CN105956571B - A kind of age estimation method of facial image - Google Patents

A kind of age estimation method of facial image Download PDF

Info

Publication number
CN105956571B
CN105956571B CN201610317460.6A CN201610317460A CN105956571B CN 105956571 B CN105956571 B CN 105956571B CN 201610317460 A CN201610317460 A CN 201610317460A CN 105956571 B CN105956571 B CN 105956571B
Authority
CN
China
Prior art keywords
facial image
matrix
estimation method
layer
age
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
Application number
CN201610317460.6A
Other languages
Chinese (zh)
Other versions
CN105956571A (en
Inventor
杜吉祥
郑德鹏
翟传敏
范文涛
王靖
刘海建
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Huaqiao University
Original Assignee
Huaqiao University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Huaqiao University filed Critical Huaqiao University
Priority to CN201610317460.6A priority Critical patent/CN105956571B/en
Publication of CN105956571A publication Critical patent/CN105956571A/en
Application granted granted Critical
Publication of CN105956571B publication Critical patent/CN105956571B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • 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
    • G06V40/168—Feature extraction; Face representation
    • 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)
  • Oral & Maxillofacial Surgery (AREA)
  • General Health & Medical Sciences (AREA)
  • Human Computer Interaction (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Image Analysis (AREA)

Abstract

A kind of age estimation method of facial image of the invention, this method include five parts: (1) piecemeal of image；(2) with PCA algorithm study piecemeal, treated that matrix obtains the convolution kernel of convolution operation；(3) convolution operation then is carried out with the convolution kernel that PCA algorithm learns；(4) use the mode of binaryzation in second convolutional layer followed by Nonlinear Processing；(5) feature extraction is carried out with the method for statistics with histogram.Age value is estimated with non-linear support vector regression K-SVR (Kernel function Support Vector Regression) after this method extraction feature, shows the accuracy rate that can greatly improve age estimation by experiment.

Description

A kind of age estimation method of facial image
Technical field
The present invention relates to the face ages to estimate field, the age of especially a kind of facial image based on deep learning model Estimation method.
Background technique
In face age automatic estimating system, two stages are generally divided into, first stage is to extract age characteristics, the Two stages are the estimation ages, and the emphasis usually studied is how to extract optimal age characteristics.Existing age estimating system In, study or the method for extracting age characteristics can be generally divided into manual extraction age characteristics and automatic learning age feature.Hand The dynamic representative method of age characteristics of extracting has LBP (Local Binary Patterns, LBP), SIFT (Scale- Invariant Feature Transform, SIFT), subspace model etc., manual extraction feature there are the shortcomings that for by people For the influence of subjectivity selection.Although extraordinary effect may be obtained in certain fields by manually selecting feature, being primarily due to can With the feature adapted to for a certain specific data type selection, but facing to new data or it is new under conditions of, manually Feature mode is different is surely suitble to for selection, is difficult the application of adaptation in practice in this way.
The deep learning that nearly this year rises, is also used to learn or extract age characteristics.Deep learning is in machine learning A field, motivation establish, the neural network analyzed of simulation human brain, it imitates the mechanism of human brain to explain number According to.Deep learning structure contains more hidden layers, by combination low-level feature formed it is more abstract it is high-rise indicate attribute classification or Feature, to find that the distribution characteristics of data indicates.But existing deep learning structure parameter when hidden layer number increases sharply increases, Higher requirement is proposed to operational performance.This is also to be faced using deep learning and problem to be solved.
Summary of the invention
It is a primary object of the present invention to overcome drawbacks described above in the prior art, propose a kind of based on deep learning The age estimation method of the facial image of DLPCANet model, this method are capable of providing a kind of efficient learning age feature, have Effect solves the problems, such as that model parameter is excessive, and realizes Fast Learning and high-precision estimation age.
The present invention adopts the following technical scheme:
A kind of age estimation method of facial image, it is characterised in that:
1) after the facial image of input being pre-processed, then piecemeal, equalization is carried out to each block, is successively handled The facial image of all inputs, to obtain the big matrix that one includes all pieces；
2) with PCA algorithm from the acquistion of big matrix middle school to the convolution kernel of first layer convolution operation；
3) the output feature maps after first layer convolution operation is subjected to piecemeal and equalization is gone to handle, obtain another big square Battle array, the convolution kernel using PCA algorithm from the big matrix middle school acquistion to second layer convolution operation；
4) the feature maps exported after second layer convolution operation is subjected to Nonlinear Processing；
5) output valve after conversion is divided into several pieces, each block is counted with histogram, all pieces are connected At a vector, the feature of input picture is obtained；
6) feature of input picture is estimated into the age with non-linear support vector regression.
Preferably, the pretreatment described in step 1) be binaryzation, smooth and standardization, obtain gray level image.
Preferably, in step 1), it is assumed that for input picture after pretreatment, obtaining size is m × n-pixel grayscale image All obtained gray level images are carried out piecemeal by picture respectively, it is assumed that the size of each block is p1×p2, step in blocking process It cuts down as s1=s2=1, then the piecemeal result of the i-th-th input pictures is expressed asWhereinTo AiIn each block carry out equalization after be It successively handles all input pictures and obtains a big matrix, be expressed asWhereinThe size for indicating each block is p1p2× 1,The size of representing matrix A is p1p2×Nm1n1, the value model of i It encloses for 1≤i≤N.
Preferably, in step 2), it is described with PCA algorithm from the acquistion of big matrix middle school to the volume of first layer convolution operation Product core, it is specific as follows, it is assumed that the 1st layer of convolution kernel number is L1, PCA algorithm be so that following objective functions reconstructed error most It is small, objective function are as follows:
WhereinFor unit matrix, size L1×L1, solving the objective function is to matrix A ATCarry out characteristic value It decomposes, chooses L1Feature vector corresponding to a maximum characteristic value is as follows as its expression formula of the convolution kernel of convolution operation:
WhereinIndicate handleIt is transformed into matrixFormat, pl(ATA solution matrix) is indicated ATThe main feature vector of A.
Preferably, in step 3), it is assumed that the convolution operation of second convolutional layer isWhereinTable Show the output valve of first convolutional layer, Wl 2It is convolution kernel.
Preferably, in step 4), the Nonlinear Processing includes first binaryzation, is reconverted into decimal value, table It is up to formulaWherein δ indicates binary conversion treatment,Value range be
Preferably, in step 6), using the Selection of kernel function radial basis function in non-linear support vector regression.
By the above-mentioned description of this invention it is found that compared with prior art, the invention has the following beneficial effects:
Method of the invention carries out feature extraction based on deep learning model, with non-linear support vector regression K-SVR (Kernel function Support Vector Regression) estimates age value, can effectively solve model parameter Excessive problem, and realize Fast Learning and high-precision estimation age.
Detailed description of the invention
Fig. 1 is deep learning model flow figure of the invention, wherein patch-mean removal expression go equalization, Convolutional Layer C1 indicates that first convolutional layer, Convolutional Layer C2 indicate second convolution Layer, Binary quantization and mapping indicate binaryzation and its mapping, Concatenated image Block-wise histogram indicates all pieces in connection image of statistics with histogram.
Fig. 2 is input picture piecemeal (patch) processing schematic.
Fig. 3 is the window form for acquiring feature vector and being transformed into convolution kernel.
Fig. 4 is first convolutional layer operation chart.
Fig. 5 is second convolutional layer operation chart.
Fig. 6 is the schematic diagram of non-linear layer operation.
Fig. 7 is that feature extraction layer carries out piecemeal schematic diagram.
Fig. 8 is the sample schematic diagram of two ages public database MORPH and FG-NET.
Fig. 9 is in two public database MORPH and FG-NET part estimation age values.
Specific embodiment
Below by way of specific embodiment, the invention will be further described.
In Fig. 1, it is known that, the present invention is the age estimation method based on deep learning model DLPCANet, the depth Practising model includes three layers, is convolutional layer (The Convolutional Layer), non-linear layer (The Nonlinear respectively Process Layer), feature extraction layer (Feature Pooling Layer).
Referring to Figure 1 to Figure 7, the age estimation method of a kind of facial image of the invention, includes the following steps:
1) facial image of input is pre-processed, including binaryzation, smooth and standardization, obtaining size is m × n picture All obtained gray level images are carried out piecemeal, as illustrated in fig. 2, it is assumed that the size of each block is by the gray level image of element respectively p1×p2, the paces in blocking process are s1=s2=1, blocking process can also regard that with a window size be s as1=s2= 1 is scanned the row and column of image.Then the piecemeal result of the i-th-th input pictures is expressed asWhereinTo AiIn each block gone It is after equalizationIt successively handles all input pictures and obtains a big matrix, indicate Are as follows:
WhereinThe size for indicating each block is p1p2× 1,The size of representing matrix A is p1p2× Nm1n1, the value range of i is 1≤i≤N.
2) with PCA algorithm from the acquistion of big matrix A middle school to the convolution kernel of first layer convolution operation.Wherein convolution kernel number is just It is that PCA with algorithm acquires feature vector.As shown in figure 3, be solution to feature vector be transformed into the window-shaped of convolution kernel W Formula.Assuming that the 1st layer of convolution kernel number is L1, PCA algorithm is so that following objective functions reconstructed error minimum, objective function Are as follows:
WhereinFor unit matrix, size L1×L1, solving the objective function in fact is exactly to matrix A ATIt carries out special Value indicative is decomposed, and L is chosen1Convolution kernel of the feature vector as convolution operation corresponding to a maximum characteristic value, expression formula is such as Under:
WhereinIndicate handleIt is transformed into matrixFormat, pl(A TA solution matrix) is indicated ATThe main feature vector of A.
Fig. 4 is first convolutional layer operation chart, it is assumed that input picture isLearn to obtain using PCA algorithm The convolution kernel of one layer convolution operation is Wl 1, l=1,2 ..., L1, go to be rolled up with input picture using different convolution kernels When product, so that it may obtain different feature maps.So the convolution output of first convolutional layer can be write as:
Wherein, * indicates convolution operation, exports feature mapsSize and input IiEqually, the reason is that carrying out convolution Zero padding operation in edge is carried out to input picture when operation, so that the size after convolution is such as input.
3) the output feature maps after first layer convolution operation is subjected to piecemeal and goes equalization, obtain another big matrix, Convolution kernel using PCA algorithm from the big matrix middle school acquistion to second layer convolution operation.Assuming that the convolution of second convolutional layer Operation isWhereinIndicate the output valve of first convolutional layer, Wl 2It is convolution kernel.It is specific: Fig. 5 table Show second convolutional layer operation chart, the output of first convolutional layerInput and first layer as second convolutional layer Convolution equally does same operation, by inputIt carries out piecemeal and can be indicated after going equalization are as follows:
Wherein l=1,2 ... .L2, the input of all second layers is handled in the same way, it is available
WhereinSpy corresponding to dominant eigenvalue with PCA algorithm solution matrix Y Levy convolution kernel W of the vector as second convolutional layerl 2.So the convolution output of second convolutional layer are as follows:
4) Fig. 6 indicates non-linear layer operation chart, the feature maps exported after second layer convolution operation is carried out non-thread Property processing, comprising: first binaryzation is reconverted into decimal value, and expression formula is
Wherein δ indicates binary conversion treatment,Value range be
5) output valve after conversion is divided into several pieces, each block is counted with histogram, all pieces are connected At a vector, the feature of input picture is obtained.Specifically: referring to Fig. 7, it is characterized abstraction, layer and carries out piecemeal schematic diagram, it is non-thread The output of property layerAs the input value of feature extraction layer,It is divided into Z-block, size h1×h2.Each block is used The method of statistics with histogram is counted, and then all blocks is connected into the form of a vector, is expressed asSuccessively handleL=1,2 ..., L1, input picture I may finally be obtainediFeature are as follows:
The sample of all inputs is handled in the same way, may finally extract the feature of input sample, is so being used Non-linear support vector regression.
6) by the feature of input picture with non-linear support vector regression K-SVR (Kernel function Supportive Vector Regression, K-SVR) estimate the age.Using the core letter in non-linear support vector regression Number selection radial basis function RBF (Radial-Basis Function).
It illustrates
Referring to Fig. 8, face age public database MORPH and FG-NET, the facial image of the two lane databases has Corresponding age information.MORPH database includes 55332 facial images, while also including not agnate people, age From 16 years old to 77 years old.FG-NET database has 1002 facial images, and the image of the inside has highly variable (such as illumination, appearance Gesture, expression etc.), age several years was from 0 years old to 69 years old.
5000 images are randomly selected as training set 5000 to MORPH database and open image as test set, and for The image that small FG-NET database then randomly selects 80% is used as test set as training set, remaining 20%.All figures As based on become after pretreatment size be 28*28 gray level image.Final prediction result is respectively as follows: in MORPH lane database Age mean absolute error reached for 4.72 (ages), reached for 4.66 (years in the age mean absolute error of FG-NET lane database Age), wherein mean absolute error is calculated asTkIndicate real age value, PkThe year estimated Age value.Fig. 9 is in two public database MORPH and FG-NET part estimation age values.
The above is only a specific embodiment of the present invention, but the design concept of the present invention is not limited to this, all to utilize this Design makes a non-material change to the present invention, and should all belong to behavior that violates the scope of protection of the present invention.

Claims (7)

1. a kind of age estimation method of facial image, it is characterised in that:
1) after the facial image of input being pre-processed, then piecemeal, equalization is carried out to each block, successively processing is all The facial image of input, to obtain the big matrix that one includes all pieces；
2) with PCA algorithm from the acquistion of big matrix middle school to the convolution kernel of first layer convolution operation；
3) the output feature maps after first layer convolution operation is subjected to piecemeal and equalization is gone to handle, obtain another big matrix, Convolution kernel using PCA algorithm from the big matrix middle school acquistion to second layer convolution operation；
4) the feature maps exported after second layer convolution operation is subjected to Nonlinear Processing；
5) output valve after conversion is divided into several pieces, each block is counted with histogram, connects into one for all pieces A vector obtains the feature of input picture；
6) feature of input picture is estimated into the age with non-linear support vector regression.
2. a kind of age estimation method of facial image as described in claim 1, it is characterised in that: described in step 1) Pretreatment is smooth and standardization, obtains gray level image.
3. a kind of age estimation method of facial image as described in claim 1, it is characterised in that: in step 1), it is assumed that Input picture by pretreatment after, obtain size be m × n-pixel gray level image, by all obtained gray level images respectively into Row piecemeal, it is assumed that the size of each block is p1×p2, the paces in blocking process are s1=s2=1, then the i-th-th inputs are schemed The piecemeal result of picture is expressed asWhereinIt is right AiIn each block carry out equalization after beAll input pictures are successively handled to obtain To a big matrix, it is expressed asWhereinIndicate the size of each block For p1p2× 1,The size of representing matrix A is p1p2×Nm1n1, the value range of i is 1≤i≤N.
4. a kind of age estimation method of facial image as claimed in claim 1 or 3, it is characterised in that: in step 2), institute State with PCA algorithm from the acquistion of big matrix middle school to the convolution kernel of first layer convolution operation, it is specific as follows, it is assumed that the 1st layer of volume Product core number is L1, PCA algorithm is so that following objective functions reconstructed error minimum, objective function are as follows:
WhereinFor unit matrix, size L1×L1, solving the objective function is to matrix A ATEigenvalues Decomposition is carried out, Choose L1Feature vector corresponding to a maximum characteristic value is as follows as its expression formula of the convolution kernel of convolution operation:
WhereinIndicate handleIt is transformed into matrixFormat, pl(A TA solution matrix A) is indicatedTA Main feature vector.
5. a kind of age estimation method of facial image as claimed in claim 4, it is characterised in that: in step 3), it is assumed that The convolution operation of second convolutional layer isWhereinIndicate the output valve of first convolutional layer, Wl 2It is volume Product core.
6. a kind of age estimation method of facial image as claimed in claim 5, it is characterised in that: described in step 4) Nonlinear Processing include first binaryzation, be reconverted into decimal value, expression formula isWherein δ Indicate binary conversion treatment,Value range be
7. a kind of age estimation method of facial image as described in claim 1, it is characterised in that: in step 6), use Selection of kernel function radial basis function in non-linear support vector regression.
CN201610317460.6A 2016-05-13 2016-05-13 A kind of age estimation method of facial image Active CN105956571B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201610317460.6A CN105956571B (en) 2016-05-13 2016-05-13 A kind of age estimation method of facial image

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201610317460.6A CN105956571B (en) 2016-05-13 2016-05-13 A kind of age estimation method of facial image

Publications (2)

Publication Number Publication Date
CN105956571A CN105956571A (en) 2016-09-21
CN105956571B true CN105956571B (en) 2019-03-12

Family

ID=56912476

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201610317460.6A Active CN105956571B (en) 2016-05-13 2016-05-13 A kind of age estimation method of facial image

Country Status (1)

Country Link
CN (1) CN105956571B (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106454315A (en) * 2016-10-26 2017-02-22 深圳市魔眼科技有限公司 Adaptive virtual view-to-stereoscopic view method and apparatus, and display device
CN108629264B (en) * 2017-03-18 2022-09-27 上海荆虹电子科技有限公司 Method and apparatus for image processing

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101877054A (en) * 2009-11-23 2010-11-03 北京中星微电子有限公司 Method and device for determining age of face image
CN104036247A (en) * 2014-06-11 2014-09-10 杭州巨峰科技有限公司 Facial feature based face racial classification method
CN105426872A (en) * 2015-12-17 2016-03-23 电子科技大学 Face age estimation method based on correlation Gaussian process regression
CN105447473A (en) * 2015-12-14 2016-03-30 江苏大学 PCANet-CNN-based arbitrary attitude facial expression recognition method

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8582807B2 (en) * 2010-03-15 2013-11-12 Nec Laboratories America, Inc. Systems and methods for determining personal characteristics
US9400925B2 (en) * 2013-11-15 2016-07-26 Facebook, Inc. Pose-aligned networks for deep attribute modeling

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101877054A (en) * 2009-11-23 2010-11-03 北京中星微电子有限公司 Method and device for determining age of face image
CN104036247A (en) * 2014-06-11 2014-09-10 杭州巨峰科技有限公司 Facial feature based face racial classification method
CN105447473A (en) * 2015-12-14 2016-03-30 江苏大学 PCANet-CNN-based arbitrary attitude facial expression recognition method
CN105426872A (en) * 2015-12-17 2016-03-23 电子科技大学 Face age estimation method based on correlation Gaussian process regression

Also Published As

Publication number Publication date
CN105956571A (en) 2016-09-21

Similar Documents

Publication Publication Date Title
CN110599409B (en) Convolutional neural network image denoising method based on multi-scale convolutional groups and parallel
CN105657402B (en) A kind of depth map restoration methods
CN111242841B (en) Image background style migration method based on semantic segmentation and deep learning
CN108734138B (en) Melanoma skin disease image classification method based on ensemble learning
CN110533683B (en) Image omics analysis method fusing traditional features and depth features
CN104285239B (en) Image processing apparatus, image processing method and printed medium
CN105844285A (en) Cucumber disease identification method and apparatus based on image information
CN105654436A (en) Backlight image enhancement and denoising method based on foreground-background separation
CN107871099A (en) Face detection method and apparatus
CN111127387B (en) Quality evaluation method for reference-free image
CN108121962B (en) Face recognition method, device and equipment based on nonnegative adaptive feature extraction
CN102289670B (en) Image characteristic extraction method with illumination robustness
CN109344898A (en) Convolutional neural networks image classification method based on sparse coding pre-training
CN113066025B (en) Image defogging method based on incremental learning and feature and attention transfer
CN107832786A (en) A kind of recognition of face sorting technique based on dictionary learning
CN109472733A (en) Image latent writing analysis method based on convolutional neural networks
CN116051408A (en) Image depth denoising method based on residual error self-coding
CN105956571B (en) A kind of age estimation method of facial image
CN106803105B (en) Image classification method based on sparse representation dictionary learning
CN114897884A (en) No-reference screen content image quality evaluation method based on multi-scale edge feature fusion
CN107967674A (en) Nuclear magnetic resonance image denoising method based on image block self-similarity priori
CN113011506B (en) Texture image classification method based on deep fractal spectrum network
CN109242879A (en) Brain glioma nuclear-magnetism image partition method based on depth convolutional neural networks
CN111652238B (en) Multi-model integration method and system
CN109887023B (en) Binocular fusion stereo image quality evaluation method based on weighted gradient amplitude

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant