CN104850825B - A kind of facial image face value calculating method based on convolutional neural networks - Google Patents
A kind of facial image face value calculating method based on convolutional neural networks Download PDFInfo
- Publication number
- CN104850825B CN104850825B CN201510185613.1A CN201510185613A CN104850825B CN 104850825 B CN104850825 B CN 104850825B CN 201510185613 A CN201510185613 A CN 201510185613A CN 104850825 B CN104850825 B CN 104850825B
- Authority
- CN
- China
- Prior art keywords
- face
- facial image
- face value
- convolutional neural
- neural networks
- 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.)
- Active
Links
Abstract
The invention discloses a kind of facial image face value calculating method based on convolutional neural networks.Gather facial image, including without face value label and with face value label；Pretreatment early period is carried out, the key point of face is obtained and extracts global and local facial image block；Then pre-training convolutional neural networks are finely adjusted, extract the depth characteristic of face, extract shape facility, both are combined as face value tag；Face value tag is input to trained face value grader in grader；Tested facial image is carried out to above-mentioned steps again successively and obtains its face value tag, its face value is calculated to obtain to the face value tag for being tested facial image with face value grader.The present invention utilizes convolutional neural networks to extract the depth characteristic of global face and local face image, and combines face shape feature, and the face value overcome in complex situations calculates uncertainty, and robustness is high, has good result in engineer application.
Description
Technical field
A kind of one kind the present invention relates to image processing method, more particularly to Computer Vision Recognition technical field is based on
The facial image face value calculating method of convolutional neural networks.
Background technology
Face value represents the handsome or beautiful numerical value of personage Yan Rong, for evaluating personage's appearance." to the right " social software is initiated
Face value ranking list, ' noninductive ' and ' liking ' quantity received according to user calculate the user's face value height, such as：At one
There are 70 ' liking ' in received 100 feedbacks from other users of user, then the face value of the user is 0.7 (0.0-1.0).
Face value is higher to represent that appearance is more good-looking, lower to represent more plain.The calculating of face value, be primarily referred to as face's face ratio whether
Coordinate, and the penetrating concept that when ancient Chinese artist draws a portrait sums up " three five, the front yards " come defines facial standard ratio
Example relation, is commonly referred to as " golden section " of face in the world --- and 1:0.618.The face value to facial image is all to pass through people at present
Judge obtains, and one automation face value calculating method of design, which seems, to be even more important, such as can mistake in photo-editing software
Filter the unhandsome photo of same person, select appearance beautiful in image search engine or handsome photo etc..
In recent years, face value calculating method was divided into the method based on shape facility and the method based on shallow-layer feature.Based on shape
The face value calculating method of shape feature refers to that the ratio between human face five-sense-organ calculates feature, " three front yards as described above as face value
Five ".Face value calculating method based on shallow-layer feature refers to that calculating the shallow-layer features such as LBP, GIST or HOG of face is used as
The feature of face value.
The difficulty that face face value calculates is embodied in following aspects：
1st, the imaging angle of camera, i.e. posture all have an impact most face value calculating method, are based particularly on shape
The method of feature.Two facial images for belonging to same person are likely to result in the face value of this two images due to the influence of posture
As a result it is different.
2nd, the change of illumination can change the half-tone information of facial image, therefore face value can be calculated and can had an impact.
3rd, facial image may also be subjected to the age, block etc. therefore influencing, and can influence what face value calculated to varying degrees
Accuracy.
Convolutional neural networks are one kind of artificial neural network, it has also become current speech analysis is ground with field of image recognition
Study carefully hot spot.Its weights share network structure and are allowed to be more closely similar to biological neural network, reduce the complexity of network model, subtract
The quantity of weights is lacked.What the advantage was showed when the input of network is multidimensional image becomes apparent, and image is directly made
For the input of network, feature extraction complicated in traditional recognition method and data reconstruction processes are avoided.Convolutional network is to know
One multilayer perceptron of other two-dimensional shapes and special designing, this network structure is to translation, proportional zoom, inclination or is total to him
The deformation of form has height consistency, and the training method of convolutional neural networks uses BP methods.
The content of the invention
In order to solve the problems, such as present in background technology, it is an object of the invention to provide one kind to be based on convolutional neural networks
Facial image face value calculating method, it is possible to increase facial image face value calculate accuracy and robustness.
The technical solution adopted by the present invention includes the following steps, as shown in Figure 1：
1) facial image of different people is gathered, the different facial images and without face value label are gathered for everyone
Different facial images with face value label are as image pattern；Face value label is face value grade mark, with face value label
The face value grade of facial image can be divided into ten grades；
2) pretreatment early period is carried out to face images, then obtains the key point of face and extract every facial image
Global Face image block and local facial image block；
3) for the facial image without face value label, for each Global Face image block and local facial image
Block, pre-training convolutional neural networks；
4) since the facial image sample with face value label is fewer, it is necessary to which pre-training has the volume of face identification functions
Product neutral net；For the facial image with face value label, above-mentioned steps 3 are finely tuned) the obtained convolutional neural networks of pre-training；
5) the full articulamentum output of the convolutional neural networks structure end after extraction fine setting is special as the depth of facial image
Sign, extracts the shape facility of facial image, combined depth feature and shape facility are as face value tag；
6) the face value tag that step 5) obtains is input in support vector machine classifier (SVM) and be trained, obtain face
It is worth grader；
7) tested facial image is subjected to above-mentioned steps 2 again successively) and its face value tag 5) is obtained, after step 6) training
Face value grader to be tested facial image face value tag carry out that its face value is calculated.
Pretreatment early period of the step 2) facial image refers to carry out dimension normalization to face images so that institute
There is the image resolution ratio of facial image identical.
The step 2) obtains the key point of face and extracts the Global Face image block of every facial image and local people
The detailed process of face image block is：Detect that five face key points are put using active shape model (ASM), five faces close
Key point is left eye central point, right eye central point, nose central point, corners of the mouth left hand edge point and corners of the mouth right hand edge point, further according to five
Face key point position extracts at least one Global Face image block and five local facial image blocks as pre-training convolution
The input of neutral net.
The Global Face image block refers to refer to comprising eyes, nose, the image block of face, local facial image block
By five face key points each centered on square image blocks, the length of side of local facial image block is left in facial image
2 times of the distance between eye, right eye center.
As shown in the left side of fig 2, step 3) the pre-training convolutional neural networks are specially：The convolutional Neural net of pre-training
The Global Face image block and the total quantity of local facial image block that the sum of network obtains for the middle extraction of step 2), each image block
A corresponding convolutional neural networks, the structure of all convolutional neural networks are identical；The pond layer of convolutional neural networks is using most
Great Chiization layer, loss layer are softmax layers, number when number of nodes gathers for image pattern, the company in convolutional neural networks
Connect weights to be initialized by normal Gaussian function, bias is initialized as zero.
As illustrated at the right side of figure 2, the convolutional neural networks that the step 4) fine setting pre-training obtains are specially：With step 3)
The connection weight of convolutional neural networks is replaced as the connection weight initial value after fine setting, and with new softmax layers after pre-training
The softmax layers of convolutional neural networks after step 3) pre-training；New softmax layer of number of nodes and all face value labels that carry
Face value grade quantity in facial image, connection weight softmax layers new are initialized by normal Gaussian function, softmax layers new
Bias be initialized as zero.
The step 5) is specially：For each facial image for carrying face value, all Global Face images are extracted
Block and local facial image block are input in the convolutional neural networks after corresponding fine setting, then each convolutional neural networks end
Full articulamentum output be together in series form depth characteristic, extraction with face value label facial image shape facility, and
With depth characteristic last face value tag in series, then using principal component analytical method (PCA methods) to face value tag carry out
Dimensionality reduction.
The shape facility is the coordinate of multiple face key points, and multiple face key points include at least left eye center
Point, right eye central point, nose central point, corners of the mouth left hand edge point and corners of the mouth right hand edge point,
The kernel function of the support vector machine classifier of the step 6) uses Radial basis kernel function, and trained sample data is
Face value tag vector after step 5) dimensionality reduction.
Compared with prior art, the beneficial effects of the present invention are：
In existing face face value calculating method, not yet there is the face value based on convolutional neural networks and calculate correlation technique,
Present invention incorporates traditional face shape feature and brand-new depth characteristic, be to one of current face value calculating method comprehensively
Upgrading.
Convolutional neural networks of the present invention have translation, proportional zoom, inclination or the deformation of his common form highly constant
Property, using factors such as the effective change that must overcome human face posture of depth characteristic, the changes of light；Face global image block drawn game
The combination of portion's image block depth characteristic forms higher-dimension redundancy feature, high to the face face value discriminant classification under complex background, Shandong
Rod is good；The stronger low-dimensional feature vector of ability to express is obtained to high dimensional feature dimensionality reduction using PCA, improves face value calculating speed.
Brief description of the drawings
Fig. 1 is the flow diagram of the present invention.
Fig. 2 is the pre-training and fine setting convolutional neural networks structure chart of the embodiment of the present invention.
Fig. 3 is five key point position views of face of the embodiment of the present invention.
Fig. 4 is the local facial image block schematic diagram of the embodiment of the present invention.
Fig. 5 is the Global Face image block schematic diagram of the embodiment of the present invention.
Embodiment
Hereinafter reference will be made to the drawings, and the preferred embodiment of the present invention is described in detail.
As shown in Figure 1, the embodiment of the method for the present invention specifically includes following steps：
1) prepare the facial image of different people and carry out necessary pretreatment early period, obtain preferable facial image；Specifically
, due to using this deep learning model of convolutional neural networks, complicated pretreatment need not be carried out to image, is only needed
Dimension normalization is carried out, normalized image to 100 × 120 resolution ratio, picture format is RGB color image.
2) the Global Face image block of every facial image and local facial image block are extracted according to the key point of face；Tool
Body, first by critical point detection method locating human face's key point position, the positions of 5 key points as shown in figure 3, further according to
Key point position extraction Global Face image block and local image block, Global Face image block and local facial image block have
5 kinds.
As shown in figure 5, wherein Global Face image block contains the information of face entirety, as shown in figure 4, local facial figure
As block is extracted centered on key point.The image block of extraction will normalize to 31 × 39, as convolutional neural networks
Input.
3) every kind of image block, pre-training convolutional neural networks are directed to；Specifically, a total of 10 kinds of facial image blocks, it is necessary to
10 convolutional neural networks of pre-training, the structure of 10 convolutional neural networks is identical, structure such as Fig. 2 institutes of convolutional neural networks
Show, left side is illustrated for the convolutional neural networks of pre-training, and right side is illustrated for the convolutional neural networks of fine setting, wherein the ginseng of convolutional layer
Number expression-form：4 × 4 × 20+1 (step-length), the size for representing convolution kernel are 4 × 4, quantity 20, step-length 1；Setting is local
It is 3 that corresponding normalization layer, which wants normalized adjacent convolution layer number,；Pond layer is using maximum pond mode, 2 × 2+2 (step-length)
It is 2 × 2 to represent pond core size, step-length 2；Network includes a full articulamentum, shares 160 node units, full articulamentum
It is connected with loss layer, loss layer uses Softmax loss functions.Face sample size is 500,000, share 10,000 it is different
People, everyone face picture has 50, therefore softmax layers of number of nodes is 10,000.
4) prepare the facial image with face value label, finely tune convolutional neural networks；Specifically, the people with face value label
Face image has 20,000, and every all corresponds to a face value, and face value is divided into 10 grades, and minimum face is represented with 0 to 9 digital representations, 0
Value, appearance is ugly, and 9 represent highest face value, have a handsome countenance beautiful.Using Deep self-taught learning for
Method in facial beauty prediction (Junying Gan, 2014) carries out face value grade classification, obtains respective
Face value label.Every in 20000 face pictures is scored by 10 people respectively, and rounding represents last to the average of 10 scorings again
Face value grade.Fine setting convolutional neural networks only need to pre-training it is good 3) in softmax layers replace with it is new with 10
The softmax layers of node, new softmax layers and the weights of full articulamentum are initialized by Gaussian function again, and are above owned
The initialization weight setting of layer is the weights of the good convolutional neural networks of pre-training, therefore except softmax node layers change
Outside, the convolutional neural networks structure of fine setting is consistent with pre-training convolutional neural networks, as shown in Figure 2.
5) depth characteristic of the output of the full articulamentum of convolutional neural networks structure end as facial image, extraction are extracted
The shape facility of facial image, the feature that two feature vectors of combination are calculated as face value；Specifically, the convolution after extraction fine setting
The output valve of the full articulamentum of neutral net is formed as depth characteristic vector, the depth characteristic vector for 10 160 dimensions of connecting
1600 dimension depth characteristic vectors, extract shape facility, shape facility has 68 dimensions, represents the position of key point.Finally connect depth
Feature and shape facility form 1668 dimensional feature vectors as face value tag.Since 1668 dimensions are the high dimensional features of redundancy, directly
Calculating can influence speed, therefore face value tag dimensionality reduction to 150 dimensional feature vectors calculates face face value using PCA with PCA
It is as shown in table 1 to lift effect.
6) it is used as face value grader by the use of the feature vector Training Support Vector Machines grader (SVM) in 5) step；Specifically
, sample size is 20,000, and each sample is 1668 dimensional feature vectors, and label is face value grade.Select radial direction base SVM conducts
Grader, 10 decile of sample size, optimal classifier parameters are determined using cross-validation method.
Influence of 1 principal component analysis of table to calculating speed and nicety of grading
| Face value grade separation precision | Every image calculating speed | |
| Without using principal component analysis | 93.57% | 45ms |
| Using principal component analysis | 95.83% | 30ms |
The present invention combines method accuracy with higher of the method compared to conventional shape feature of depth characteristic, robustness
It is good, it is as shown in table 2 below：
Influence of 2 depth characteristic of table to face value grade separation precision
| Feature | Face value grade separation precision |
| Shape facility | 87.52% |
| Shape facility+depth characteristic | 95.83% |
It can be seen from the above that the present invention makes use of the depth characteristic that convolutional neural networks extract global face and local face image, and
With reference to face shape feature, the combination of two kinds of features had both remained traditional face aesthetic approach on human face five-sense-organ ratio,
The advantages of convolutional neural networks can learn deep layer face semantic information is absorbed again, and the face value overcome in complex situations calculates
The combination of uncertainty, depth characteristic and shape facility can improve the precision of face value grade separation, and robustness is good, have prominent aobvious
The technique effect of work.
Above-mentioned embodiment is used for illustrating the present invention, rather than limits the invention, the present invention's
In spirit and scope of the claims, to any modifications and changes of the invention made, protection model of the invention is both fallen within
Enclose.
Claims (8)
1. a kind of facial image face value calculating method based on convolutional neural networks, it is characterised in that comprise the following steps：
1）Gather different people facial image, for everyone gather without face value label different facial images and carried
For the different facial images of face value label as image pattern, face value label is face value grade mark；
2）Pretreatment early period is carried out to face images, the key point of face is then obtained and extracts the complete of every facial image
Office's facial image block and local facial image block；
3）For the facial image without face value label, for each Global Face image block and local facial image block, in advance
Training convolutional neural networks；
4）For the facial image with face value label, above-mentioned steps 3 are finely tuned）The convolutional neural networks that pre-training obtains；
5）The full articulamentum of convolutional neural networks structure end after extraction fine setting exports the depth characteristic as facial image, carries
The shape facility of facial image is taken, combined depth feature and shape facility are as face value tag；
The step 5）Specially：For each facial image for carrying face value label, all Global Face images are extracted
Block and local facial image block are input in the convolutional neural networks after corresponding fine setting, then each convolutional neural networks end
Full articulamentum output be together in series form depth characteristic, extraction with face value label facial image shape facility, and
With depth characteristic last face value tag in series, then using principal component analytical method to face value tag carry out dimensionality reduction；
6）By step 5）The face value tag obtained after dimensionality reduction, which is input in support vector machine classifier, to be trained, and obtains face value point
Class device；
7）Tested facial image is subjected to above-mentioned steps 2 again successively）With 5）Its face value tag is obtained, with step 6）Face after training
Value grader to the face value tag for being tested facial image carries out that its face value is calculated.
2. a kind of facial image face value calculating method based on convolutional neural networks according to claim 1, its feature exist
In：The step 2）Pretreatment early period of facial image refers to carry out dimension normalization to face images so that owner
The image resolution ratio of face image is identical.
3. a kind of facial image face value calculating method based on convolutional neural networks according to claim 1, its feature exist
In：The step 2）Obtain the key point of face and extract the Global Face image block of every facial image and local facial image
The detailed process of block is：Detect that five face key points are put using active shape model, five face key points are left eye
Central point, right eye central point, nose central point, corners of the mouth left hand edge point and corners of the mouth right hand edge point, further according to five face key points
Position extracts at least one Global Face image block and five local facial image blocks as pre-training convolutional neural networks
Input.
4. a kind of facial image face value calculating method based on convolutional neural networks according to claim 3, its feature exist
In：The Global Face image block refers to refer to five comprising eyes, nose, the image block of face, local facial image block
A face key point each centered on square image blocks, the length of side of local facial image block is left eye, the right side in facial image
2 times of the distance between eye center.
5. a kind of facial image face value calculating method based on convolutional neural networks according to claim 1, its feature exist
In：The step 3）Pre-training convolutional neural networks are specially：The sum of the convolutional neural networks of pre-training is step 2）In carry
The Global Face image block and the total quantity of local facial image block obtained, each image block correspond to a convolutional Neural net
Network, the structure of all convolutional neural networks are identical；The pond layer of convolutional neural networks uses maximum pond layer, and loss layer is
Softmax layers, number when number of nodes gathers for image pattern, the connection weight in convolutional neural networks is by normal Gaussian letter
Number initialization, bias are initialized as zero.
6. a kind of facial image face value calculating method based on convolutional neural networks according to claim 1, its feature exist
In：The step 4）Finely tuning the convolutional neural networks that pre-training obtains is specially：With step 3）Convolutional neural networks after pre-training
Connection weight as the connection weight initial value after fine setting, and replace step 3 with new softmax layers）Convolution god after pre-training
Softmax layers through network；Number of nodes softmax layers new and face value number of degrees in all facial images with face value label
Measure identical, softmax layers new connection weight to be initialized by normal Gaussian function, bias softmax layers new is initialized as
Zero.
7. a kind of facial image face value calculating method based on convolutional neural networks according to claim 1, its feature exist
In：The shape facility is the coordinate of multiple face key points, and multiple face key points include at least left eye central point, right eye
Central point, nose central point, corners of the mouth left hand edge point and corners of the mouth right hand edge point.
8. a kind of facial image face value calculating method based on convolutional neural networks according to claim 1, its feature exist
In：The step 6）The kernel function of support vector machine classifier use Radial basis kernel function.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510185613.1A CN104850825B (en) | 2015-04-18 | 2015-04-18 | A kind of facial image face value calculating method based on convolutional neural networks |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510185613.1A CN104850825B (en) | 2015-04-18 | 2015-04-18 | A kind of facial image face value calculating method based on convolutional neural networks |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104850825A CN104850825A (en) | 2015-08-19 |
| CN104850825B true CN104850825B (en) | 2018-04-27 |
Family
ID=53850460
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510185613.1A Active CN104850825B (en) | 2015-04-18 | 2015-04-18 | A kind of facial image face value calculating method based on convolutional neural networks |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104850825B (en) |
Families Citing this family (58)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9542626B2 (en) * | 2013-09-06 | 2017-01-10 | Toyota Jidosha Kabushiki Kaisha | Augmenting layer-based object detection with deep convolutional neural networks |
| CN105117611B (en) * | 2015-09-23 | 2018-06-12 | 北京科技大学 | Based on the determining method and system of the TCM tongue diagnosis model of convolutional Neural metanetwork |
| CN105205479A (en) * | 2015-10-28 | 2015-12-30 | 小米科技有限责任公司 | Human face value evaluation method, device and terminal device |
| CN106709411A (en) * | 2015-11-17 | 2017-05-24 | 腾讯科技（深圳）有限公司 | Appearance level acquisition method and device |
| CN105354560A (en) * | 2015-11-25 | 2016-02-24 | 小米科技有限责任公司 | Fingerprint identification method and device |
| US10496897B2 (en) | 2015-11-25 | 2019-12-03 | Institute Of Automation Chinese Academy Of Sciences | Method and apparatus for recognizing RGB-D objects based on adaptive similarity measure of dense matching item |
| CN105477859B (en) * | 2015-11-26 | 2019-02-19 | 北京像素软件科技股份有限公司 | A kind of game control method and device based on user's face value |
| CN105472155B (en) * | 2015-12-04 | 2019-02-12 | 小米科技有限责任公司 | Background setting method and device |
| CN105447190B (en) * | 2015-12-18 | 2019-03-15 | 小米科技有限责任公司 | Picture retrieval method, device and server based on convolutional neural networks |
| CN105718869B (en) * | 2016-01-15 | 2019-07-02 | 网易（杭州）网络有限公司 | The method and apparatus of face face value in a kind of assessment picture |
| CN107045618B (en) * | 2016-02-05 | 2020-07-03 | 北京陌上花科技有限公司 | Facial expression recognition method and device |
| CN107122356B (en) * | 2016-02-24 | 2020-10-09 | 北京小米移动软件有限公司 | Method and device for displaying face value and electronic equipment |
| CN105574215B (en) * | 2016-03-04 | 2019-11-12 | 哈尔滨工业大学深圳研究生院 | A kind of instance-level image search method indicated based on multilayer feature |
| CN107305622B (en) * | 2016-04-15 | 2021-08-31 | 北京市商汤科技开发有限公司 | Face facial feature recognition method, device and system |
| CN107516061B (en) * | 2016-06-17 | 2020-04-07 | 北京市商汤科技开发有限公司 | Image classification method and system |
| CN106228137A (en) * | 2016-07-26 | 2016-12-14 | 广州市维安科技股份有限公司 | A kind of ATM abnormal human face detection based on key point location |
| CN107730483A (en) * | 2016-08-10 | 2018-02-23 | 阿里巴巴集团控股有限公司 | The methods, devices and systems of mobile device, processing face biological characteristic |
| CN106295566B (en) * | 2016-08-10 | 2019-07-09 | 北京小米移动软件有限公司 | Facial expression recognizing method and device |
| WO2018109505A1 (en) * | 2016-12-15 | 2018-06-21 | Google Llc | Transforming source domain images into target domain images |
| CN106778627B (en) * | 2016-12-20 | 2019-09-03 | 北京安云世纪科技有限公司 | Detect the method, apparatus and mobile terminal of face face value |
| CN106709448A (en) * | 2016-12-21 | 2017-05-24 | 华南理工大学 | Face convolutional neural network-based traditional Chinese medicine constitution recognition method |
| CN106647758A (en) * | 2016-12-27 | 2017-05-10 | 深圳市盛世智能装备有限公司 | Target object detection method and device and automatic guiding vehicle following method |
| CN106780906B (en) * | 2016-12-28 | 2019-06-21 | 北京品恩科技股份有限公司 | A kind of testimony of a witness unification recognition methods and system based on depth convolutional neural networks |
| CN106940790B (en) * | 2017-03-13 | 2020-05-19 | 重庆文理学院 | People flow congestion prediction method and system |
| CN107169408A (en) * | 2017-03-31 | 2017-09-15 | 北京奇艺世纪科技有限公司 | A kind of face value decision method and device |
| CN107122054A (en) * | 2017-04-27 | 2017-09-01 | 青岛海信医疗设备股份有限公司 | A kind of detection method and device of face deflection angle and luffing angle |
| CN107123083B (en) * | 2017-05-02 | 2019-08-27 | 中国科学技术大学 | Face edit methods |
| CN108932459B (en) * | 2017-05-26 | 2021-12-10 | 富士通株式会社 | Face recognition model training method and device and face recognition method |
| CN107423665A (en) * | 2017-06-05 | 2017-12-01 | 中南大学 | Three-dimensional face analysis method and its analysis system based on BP neural network |
| CN107368671A (en) * | 2017-06-07 | 2017-11-21 | 万香波 | System and method are supported in benign gastritis pathological diagnosis based on big data deep learning |
| CN107368670A (en) * | 2017-06-07 | 2017-11-21 | 万香波 | Stomach cancer pathology diagnostic support system and method based on big data deep learning |
| CN109003659A (en) * | 2017-06-07 | 2018-12-14 | 万香波 | Stomach Helicobacter pylori infects pathological diagnosis and supports system and method |
| CN107342962B (en) * | 2017-07-03 | 2019-12-13 | 北京邮电大学 | deep learning intelligent constellation diagram analysis method based on convolutional neural network |
| CN107463943B (en) * | 2017-07-10 | 2020-07-21 | 北京小米移动软件有限公司 | Color value scoring method, training method and device of same-person color value difference classifier |
| CN107424130B (en) * | 2017-07-10 | 2020-09-11 | 北京小米移动软件有限公司 | Picture beautifying method and device |
| CN107622240B (en) | 2017-09-21 | 2023-10-13 | 百度在线网络技术（北京）有限公司 | Face detection method and device |
| CN107886070A (en) * | 2017-11-10 | 2018-04-06 | 北京小米移动软件有限公司 | Verification method, device and the equipment of facial image |
| CN107818319A (en) * | 2017-12-06 | 2018-03-20 | 成都睿码科技有限责任公司 | A kind of method of automatic discrimination face beauty degree |
| CN109934047A (en) * | 2017-12-15 | 2019-06-25 | 浙江舜宇智能光学技术有限公司 | Face identification system and its face identification method based on deep learning |
| CN108121969B (en) * | 2017-12-22 | 2021-12-28 | 百度在线网络技术（北京）有限公司 | Method and apparatus for processing image |
| CN108764334A (en) * | 2018-05-28 | 2018-11-06 | 北京达佳互联信息技术有限公司 | Facial image face value judgment method, device, computer equipment and storage medium |
| CN109033938A (en) * | 2018-06-01 | 2018-12-18 | 上海阅面网络科技有限公司 | A kind of face identification method based on ga s safety degree Fusion Features |
| CN108629338B (en) * | 2018-06-14 | 2021-10-19 | 五邑大学 | Face beauty prediction method based on LBP and convolutional neural network |
| CN109214298B (en) * | 2018-08-09 | 2021-06-08 | 盈盈（杭州）网络技术有限公司 | Asian female color value scoring model method based on deep convolutional network |
| CN109465676B (en) * | 2018-09-13 | 2020-10-16 | 成都数之联科技有限公司 | Tool life prediction method |
| CN109657539B (en) * | 2018-11-05 | 2022-01-25 | 达闼机器人有限公司 | Face value evaluation method and device, readable storage medium and electronic equipment |
| CN109711358B (en) * | 2018-12-28 | 2020-09-04 | 北京远鉴信息技术有限公司 | Neural network training method, face recognition system and storage medium |
| CN110222571B (en) * | 2019-05-06 | 2023-04-07 | 平安科技（深圳）有限公司 | Intelligent judgment method and device for black eye and computer readable storage medium |
| CN110189249B (en) * | 2019-05-24 | 2022-02-18 | 深圳市商汤科技有限公司 | Image processing method and device, electronic equipment and storage medium |
| CN110222623A (en) * | 2019-05-31 | 2019-09-10 | 深圳市恩钛控股有限公司 | Micro- expression analysis method and system |
| CN110472489B (en) * | 2019-07-05 | 2023-05-05 | 五邑大学 | Face beauty grade prediction method, device and storage medium |
| CN110430397A (en) * | 2019-07-31 | 2019-11-08 | 广州竞德信息技术有限公司 | Based on multi-modal Activity recognition method |
| CN110503149B (en) * | 2019-08-26 | 2022-06-14 | 咪咕文化科技有限公司 | Method and system for classifying local features in image |
| CN110866437A (en) * | 2019-09-23 | 2020-03-06 | 平安科技（深圳）有限公司 | Color value determination model optimization method and device, electronic equipment and storage medium |
| CN111062260B (en) * | 2019-11-25 | 2024-03-05 | 杭州绿度信息技术有限公司 | Automatic generation method of face-beautifying recommendation scheme |
| CN111177469A (en) * | 2019-12-20 | 2020-05-19 | 国久大数据有限公司 | Face retrieval method and face retrieval device |
| CN112183669B (en) * | 2020-11-04 | 2024-02-13 | 航天科工(北京)空间信息应用股份有限公司 | Image classification method, device, equipment and storage medium |
| CN112733705A (en) * | 2021-01-07 | 2021-04-30 | 中科魔镜(深圳)科技发展有限公司 | 3D intelligent analysis system based on human face |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103824054A (en) * | 2014-02-17 | 2014-05-28 | 北京旷视科技有限公司 | Cascaded depth neural network-based face attribute recognition method |
| CN104463243A (en) * | 2014-12-01 | 2015-03-25 | 中科创达软件股份有限公司 | Sex detection method based on average face features |
| CN104463172A (en) * | 2014-12-09 | 2015-03-25 | 中国科学院重庆绿色智能技术研究院 | Face feature extraction method based on face feature point shape drive depth model |
| CN104504376A (en) * | 2014-12-22 | 2015-04-08 | 厦门美图之家科技有限公司 | Age classification method and system for face images |
-
2015
- 2015-04-18 CN CN201510185613.1A patent/CN104850825B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103824054A (en) * | 2014-02-17 | 2014-05-28 | 北京旷视科技有限公司 | Cascaded depth neural network-based face attribute recognition method |
| CN104463243A (en) * | 2014-12-01 | 2015-03-25 | 中科创达软件股份有限公司 | Sex detection method based on average face features |
| CN104463172A (en) * | 2014-12-09 | 2015-03-25 | 中国科学院重庆绿色智能技术研究院 | Face feature extraction method based on face feature point shape drive depth model |
| CN104504376A (en) * | 2014-12-22 | 2015-04-08 | 厦门美图之家科技有限公司 | Age classification method and system for face images |
Non-Patent Citations (1)
| Title |
|---|
| Deep Learning Face Representation from Predicting 10000 Classes;Yi Sun etc.;《Computer Vision and Pattern Recognition(CVPR),2014IEEE Conference on.》;20140925;正文第1891－1898页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104850825A (en) | 2015-08-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104850825B (en) | A kind of facial image face value calculating method based on convolutional neural networks | |
| CN108985377B (en) | A kind of image high-level semantics recognition methods of the multiple features fusion based on deep layer network | |
| CN108182441B (en) | Parallel multichannel convolutional neural network, construction method and image feature extraction method | |
| CN107437092B (en) | The classification method of retina OCT image based on Three dimensional convolution neural network | |
| CN104463100B (en) | Intelligent wheel chair man-machine interactive system and method based on human facial expression recognition pattern | |
| CN101819628B (en) | Method for performing face recognition by combining rarefaction of shape characteristic | |
| CN106778468B (en) | 3D face identification method and equipment | |
| CN109165674A (en) | A kind of certificate photo classification method based on multi-tag depth convolutional network | |
| Youssif et al. | Arabic sign language (arsl) recognition system using hmm | |
| CN109815826A (en) | The generation method and device of face character model | |
| CN106909220A (en) | A kind of sight line exchange method suitable for touch-control | |
| CN105005765A (en) | Facial expression identification method based on Gabor wavelet and gray-level co-occurrence matrix | |
| CN104992142A (en) | Pedestrian recognition method based on combination of depth learning and property learning | |
| CN106529504B (en) | A kind of bimodal video feeling recognition methods of compound space-time characteristic | |
| CN104008375B (en) | The integrated face identification method of feature based fusion | |
| CN105678235B (en) | Three-dimensional face expression recognition methods based on representative region various dimensions feature | |
| CN108052884A (en) | A kind of gesture identification method based on improvement residual error neutral net | |
| CN104143079A (en) | Method and system for face attribute recognition | |
| CN109344692A (en) | A kind of motion quality evaluation method and system | |
| CN107862249A (en) | A kind of bifurcated palm grain identification method and device | |
| CN105718873A (en) | People stream analysis method based on binocular vision | |
| CN105138974B (en) | A kind of multi-modal Feature fusion of finger based on Gabor coding | |
| CN105956570B (en) | Smiling face's recognition methods based on lip feature and deep learning | |
| Singh et al. | Indian Sign Language Recognition System for Dynamic Signs | |
| CN110046544A (en) | Digital gesture identification method based on convolutional neural networks |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| EXSB | Decision made by sipo to initiate substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |