) * | 2017-02-01 | 2018-08-09 | 삼성전자 주식회사 | Device and method for recommending product |
| CN108509941A (en) * | 2018-04-20 | 2018-09-07 | 北京京东金融科技控股有限公司 | Emotional information generation method and device |
| CN108858245A (en) * | 2018-08-20 | 2018-11-23 | 深圳威琳懋生物科技有限公司 | A kind of shopping guide robot |
| CN109048934A (en) * | 2018-08-20 | 2018-12-21 | 深圳威琳懋生物科技有限公司 | A kind of intelligent shopping guide robot system |
| CN109191190A (en) * | 2018-08-20 | 2019-01-11 | 深圳威琳懋生物科技有限公司 | A kind of control method and computer readable storage medium of shopping guide robot |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109815873A (en) * | 2019-01-17 | 2019-05-28 | 深圳壹账通智能科技有限公司 | Merchandise display method, apparatus, equipment and medium based on image recognition |
-
2019
- 2019-01-17 CN CN201910042541.3A patent/CN109815873A/en active Pending
- 2019-11-26 WO PCT/CN2019/120988 patent/WO2020147430A1/en active Application Filing
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106843028A (en) * | 2016-12-04 | 2017-06-13 | 徐州乐泰机电科技有限公司 | A kind of Intelligent clothes cabinet control method comprising advertisement and its intelligent screening |
| WO2018143630A1 (en) * | 2017-02-01 | 2018-08-09 | 삼성전자 주식회사 | Device and method for recommending product |
| CN108509941A (en) * | 2018-04-20 | 2018-09-07 | 北京京东金融科技控股有限公司 | Emotional information generation method and device |
| CN108858245A (en) * | 2018-08-20 | 2018-11-23 | 深圳威琳懋生物科技有限公司 | A kind of shopping guide robot |
| CN109048934A (en) * | 2018-08-20 | 2018-12-21 | 深圳威琳懋生物科技有限公司 | A kind of intelligent shopping guide robot system |
| CN109191190A (en) * | 2018-08-20 | 2019-01-11 | 深圳威琳懋生物科技有限公司 | A kind of control method and computer readable storage medium of shopping guide robot |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2020147430A1 (en) * | 2019-01-17 | 2020-07-23 | 深圳壹账通智能科技有限公司 | Image identification-based product display method, device, apparatus, and medium |
| CN110650306A (en) * | 2019-09-03 | 2020-01-03 | 平安科技（深圳）有限公司 | Method and device for adding expression in video chat, computer equipment and storage medium |
| CN110598790A (en) * | 2019-09-12 | 2019-12-20 | 北京达佳互联信息技术有限公司 | Image identification method and device, electronic equipment and storage medium |
| CN111144241A (en) * | 2019-12-13 | 2020-05-12 | 深圳奇迹智慧网络有限公司 | Target identification method and device based on image verification and computer equipment |
| CN111062786A (en) * | 2019-12-25 | 2020-04-24 | 创新奇智(青岛)科技有限公司 | Model updating method based on establishment of commodity appearance characteristic mapping table |
| CN111291623A (en) * | 2020-01-15 | 2020-06-16 | 浙江连信科技有限公司 | Heart physiological characteristic prediction method and device based on face information |
| CN111310602A (en) * | 2020-01-20 | 2020-06-19 | 北京正和恒基滨水生态环境治理股份有限公司 | System and method for analyzing attention of exhibit based on emotion recognition |
| CN111563503A (en) * | 2020-05-09 | 2020-08-21 | 南宁市第三中学 | Minority culture symbol identification method |
| CN114153342A (en) * | 2020-08-18 | 2022-03-08 | 深圳市万普拉斯科技有限公司 | Visual information display method and device, computer equipment and storage medium |
| CN112363624A (en) * | 2020-11-16 | 2021-02-12 | 新之航传媒科技集团有限公司 | Interactive exhibition hall system based on emotion analysis |
| CN112363624B (en) * | 2020-11-16 | 2022-09-09 | 新之航传媒科技集团有限公司 | Interactive exhibition hall system based on emotion analysis |
| CN113177603A (en) * | 2021-05-12 | 2021-07-27 | 中移智行网络科技有限公司 | Training method of classification model, video classification method and related equipment |
| CN113177603B (en) * | 2021-05-12 | 2022-05-06 | 中移智行网络科技有限公司 | Training method of classification model, video classification method and related equipment |
| CN113762156A (en) * | 2021-09-08 | 2021-12-07 | 北京优酷科技有限公司 | Viewing data processing method, device and storage medium |
| CN113762156B (en) * | 2021-09-08 | 2023-10-24 | 北京优酷科技有限公司 | Video data processing method, device and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2020147430A1 (en) | 2020-07-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109815873A (en) | Merchandise display method, apparatus, equipment and medium based on image recognition | |
| Pérez-Borrero et al. | A fast and accurate deep learning method for strawberry instance segmentation | |
| US20160350336A1 (en) | Automated image searching, exploration and discovery | |
| Othmani et al. | Age estimation from faces using deep learning: A comparative analysis | |
| WO2016190814A1 (en) | Method and system for facial recognition | |
| Anandhakrishnan et al. | Deep Convolutional Neural Networks for image based tomato leaf disease detection | |
| CN108960167A (en) | Hair style recognition methods, device, computer readable storage medium and computer equipment | |
| CN110413825B (en) | Street-clapping recommendation system oriented to fashion electronic commerce | |
| Karanam et al. | Study on image processing using deep learning techniques | |
| CN110580446A (en) | Behavior semantic subdivision understanding method, system, computer device and medium | |
| Biswas et al. | Recognition of local birds using different cnn architectures with transfer learning | |
| Benhabiles et al. | Deep learning based detection of hair loss levels from facial images | |
| Bondre et al. | Review on leaf diseases detection using deep learning | |
| Guehairia et al. | Deep random forest for facial age estimation based on face images | |
| CN110956157A (en) | Deep learning remote sensing image target detection method and device based on candidate frame selection | |
| Kukreja et al. | Early Detection and Prevention of Mustard Downy Mildew Disease using a Hybrid CNN-LSTM Model | |
| Rahaman et al. | A deep learning based smartphone application for detecting mango diseases and pesticide suggestions | |
| CN114239754A (en) | Pedestrian attribute identification method and system based on attribute feature learning decoupling | |
| Raja et al. | A novel fuzzy-based modified GAN and faster RCNN for classification of banana leaf disease | |
| Bashir et al. | A comprehensive review on apple leaf disease detection | |
| Parameshachari et al. | Plant Disease Detection and Classification Using Transfer Learning Inception Technique | |
| Milke et al. | Development of a coffee wilt disease identification model using deep learning | |
| Hall | An Automated Approach to Bee Identi cation from Wing Venation | |
| Park et al. | Clothing classification using CNN and shopping mall search system | |
| Chang et al. | Research on classification and recognition method of plant leaves based on deep learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190528 |