CN108763423A - A kind of jade recommendation method and device based on user picture - Google Patents
A kind of jade recommendation method and device based on user picture Download PDFInfo
- Publication number
- CN108763423A CN108763423A CN201810507261.0A CN201810507261A CN108763423A CN 108763423 A CN108763423 A CN 108763423A CN 201810507261 A CN201810507261 A CN 201810507261A CN 108763423 A CN108763423 A CN 108763423A
- Authority
- CN
- China
- Prior art keywords
- user
- picture
- identification
- test set
- jade
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/172—Classification, e.g. identification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/178—Human faces, e.g. facial parts, sketches or expressions estimating age from face image; using age information for improving recognition
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of, and the jade based on user picture recommends method and device, the method includes：Data are obtained, and the data of acquisition are normalized；Construct training set and test set；The second deep learning model is trained using the training set；Preset first deep learning model is evaluated using first test set；The the second deep learning model for obtain in evaluation procedure 3 using the second test set；Obtain pre-rendered jade product and user information relation table；User picture is obtained, the user picture is inputted to the second deep learning model after evaluation respectively；The user picture is inputted into the first deep learning model；In the table drawn in steps of 5 later, the jade product for being suitble to this user is searched for, then exports the information of corresponding with user information jade product, wherein the user information includes：The person characteristic information of user.Using the embodiment of the present invention, the accuracy rate of jade recommendation is improved.
Description
Technical field
The present invention relates to a kind of jades to recommend method and device, is more particularly to a kind of jade recommendation based on user picture
Method and device.
Background technology
Currently, China is jade consumption big country, possess the world's largest jade consumption market.It reforms and opens up to the outside world in the past 40 years,
Chinese jewels and jade industry achieves great development, and the engraving technology of jade shows unique characteristics, and jade product is dazzling, and category is numerous
It is more.With the continuous improvement of people's living standard, more and more people are keen to purchase jade product, and also have purchase
The economic strength bought.However ordinary consumer goes to carefully choose one by one there is no the too many time, and it is relevant specially not have jade yet
Industry knowledge, when buying jade, it is not known which kind of jade product is suitble to oneself.
Currently, many Products Show method and systems are all based on, personal information, the browsing that user fills in record, product closes
Data, the processes such as connection property are as follows：Existing Products Show technology depend on the personal information that user fills in, browsing record,
Then the data such as product relevance carry out jade recommendation according to these data and the matching degree of jade.
But the prior art is needed by the way that personal information is filled in manually, and the information that family is filled in registration is not necessarily
All effectively (such as gender, age), the information filled in often differs greatly with real information, to influence to recommend accuracy.Cause
This, the jade of recommendation and the truthful data of user have larger gap, and then cause jade recommendation results not accurate enough just
Technical problem.
Invention content
Technical problem to be solved by the present invention lies in provide a kind of jade recommendation method and dress based on user picture
It sets, it is not accurate enough with regard to technical problem to solve jade recommendation results in the prior art.
The present invention is to solve above-mentioned technical problem by the following technical programs：
An embodiment of the present invention provides a kind of, and the jade based on user picture recommends method, the method includes：
Step 1：Data are obtained, and the data of acquisition are normalized, wherein the data include：Containing someone
The picture of feature in object；
Step 2：Construct training set and test set, wherein the test set includes：First test set；With with the training
Collect corresponding second test set；
Step 3：The first deep learning model is obtained, and the second deep learning model is trained using the training set；
Step 4：Preset first deep learning model is evaluated using first test set；It is commented using the second test set
The the second deep learning model obtained in valence step 3；
Step 5：Obtain pre-rendered jade product and user information relation table；
Step 6：User picture is obtained, the user picture is inputted to the second deep learning model after evaluation respectively, and
Input the first deep learning model；The user picture is inputted into the first deep learning model；It is painted in steps of 5 later
In the table of system, it is suitble to according to the user information search of the first deep learning model and the second deep learning model identified
Then the jade product of this user exports the information of jade product corresponding with the user information, wherein the user information
Including：The person characteristic information of user.
Optionally, the character features include one or more combination of following characteristics：
The color of clothes, the color of lipstick, man, woman, woman's oiling lipstick, woman do not use lipstick, a variety of colors,
The each age group of personage, personage are with smiling or personage is without smile.
Optionally, the step 1, including：
It will be stored in different data sets after these picture classifications, according to different functional requirements, by required picture
Picture size normalized is carried out, i.e., picture is all converted into same size, and enhancing processing is carried out to image.
Optionally, the training set in the step 2 includes：The training set of age identification, the training set and clothing of lipstick identification
Take the training set of color identification；
First test set includes：Test set, the test set identified of smiling for gender identification；
Second test set includes：What the test set of age identification, the test set of lipstick identification and clothes color identified
Test set.
Optionally, the step 3, including：
Based on TensorFlow deep learning frames, age identification, lipstick identification is respectively trained in training convolutional neural networks
Corresponding deep learning model is identified with clothes color.
Optionally, the step 4, including：
Gender identification model is evaluated using the gender identification test set of gained in step 2；Use the test for identification of smiling
Collect to evaluate smile identification model；Test set is identified using clothes color to evaluate clothes color identification model；Known using lipstick
Other test set evaluates lipstick identification model；
Accuracy rate is more than the identification model of predetermined threshold value as the identification model after evaluation.
Optionally, the step 6, including：
The first step：The human face region in user picture is detected first with the human face detection tech of OpenCV；
Second step：Whether smiled using the user in the smile identification model judgement user picture of OpenCV；
Third walks：Using the gender identification model of OpenCV, user's gender in photo is identified, if user is female
Property, then the Mouth detection technical limit spacing face region of OpenCV is reused, face region is inputted into lipstick identification model, with judgement
Whether this female user has applied lipstick；
4th step：The clothes picture in user picture is obtained, this clothes picture is input in clothes color identification model,
Obtain the clothes color of this user；
5th step：User picture is inputted in age identification model, age of user section is obtained；
6th step：Get gender, the age, clothes color, smile, without smiling, applying lipstick and be not coated with lipstick after, then according to
According to the jade product and user information relation table obtained by step 5, the jade product of this user is suitble to according to user information search, so
The information of jade product corresponding with the user information is exported afterwards.
An embodiment of the present invention provides a kind of jade recommendation apparatus based on user picture, described device include：
First acquisition module is normalized, wherein the data for obtaining data, and by the data of acquisition
Including：Picture containing feature in personage；
Constructing module, for constructing training set and test set, wherein the test set includes：First test set；With with institute
State corresponding second test set of training set；
Training module trains the second deep learning mould for obtaining the first deep learning model, and using the training set
Type；
Evaluation module, for evaluating preset first deep learning model using first test set；It is surveyed using second
Examination collects to evaluate the second deep learning model obtained in the training module；
Second acquisition module, for obtaining pre-rendered jade product and user information relation table；
The user picture is inputted the second depth after evaluation by third acquisition module respectively for obtaining user picture
Learning model, and input the first deep learning model；The user picture is inputted into the first deep learning model；Exist later
In the table drawn in step 5, according to the user of the first deep learning model and the second deep learning model identified
Information search is suitble to the jade product of this user, then exports the information of jade product corresponding with the user information, wherein
The user information includes：The person characteristic information of user.
Optionally, the character features include one or more combination of following characteristics：
The color of clothes, the color of lipstick, man, woman, woman's oiling lipstick, woman do not use lipstick, a variety of colors,
The each age group of personage, personage are with smiling or personage is without smile.
First acquisition module, is additionally operable to：
It will be stored in different data sets after these picture classifications, according to different functional requirements, by required picture
Picture size normalized is carried out, i.e., picture is all converted into same size, and enhancing processing is carried out to image.
Optionally, the training set includes：The training set of age identification, the training set of lipstick identification and the identification of clothes color
Training set；
First test set includes：Test set, the test set identified of smiling for gender identification；
Second test set includes：What the test set of age identification, the test set of lipstick identification and clothes color identified
Test set.
Optionally, the training module, is additionally operable to：
Based on TensorFlow deep learning frames, age identification, lipstick identification is respectively trained in training convolutional neural networks
Corresponding deep learning model is identified with clothes color.
Optionally, the evaluation module, is additionally operable to：
Gender identification model is evaluated using the gender identification test set of gained in the constructing module；It is identified using the age
Test set carrys out appraisal age identification model；Test set is identified using clothes color to evaluate clothes color identification model；Use lip
Cream identifies test set to evaluate lipstick identification model；
Accuracy rate is more than the identification model of predetermined threshold value as the identification model after evaluation.
Optionally, the third acquisition module, is used for：
The first step：The human face region in user picture is detected first with the human face detection tech of OpenCV；
Second step：Whether smiled using the user in the smile identification model judgement user picture of OpenCV；
Third walks：Using the gender identification model of OpenCV, user's gender in photo is identified, if user is female
Property, then the Mouth detection technical limit spacing face region of OpenCV is reused, face region is inputted into lipstick identification model, with judgement
Whether this female user has applied lipstick；
4th step：The clothes picture in user picture is obtained, this clothes picture is input in clothes color identification model,
Obtain the clothes color of this user；
5th step：User picture is inputted in age identification model, age of user section is obtained；
6th step：Get gender, the age, clothes color, smile, without smiling, applying lipstick and be not coated with lipstick after, then according to
According to the jade product and user information relation table obtained by step 5, the jade product of this user is suitble to according to user information search, so
The information of jade product corresponding with the user information is exported afterwards.
The present invention has the following advantages compared with prior art：
(1) apply the embodiment of the present invention that can automatically extract the true master data of user, in turn by user picture
According to user data with the correspondence of jade product, recommend jade to user, compared with the existing technology in, filled out dependent on user
The authenticity of the data write, the data that the embodiment of the present invention obtains is more preferable, and then can improve the accuracy rate of jade recommendation.
(2) embodiment of the present invention is applied, the use difficulty and complexity of system can be reduced.
Description of the drawings
Fig. 1 is the flow diagram that a kind of jade based on user picture provided in an embodiment of the present invention recommends method；
Fig. 2 is a kind of structural schematic diagram of the jade recommendation apparatus based on user picture provided in an embodiment of the present invention；
Fig. 3 is a kind of structural schematic diagram of the jade commending system based on user picture provided in an embodiment of the present invention.
Specific implementation mode
It elaborates below to the embodiment of the present invention, the present embodiment is carried out lower based on the technical solution of the present invention
Implement, gives detailed embodiment and specific operating process, but protection scope of the present invention is not limited to following implementation
Example.
To solve prior art problem, the jade that an embodiment of the present invention provides a kind of based on user picture recommend method and
Device, just a kind of jade based on user picture provided in an embodiment of the present invention recommends method to be introduced first below.
Fig. 1 is the flow diagram that a kind of jade based on user picture provided in an embodiment of the present invention recommends method, such as
Shown in Fig. 1, the method includes：
S101：Data are obtained, and the data of acquisition are normalized, wherein the data include：Contain personage
The picture of middle feature.
Specifically, the character features include one or more combination of following characteristics：The color of clothes, lipstick
Color, man, woman, woman's oiling lipstick, woman do not use lipstick, a variety of colors, each age group of personage, personage is with smile
Or personage is without smile.
Specifically, by being stored in different data sets after these picture classifications, it, will be required according to different functional requirements
Picture carry out picture size normalized, i.e., picture is all converted into same size, and enhancing processing is carried out to image.
S102：Construct training set and test set, wherein the test set includes：First test set；With with the training set
Corresponding second test set.
Specifically, the training set includes：The training set of age identification, the training set of lipstick identification and the identification of clothes color
Training set；First test set includes：Test set, the test set identified of smiling for gender identification；Described second surveys
Examination collects：The test set of the test set of age identification, the test set and the identification of clothes color of lipstick identification.
Illustratively, it can gather the data identified for gender as one, know using the set as gender
Other data acquisition system；The data that can identify the age are gathered as one, using a part of data in the set as the age
The training set of identification, the test set that another part data in the set are identified as the age.It is understood that the age knows
Data part having the same in the test set that data in other training set can be identified with the age, can also completely not
Together.
In practical applications, each aforementioned training set can be subdivided into picture training set and label training set, it is aforementioned
Each test set is subdivided into picture test set and label test set, and then is stored separately.
For example, age recognition training collection may include：For the picture training set of age identification and for age identification
Then label training set will：It is stored respectively for the picture training set of age identification and the label training set for age identification
Into two binary files.For the training set of other functions, training set, the training identified of smiling for gender identification
Collection, the training set of age identification, the training set of lipstick identification and the training set of clothes color identification can also be according to the method described above
Storage.
For each in the test set of the test set of age identification, the test set of lipstick identification and the identification of clothes color
Each test set can be divided into picture test set and label test set according to the method described above, then stored respectively by test set
Into a binary file.
In a first aspect, the image data in picture training set can be all stored in a binary file, format
As shown in table 1.
Table 1 is a kind of storage format of picture training set for age identification provided in an embodiment of the present invention.
Table 1
| Offset | Type | Value | Description |
| 0000 | Int32 | 2029 | Magic number |
| 0004 | Int32 | 43877 | Picture sum |
| 0008 | Int32 | 512 | The line number of every width picture |
| 0012 | Int32 | 512 | The columns of every width picture |
| 0016 | Uint8 | Pixel | |
| 0017 | Uint8 | Pixel | |
| ...... | Uint8 | Pixel | |
| xxx | Uint8 | Pixel |
As shown in table 1, for each training set, in sequence by the pixel in each picture for including in the training set
Storage, wherein offset are in storage, and the signal of first storage byte position of 1 corresponding data of table is zero, by storage number
According to corresponding banner word section serial number；For example, the storage format of the corresponding picture training set for age identification of table 1, offset
Amount " 0000 " is the banner word section serial number for the picture training set magic number for storing age identification；" 0004 " is the identification of storage age
Picture training set in contain picture sum, 43877 banner word section serial number；" 0008 " is the picture for storing age identification
The line number of the pixel of the every pictures contained in training set, 512 banner word section serial number；" 0012 " is to identify at the storage age
The columns of the pixel of the every pictures contained in picture training set, 512 banner word section serial number；" 0016 " is to know at the storage age
The banner word section serial number of the pixel of the first pictures contained in other picture training set；" 0017 " is to identify at the storage age
The banner word section serial number of the pixel contained in picture training set, and so on.
In practical applications, the pixel of each pictures can store line by line, per a line by comprising pixel pixel
The value range of value composition, pixel is 0 to 255.For example, the pixel value of the first row pixel is arranged as from left to right：125；252；
0；47；The pixel value of second row pixel is arranged as from left to right：155；262；145；40.Then the first pictures, can in storage
Think：125；252；0；47；155；262；145；40.
It should be noted that magic number is used to characterize 1 corresponding identification function of table, for example, the picture training of age identification
The identification function integrated identifies as the age, it is assumed that the magic number of age identification is 2029, then table 1 is possessed magic number as " 2009 ".Type
For the type of data storage.In practical applications, magic number, as long as can be the identification work(that can distinguish each picture training set
Other numerical value of energy, the embodiment of the present invention do not make it restriction herein.
Label data corresponding with picture training set in label training set can be all stored in one two by second aspect
In binary file, format is as shown in table 2.
Table 2 is a kind of storage format of label training set for age identification provided in an embodiment of the present invention.
Table 2
As shown in table 2, in table 2, offset " 0000 " is the starting for the picture training set magic number for storing age identification
Byte position serial number；" 0004 " is storage for the total quantity 49877 containing label training in the label training set of age identification
Banner word section serial number；" 0008 " is label of the storage for age bracket one of in the label training set of age identification
Value, 0 banner word section serial number；" 0009 " is that another age bracket of storage in the label training set of age identification corresponds to
Label value, 1 banner word section serial number.
Image data in picture test set can be all stored in a binary file, format by the third aspect
As shown in table 1.The magic number of the corresponding file of picture test set for age identification is also 2029, and picture sum is 6000,
Picture line number columns is also all 512.
The label data identified for the age can be all stored in a binary file, format by fourth aspect
As shown in table 2.It is understood that the magic number of the corresponding file of label test set for age identification is also 2031, mark
Label sum is 6000；And label value range is 0 to 11, concrete meaning is as shown in table 3, and table 3 is provided in an embodiment of the present invention
A kind of mapping table between label value and age bracket for age identification.
Table 3
5th aspect, for lipstick identification, the magic number of the corresponding file of picture training set is 2039, and picture sum is
43877, picture line number columns is all 256；The magic number of the corresponding file of label training set is 2041, total number of labels 43877,
Label value range is that 0~1,0 expression does not apply lipstick, and 1 indicates to have applied lipstick.The magic number of the corresponding file of picture test set
It is 2039, picture sum is 6000, and picture line number columns is all 256；The magic number of the corresponding file of label test set is also
2041, total number of labels 6000, label value range is 0~1,0 expression no figure lipstick, and 1 indicates to have applied lipstick.
For the identification of clothes color, the magic number of the corresponding file of picture training set is 2049, and picture sum is
43877, picture line number columns is all 64；The magic number of the corresponding file of label training set is 2051, total number of labels 43877,
Label value range is 0~18, and the concrete meaning of each label is as shown in table 4.The magic number of the corresponding file of picture test set
It is also 2049, picture sum is 6000, and picture line number columns is all 64；The magic number of the corresponding file of label test set is also
2051, total number of labels 6000, label value range is 0~18, and the correspondence of label value and color is as shown in table 4.Table 4 is
The mapping table of label value and color for color identification.
Table 4
For gender identification, the magic number of the corresponding file of picture test set is 2009, and picture sum is 49877, figure
Piece line number and columns are all 512；The magic number of the corresponding file of label test set is 2011, and total number of labels 49877, label takes
Value ranging from 0 to 1,0 indicates male, and 1 indicates women；
For smiling and identifying, the magic number of the corresponding file of picture test set is 2019, and picture sum is 49877,
Picture line number and columns are all 512；The corresponding file magic number of label test set is 2021, and total number of labels 49877, label takes
Value ranging from 0 to 1,0:Expression is not smiled, and 1 indicates to smile；
S103：The first deep learning model is obtained, and the second deep learning model is trained using the training set.
Specifically, TensorFlow deep learning frames can be based on, age knowledge is respectively trained in training convolutional neural networks
Not, lipstick identification and clothes color identify corresponding second deep learning model.
TensorFlow deep learning frames are also known as tensor flow depth degree learning framework, be normally used for language identification,
The multinomial machine learning such as image recognition and computer vision and deep learning field.TensorFlow by a large amount of machine learning and
The relevant interface of deep learning is encapsulated, and perfect application programming interface is provided.That is, the first deep learning
Model can directly acquire.
S104：Preset first deep learning model is evaluated using first test set；It is commented using the second test set
The the second deep learning model obtained in valence S103 steps.
Specifically, gender identification model can be evaluated using the gender identification test set of gained in step 2；Use smile
The test set of identification evaluates smile identification model；Carry out appraisal age identification model using age identification test set；Use clothes
Color identifies test set to evaluate clothes color identification model；Test set is identified using lipstick to evaluate lipstick identification model；It will
Accuracy rate is more than the identification model of predetermined threshold value as the identification model after evaluation.
For example, the gender that the picture test set for gender identification of gained in step S102 is loaded into OpenCV identifies mould
In type, a recognition result set is obtained, this results set is compared with corresponding label test set, statistics identification is accurate
Rate is to evaluate gender identification model, and then the gender by the accuracy rate of the gender identification model in OpenCV not less than 98.2% is known
Other model is as the gender identification model for carrying out user's picture recognition.
By each width image data in the picture test set for identification of smiling of gained in S102 steps, by its first band
Enter the Face datection model of OpenCV (for example, model file can be opencv-2.4.13.4/data/h aarcascades/
Haarcascade_frontalface_default.xml in), face rectangular area coordinate in picture is obtained, then by this region
(model file is opencv-2.4.13.4/d ata/haarcascades/ to the smile identification model of coordinate loading OpenCV
Haarcascade_smile.xml), to identify whether the user in picture smiles.For entire picture test set, just
To a recognition result set, this results set is compared with corresponding label test set, statistics recognition accuracy is to comment
Then the accuracy rate of smile identification model in OpenCV is not less than 83.3% smile identification model by valence smile identification model
As the smile identification model for carrying out user's picture recognition.
By the picture test set for age identification of gained in S102 steps, it is loaded into age identification model, just obtains one
A recognition result set compares this results set with corresponding label test set, and statistics age recognition accuracy is to comment
Valence age identification model, not by the accuracy rate based on convolutional neural networks algorithm and the age identification model of TensorFlow frames
Age identification model less than 91.6% is as the age identification model after evaluation.
The method for obtaining the evaluation clothes color identification model after evaluation and the lipstick identification model after evaluation is commented with acquisition
The method of age identification model after valence is identical, and which is not described herein again.
S105：Obtain pre-rendered jade product and user information relation table；
In practical applications, it can be closed come pre-rendered jade product and user information according to the experience of jade industry specialists
It is table.
Table 5 is a kind of jade product provided in an embodiment of the present invention and user information relation table, as shown in table 5, jade production
Category type-user information relation table contains the correspondence of jade type and user information, wherein user information includes：Property
Not, the age, clothes color, personality, dress up.
Table 5
It should be noted that for character personality, whether we mainly from smiling this angle and considering, fill personage
Play the part of, whether we are mainly from the viewpoint of applying lipstick.In addition, there may be the meanings of multigroup value to be：For example certain type is beautiful
Stone product, masculinity femininity all matters are worn, then have two class values.
S106：User picture is obtained, the user picture input to the second deep learning model and defeated after evaluating respectively
Enter the first deep learning model；The user picture is inputted into the first deep learning model；It is drawn in steps of 5 later
Table in, this is suitble to according to the user information search of the first deep learning model and the second deep learning model identified
Then the jade product of user exports the information of jade product corresponding with the user information, wherein the user information packet
It includes：The person characteristic information of user.
Specifically, the first step：The human face region in user picture is detected first with the human face detection tech of OpenCV；
Second step：Whether smiled using the user in the smile identification model judgement user picture of OpenCV；
Third walks：Using the gender identification model of OpenCV, user's gender in photo is identified, if user is female
Property, then the Mouth detection technical limit spacing face region of OpenCV is reused, face region is inputted into lipstick identification model, with judgement
Whether this female user has applied lipstick；
4th step：The clothes picture in user picture is obtained, this clothes picture is input in clothes color identification model,
Obtain the clothes color of this user；
5th step：User picture is inputted in age identification model, age of user section is obtained；
6th step：Get gender, the age, clothes color, smile, without smiling, applying lipstick and be not coated with lipstick after, then according to
According to the jade product and user information relation table obtained by step 5, the jade product of this user is suitble to according to user information search, so
The information of jade product corresponding with the user information is exported afterwards.
It can obtain and believe with the user for each single item user information in user information according to the correspondence in table 5
Corresponding jade product type number is ceased, can be had multiple.For example, for the age information of user A, obtain and the age of user
The corresponding jade product type number -1 of information, jade product type number -2；For the smile information of user A, obtains and be somebody's turn to do
The corresponding jade product type of smile information of user numbers -3；And so on, then the jade product type of acquisition is numbered
Jade product as recommendation is exported.
In practical applications, it can also be numbered according to jade product type and obtain the detailed of jade product corresponding with the number
It counts accurately according to being exported, as shown in table 6, table 6 is that a kind of jade product type number provided in an embodiment of the present invention is obtained and is somebody's turn to do
Number the mapping table of corresponding jade product.
Table 6
In practical applications, OpenCV (Open Computer Vision, calculating formula of increasing income vision library) is one and is based on
BSD licenses are increased income the cross-platform computer vision library of distribution, and what it realized in terms of image procossing and computer vision many leads to
With the algorithm in terms of algorithm, especially recognition of face.
In practical applications, user information includes：Gender, clothes color, smile, lipstick and is not coated with without smiling, applying the age
Lipstick.
The true master data of user can be automatically extracted by user picture using embodiment illustrated in fig. 1 of the present invention,
And then according to user data with the correspondence of jade product, recommend jade to user, compared with the existing technology in, dependent on using
The authenticity of the data that family is filled in, the data that the embodiment of the present invention obtains is more preferable, and then can improve the accuracy rate of jade recommendation.
In addition, applying embodiment illustrated in fig. 1 of the present invention, the use difficulty and complexity of system can be reduced.
In addition, when carrying out jade Products Show to user in the prior art, the data of active user are relied only on, can not be it
Other people provide recommendation results, for example consumer A wants through certain commending system to be that consumer B buys certain product, at this time commending system
The result that provides is suitble to consumer A, but is not suitable for consumer B, because it is based on the data of consumer A and non-consumer
B's, using embodiment illustrated in fig. 1 of the present invention, after the user picture of consumer B is uploaded to system by consumer A, so that it may to offset
The person of expense B realizes the recommendation of jade product.
Corresponding with embodiment illustrated in fig. 1 of the present invention, the embodiment of the present invention additionally provides a kind of jade based on user picture
Stone recommendation apparatus.
Fig. 2 is a kind of jade recommendation apparatus based on user picture provided in an embodiment of the present invention, as shown in Fig. 2, described
Device includes：
First acquisition module 201 is normalized, wherein the number for obtaining data, and by the data of acquisition
According to including：Picture containing feature in personage；
Constructing module 202, for constructing training set and test set, wherein the test set includes：First test set；With
The second test set corresponding with the training set；
Training module 203 trains the second deep learning for obtaining the first deep learning model, and using the training set
Model；
Evaluation module 204, for evaluating preset first deep learning model using first test set；Use second
Test set evaluates the second deep learning model obtained in the training module；
Second acquisition module 205, for obtaining pre-rendered jade product and user information relation table；
The user picture is inputted the second depth after evaluation by third acquisition module 206 respectively for obtaining user picture
Spend learning model, and input the first deep learning model；The user picture is inputted into the first deep learning model；Later
In the table drawn in steps of 5, according to the use of the first deep learning model and the second deep learning model identified
Family information search is suitble to the jade product of this user, then exports the information of jade product corresponding with the user information,
In, the user information includes：The person characteristic information of user.
The true master data of user can be automatically extracted by user picture using embodiment illustrated in fig. 2 of the present invention,
And then according to user data with the correspondence of jade product, recommend jade to user, compared with the existing technology in, dependent on using
The authenticity of the data that family is filled in, the data that the embodiment of the present invention obtains is more preferable, and then can improve the accuracy rate of jade recommendation.
In addition, applying embodiment illustrated in fig. 1 of the present invention, the use difficulty and complexity of system can be reduced.
In a kind of specific implementation mode of the embodiment of the present invention, the character features include one of following characteristics or
Multiple combinations：
The color of clothes, the color of lipstick, man, woman, woman's oiling lipstick, woman do not use lipstick, a variety of colors,
The each age group of personage, personage are with smiling or personage is without smile.
In a kind of specific implementation mode of the embodiment of the present invention, first acquisition module 201 is additionally operable to：
It will be stored in different data sets after these picture classifications, according to different functional requirements, by required picture
Picture size normalized is carried out, i.e., picture is all converted into same size, and enhancing processing is carried out to image.
In a kind of specific implementation mode of the embodiment of the present invention, the training set includes：Training set, the lip of age identification
The training set of the training set and the identification of clothes color of cream identification；
First test set includes：Test set, the test set identified of smiling for gender identification；
Second test set includes：What the test set of age identification, the test set of lipstick identification and clothes color identified
Test set.
In a kind of specific implementation mode of the embodiment of the present invention, the training module 203 is additionally operable to：
Based on TensorFlow deep learning frames, age identification, lipstick identification is respectively trained in training convolutional neural networks
Corresponding deep learning model is identified with clothes color.
In a kind of specific implementation mode of the embodiment of the present invention, the evaluation module 204 is additionally operable to：
Gender identification model is evaluated using the gender identification test set of gained in the constructing module；It is identified using the age
Test set carrys out appraisal age identification model；Test set is identified using clothes color to evaluate clothes color identification model；Use lip
Cream identifies test set to evaluate lipstick identification model；
Accuracy rate is more than the identification model of predetermined threshold value as the identification model after evaluation.
In a kind of specific implementation mode of the embodiment of the present invention, the third acquisition module 206 is used for：
The first step：The human face region in user picture is detected first with the human face detection tech of OpenCV；
Second step：Whether smiled using the user in the smile identification model judgement user picture of OpenCV；
Third walks：Using the gender identification model of OpenCV, user's gender in photo is identified, if user is female
Property, then the Mouth detection technical limit spacing face region of OpenCV is reused, face region is inputted into lipstick identification model, with judgement
Whether this female user has applied lipstick；
4th step：The clothes picture in user picture is obtained, this clothes picture is input in clothes color identification model,
Obtain the clothes color of this user；
5th step：User picture is inputted in age identification model, age of user section is obtained；
6th step：Get gender, the age, clothes color, smile, without smiling, applying lipstick and be not coated with lipstick after, then according to
According to the jade product and user information relation table obtained by step 5, the jade product of this user is suitble to according to user information search, so
The information of jade product corresponding with the user information is exported afterwards.
Corresponding with embodiment illustrated in fig. 1 of the present invention, the embodiment of the present invention additionally provides a kind of jade based on user picture
Stone commending system.
Fig. 3 is a kind of structural schematic diagram of the jade commending system based on user picture provided in an embodiment of the present invention, such as
Shown in Fig. 3, system includes high in the clouds and client, wherein high in the clouds include data acquisition and with processing module, dataset construction module,
Identification model training module, model evaluation module, jade recommending module, wherein
Data acquisition and and processing module, the content for executing S101 steps；
Dataset construction module, for executing the content in S102 steps；
Identification model training module is used to execute the content in S103 steps；
Model evaluation module, for executing the content in S104 steps；
Jade recommending module, for executing the content in S105-S106 steps.
Client may include：One or more of the ends PC, WEB terminal, mobile phone terminal, tablet computer end etc..
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all essences in the present invention
All any modification, equivalent and improvement etc., should all be included in the protection scope of the present invention made by within refreshing and principle.
Claims (14)
1. a kind of jade based on user picture recommends method, which is characterized in that the method includes：
Step 1：Data are obtained, and the data of acquisition are normalized, wherein the data include：Containing in personage
The picture of feature；
Step 2：Construct training set and test set, wherein the test set includes：First test set；With with the training set pair
The second test set answered；
Step 3：The first deep learning model is obtained, and the second deep learning model is trained using the training set；
Step 4：Preset first deep learning model is evaluated using first test set；Step is evaluated using the second test set
The the second deep learning model obtained in rapid 3；
Step 5：Obtain pre-rendered jade product and user information relation table；
Step 6：User picture is obtained, the user picture is inputted to the second deep learning model after evaluation, and input respectively
The first deep learning model；The user picture is inputted into the first deep learning model；It is drawn in steps of 5 later
In table, this use is suitble to according to the user information search of the first deep learning model and the second deep learning model identified
Then the jade product at family exports the information of jade product corresponding with the user information, wherein the user information packet
It includes：The person characteristic information of user.
2. a kind of jade based on user picture according to claim 1 recommends method, which is characterized in that the personage is special
Sign includes one or more combination of following characteristics：
The color of clothes, the color of lipstick, man, woman, woman's oiling lipstick, woman do not use lipstick, a variety of colors, personage
Each age group, personage with smile or personage without smile.
3. a kind of jade based on user picture according to claim 1 recommends method, which is characterized in that the step 1,
Including：
It will be stored in different data sets after these picture classifications, according to different functional requirements, required picture carried out
Picture is all converted to same size, and carries out enhancing processing to image by picture size normalized.
4. a kind of jade based on user picture according to claim 1 recommends method, which is characterized in that the step 2
In training set include：The training set of the training set of age identification, the training set and the identification of clothes color of lipstick identification；
First test set includes：Test set, the test set identified of smiling for gender identification；
Second test set includes：The test of the test set of age identification, the test set and the identification of clothes color of lipstick identification
Collection.
5. a kind of jade based on user picture according to claim 4 recommends method, which is characterized in that the step 3,
Including：
Based on TensorFlow deep learning frames, age identification, lipstick identification and clothing is respectively trained in training convolutional neural networks
It takes color and identifies corresponding deep learning model.
6. a kind of jade based on user picture according to claim 4 recommends method, which is characterized in that the step 4,
Including：
Gender identification model is evaluated using the gender identification test set of gained in step 2；Test set using identification of smiling comes
Evaluate smile identification model；Carry out appraisal age identification model using age identification test set；Test set is identified using clothes color
To evaluate clothes color identification model；Test set is identified using lipstick to evaluate lipstick identification model；
Accuracy rate is more than the identification model of predetermined threshold value as the identification model after evaluation.
7. a kind of jade based on user picture according to claim 1 recommends method, which is characterized in that the step 6,
Including：
The first step：The human face region in user picture is detected first with the human face detection tech of OpenCV；
Second step：Whether smiled using the user in the smile identification model judgement user picture of OpenCV；
Third walks：Using the gender identification model of OpenCV, user's gender in photo is identified, if user is women,
Face region is inputted lipstick identification model, to judge this female by the Mouth detection technical limit spacing face region for reusing OpenCV
Whether property user has applied lipstick；
4th step：The clothes picture in user picture is obtained, this clothes picture is input in clothes color identification model, is obtained
The clothes color of this user；
5th step：User picture is inputted in age identification model, age of user section is obtained；
6th step：Get gender, the age, clothes color, smile, without smiling, applying lipstick and be not coated with lipstick after, then according to step
The jade product and user information relation table of rapid 5 gained, the jade product of this user is suitble to according to user information search, then defeated
Go out the information of jade product corresponding with the user information.
8. a kind of jade recommendation apparatus based on user picture, which is characterized in that described device includes：
First acquisition module is normalized for obtaining data, and by the data of acquisition, wherein the data include：
Picture containing feature in personage；
Constructing module, for constructing training set and test set, wherein the test set includes：First test set；With with the instruction
Practice and collects corresponding second test set；
Training module trains the second deep learning model for obtaining the first deep learning model, and using the training set；
Evaluation module, for evaluating preset first deep learning model using first test set；Use the second test set
To evaluate the second deep learning model obtained in the training module；
Second acquisition module, for obtaining pre-rendered jade product and user information relation table；
The user picture is inputted the second deep learning after evaluation by third acquisition module respectively for obtaining user picture
Model, and input the first deep learning model；The user picture is inputted into the first deep learning model；Later in step
In the table drawn in 5, according to the user information of the first deep learning model and the second deep learning model identified
Search is suitble to the jade product of this user, then exports the information of jade product corresponding with the user information, wherein described
User information includes：The person characteristic information of user.
9. a kind of jade recommendation apparatus based on user picture according to claim 8, which is characterized in that the personage is special
Sign includes one or more combination of following characteristics：
The color of clothes, the color of lipstick, man, woman, woman's oiling lipstick, woman do not use lipstick, a variety of colors, personage
Each age group, personage with smile or personage without smile.
10. a kind of jade recommendation apparatus based on user picture according to claim 8, which is characterized in that described first
Acquisition module is additionally operable to：
It will be stored in different data sets after these picture classifications, according to different functional requirements, required picture carried out
Picture is all converted to same size, and carries out enhancing processing to image by picture size normalized.
11. a kind of jade recommendation apparatus based on user picture according to claim 8, which is characterized in that the training
Collection includes：The training set of the training set of age identification, the training set and the identification of clothes color of lipstick identification；
First test set includes：Test set, the test set identified of smiling for gender identification；
Second test set includes：The test of the test set of age identification, the test set and the identification of clothes color of lipstick identification
Collection.
12. a kind of jade recommendation apparatus based on user picture according to claim 11, which is characterized in that the training
Module is additionally operable to：
Based on TensorFlow deep learning frames, age identification, lipstick identification and clothing is respectively trained in training convolutional neural networks
It takes color and identifies corresponding deep learning model.
13. a kind of jade recommendation apparatus based on user picture according to claim 11, which is characterized in that the evaluation
Module is additionally operable to：
Gender identification model is evaluated using the gender identification test set of gained in the constructing module；It is identified and is tested using the age
Collection carrys out appraisal age identification model；Test set is identified using clothes color to evaluate clothes color identification model；Known using lipstick
Other test set evaluates lipstick identification model；
Accuracy rate is more than the identification model of predetermined threshold value as the identification model after evaluation.
14. a kind of jade recommendation apparatus based on user picture according to claim 1, which is characterized in that the third
Acquisition module is used for：
The first step：The human face region in user picture is detected first with the human face detection tech of OpenCV；
Second step：Whether smiled using the user in the smile identification model judgement user picture of OpenCV；
Third walks：Using the gender identification model of OpenCV, user's gender in photo is identified, if user is women,
Face region is inputted lipstick identification model, to judge this female by the Mouth detection technical limit spacing face region for reusing OpenCV
Whether property user has applied lipstick；
4th step：The clothes picture in user picture is obtained, this clothes picture is input in clothes color identification model, is obtained
The clothes color of this user；
5th step：User picture is inputted in age identification model, age of user section is obtained；
6th step：Get gender, the age, clothes color, smile, without smiling, applying lipstick and be not coated with lipstick after, then according to step
The jade product and user information relation table of rapid 5 gained, the jade product of this user is suitble to according to user information search, then defeated
Go out the information of jade product corresponding with the user information.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810507261.0A CN108763423A (en) | 2018-05-24 | 2018-05-24 | A kind of jade recommendation method and device based on user picture |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810507261.0A CN108763423A (en) | 2018-05-24 | 2018-05-24 | A kind of jade recommendation method and device based on user picture |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108763423A true CN108763423A (en) | 2018-11-06 |
Family
ID=64005795
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810507261.0A Pending CN108763423A (en) | 2018-05-24 | 2018-05-24 | A kind of jade recommendation method and device based on user picture |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108763423A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110348514A (en) * | 2019-07-10 | 2019-10-18 | 合肥晌玥科技有限公司 | A kind of jade recognition methods and device based on Faster R-CNN model |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103098079A (en) * | 2011-04-11 | 2013-05-08 | 英特尔公司 | Personalized program selection system and method |
| US20160027046A1 (en) * | 2014-07-24 | 2016-01-28 | Samsung Electronics Co., Ltd. | Method and device for playing advertisements based on relationship information between viewers |
| CN105426850A (en) * | 2015-11-23 | 2016-03-23 | 深圳市商汤科技有限公司 | Human face identification based related information pushing device and method |
-
2018
- 2018-05-24 CN CN201810507261.0A patent/CN108763423A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103098079A (en) * | 2011-04-11 | 2013-05-08 | 英特尔公司 | Personalized program selection system and method |
| US20160027046A1 (en) * | 2014-07-24 | 2016-01-28 | Samsung Electronics Co., Ltd. | Method and device for playing advertisements based on relationship information between viewers |
| CN105426850A (en) * | 2015-11-23 | 2016-03-23 | 深圳市商汤科技有限公司 | Human face identification based related information pushing device and method |
Non-Patent Citations (2)
| Title |
|---|
| WZEARW: ""见人识面，TensorFlow实现人脸性别/年龄识别"", 《HTTPS://CLOUD.TENCENT.COM/DEVELOPER/ARTICLE/1092415》 * |
| 风吴痕: ""OpenCV检测篇（二） ——笑脸检测"", 《HTTPS://BLOG.CSDN.NET/WC781708249/ARTICLE/DETAILS/78582430》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110348514A (en) * | 2019-07-10 | 2019-10-18 | 合肥晌玥科技有限公司 | A kind of jade recognition methods and device based on Faster R-CNN model |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104203042B (en) | Makeup auxiliary device, cosmetic auxiliary method and recording medium | |
| CN107341434A (en) | Processing method, device and the terminal device of video image | |
| CN107844990A (en) | A kind of approaches to IM and its system, terminal device for intelligent shops | |
| Wang et al. | Analyses of a multimodal spontaneous facial expression database | |
| CN105426850A (en) | Human face identification based related information pushing device and method | |
| EP3485762A1 (en) | Makeup assistance device and makeup assistance method | |
| US11010894B1 (en) | Deriving a skin profile from an image | |
| CN106055710A (en) | Video-based commodity recommendation method and device | |
| CN107392151A (en) | Face image various dimensions emotion judgement system and method based on neutral net | |
| CN109034099A (en) | A kind of expression recognition method and device | |
| KR102292629B1 (en) | System and method for providing beauty contents using virtual simulation contents | |
| CN106951448A (en) | A kind of personalization, which is worn, takes recommendation method and system | |
| CN108876430B (en) | Advertisement pushing method based on crowd characteristics, electronic equipment and storage medium | |
| CN108376352A (en) | The cosmetics of color with suitable user provide device and method | |
| CN107169002A (en) | A kind of personalized interface method for pushing and device recognized based on face | |
| CN106327231A (en) | Personalized commodity matching and recommending system and method | |
| Wu et al. | Understanding and modeling user-perceived brand personality from mobile application uis | |
| CN111428662A (en) | Advertisement playing change method and system based on crowd attributes | |
| CN112819718A (en) | Image processing method and device, electronic device and storage medium | |
| Yan et al. | Measuring dynamic micro-expressions via feature extraction methods | |
| Park et al. | An automatic virtual makeup scheme based on personal color analysis | |
| CN110097400A (en) | Information recommendation method, apparatus and system, storage medium, intelligent interaction device | |
| CN112889065A (en) | System and method for providing personalized product recommendations using deep learning | |
| CN110807691B (en) | Cross-commodity-class commodity recommendation method and device | |
| CN108763423A (en) | A kind of jade recommendation method and device based on user picture |
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 | ||
| TA01 | Transfer of patent application right | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20200910 Address after: 230000 2nd floor, No.4 workshop, intelligent equipment science and Technology Park, no.3963 Susong Road, Hefei Economic and Technological Development Zone, Anhui Province Applicant after: Hefei Huaiyue Technology Co.,Ltd. Address before: 236000 Anhui city of Hefei Province Economic and Technological Development Zone Cuiwei Road No. 6 Haiheng building room 6012 Applicant before: HRG INTERNATIONAL INSTITUTE FOR RESEARCH & INNOVATION |
|
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20181106 |