CN108764106A - Multiple dimensioned colour image human face comparison method based on cascade structure - Google Patents
Multiple dimensioned colour image human face comparison method based on cascade structure Download PDFInfo
- Publication number
- CN108764106A CN108764106A CN201810497963.5A CN201810497963A CN108764106A CN 108764106 A CN108764106 A CN 108764106A CN 201810497963 A CN201810497963 A CN 201810497963A CN 108764106 A CN108764106 A CN 108764106A
- Authority
- CN
- China
- Prior art keywords
- face alignment
- decision tree
- face
- rgb
- value
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
Abstract
The present invention proposes that a kind of multiple dimensioned colour image human face comparison method based on cascade structure, this method include:(1)For coloured image RGB triple channels, the facial image in each channel is to carrying out multiple dimensioned face characteristic extraction;(2)For RGB triple channels, face alignment decision tree, the corresponding triple channel face alignment decision tree classifier of training are built respectively;(3)Based on Softcascade cascade structures, joint triple channel face alignment decision tree classifier exports face alignment mathematical expectation of probability, obtains face alignment combined chance value.The present invention compares problem for multiple dimensioned colour image human face, and by extracting across dimension normalization pixel difference feature, training is based on Softcascade cascade structure face alignment decision-tree models, can effectively improve multiple dimensioned colour image human face matching identification rate.
Description
Technical field
The present invention is to belong to computer vision, face alignment and identification field, and specific design is for the multiple dimensioned coloured silk of input
The face comparison method of color image.
Background technology
Biological characteristic is attribute of people itself, has very strong individual differences and self stability.In recent years, as life
Face alignment algorithm in object identification technology, it has many characteristics, such as real-time, substantivity, convenience, is more and more calculated
Machine visual field expert and scholar's research.With the development of machine learning algorithm and the extensive use of face alignment, face alignment
Problem is just gradually developing into the hot spot of research.It is played in the fields such as automatic monitoring, alarm, authentication and finance activities simultaneously
Increasingly important role.
But current face alignment is confined to the processing of gray-scale map on algorithm research, has ignored coloured image in face alignment
Research.Multiple dimensioned different size image is cannot achieve during face alignment simultaneously to compare, needed before face alignment by
Image loses image information during scaling and amplification and causes distortion etc. and ask to redefining the uniform sizes of image pair
Topic, wastes computing resource, and information use is not enough and reduces face alignment discrimination.
Invention content
Present in existing face alignment technology (1) is studied just for gray-scale map and ignores coloured image in face ratio
To important function;(2) it needs to redefine uniform sizes in face of different input size pictures, cannot achieve multiple dimensioned face
The problems such as comparison, the present invention propose a kind of in face of multiple dimensioned colour image human face comparison method, compared to the prior art, this hair
It is bright targetedly to be applied in face in RGB triple channels using decision tree and cascade structure, while being directed to input picture pair
Size it is different, it is proposed that a kind of feature extracting method across size realizes multiple dimensioned colour image human face and compares, into
The discrimination of face alignment image algorithm is improved and improved to one step.
Multiple dimensioned colour image human face comparison method based on cascade structure, feature include the following steps:
Step 1:For coloured image RGB triple channels, the multiple dimensioned facial image in each channel returns extraction respectively
One changes pixel difference feature;
Step 2:Using the normalization pixel difference feature of extraction, face alignment decision tree, the corresponding threeway of training are built respectively
Road face alignment decision tree classifier;
Step 3:Based on Softcascade cascade structures, joint triple channel face alignment decision tree output face alignment is general
Rate mean value obtains face alignment combined chance value.Complete multiple dimensioned, colour image human face pair face alignment probability output.
Further, step 1 includes the following steps:
Step 1.1:Training data sample set D includes the P positive sample pair and N with label information of different scale size
A negative sample pair, for training dataset RGB Three Channel Color image patterns pair, respectively to the multiple dimensioned face in each channel
Any two points extraction normalization pixel difference feature of image pattern pair, realizes that across scale feature extraction, formula are as follows:
Wherein, AM×M、BN× N is the RGB color facial image for being respectively M × M sizes and N × N sizes;For R, G, B value that A image RGB respective channel coordinates are x points;For B images RGB
Corresponding channel coordinate is R, G, B value of y points;Normalization pixel difference is characterized as m2×n2The feature vector of dimension;F(B)、F(B)、F(B)
For the normalization pixel difference feature extracted for the RGB of image A and image B.
Further, step 2 includes the following steps:
Step 2.1:The training dataset RGB triple channels normalization pixel difference feature f (x, y) obtained by step 1, difference needle
Face alignment decision tree classifier is trained to RGB triple channels.Assuming that given input normalizes pixel difference feature f, decision leaf
Child node exports face alignment similarity probability S.According to split vertexes function, that is, maximum value purity function gamma, pixel will be normalized
The split vertexes of poor feature construction decision tree, input feature vector f train dual threshold α by the method for exhaustion1, α2, according to dual threshold α1, α2
Determine the left and right of each split vertexes of decision tree structure
Maximum value purity function gamma enables training sample data collection distinguish positive negative sample to the greatest extent, and formula is as follows:
Wherein, uP、vNTo be split off the number of positive sample P and negative sample N that node is distinguished in decision tree.
It is assumed that in the decision tree for N number of node that depth is κ, wherein there is L leaf node.In each leafy node lL
It can indicate the probability distribution s of image similarityl∈S.Each node of decision tree will under the constraint of maximum value purity function gamma
Each leaf node exports image similarity probability value S, it is assumed that label is distributed as σ on corresponding leaf nodel, decision tree
The face alignment similarity probability value of one training sample pair can be expressed as:
Wherein,
Step 2.2:Mechanism is cascaded based on Softcascade, trained decision tree is cascaded into face alignment strong classifier,
And decision tree leaf node face alignment mathematical expectation of probability is exported, formula is as follows:
Wherein, C1、C2、C3For the decision tree number of RGB triple channels training;For RGB channel cascade point
The face alignment probability value of i-th of tree of class device output;It is equal for RGB triple channel decision tree face alignment probability
Value.
Further, above-mentioned steps 3 include the following steps:
Step 3.1:It is general according to the Softcascade cascade classifier decision tree leaf node face alignments of RGB triple channels
Rate mean value, can respectively obtain face alignment combined chance value V isMean value, maximum value or minimum value,
Formula difference is as follows:
Wherein,For the face alignment mathematical expectation of probability of RGB triple channel decision trees;L1、L2、L3For RGB
The decision tree number of triple channel training.
Step 3.2:By targetedly training the respective face alignment decision tree classifier of triple channel and being cascaded into one
Complete strong classifier, it is comprehensive that the face alignment mathematical expectation of probability that comprehensive triple channel face alignment strong classifier exports obtains face alignment
Close probability value V, setting face alignment similarity threshold τ.Comprehensive face alignment combined chance value V and face alignment similarity threshold
τ, the judgement of face alignment similarity are as follows:
Description of the drawings
Fig. 1 is total techniqueflow chart;
Fig. 2 is Softcascade cascade structure face alignment decision tree structure figures.
Specific implementation mode
The invention will be further described below in conjunction with the accompanying drawings.
As shown in Figure 1, the multiple dimensioned colour image human face comparison method based on cascade structure of the present invention, including walk as follows
Suddenly:
Step 1:For coloured image RGB triple channels, the multiple dimensioned facial image in each channel returns extraction respectively
One changes pixel difference feature.Sub-step is as follows:
Step 1.1:Training data sample set D includes the P positive sample pair and N with label information of different scale size
A negative sample pair, for training dataset RGB Three Channel Color image patterns pair, respectively to the multiple dimensioned face in each channel
Any two points extraction normalization pixel difference feature of image pattern pair, realizes that across scale feature extraction, formula are as follows:
Wherein, AM×M、BN× N is the RGB color facial image for being respectively M × M sizes and N × N sizes;For R, G, B value that A image RGB respective channel coordinates are x points;For B images RGB
Corresponding channel coordinate is R, G, B value of y points;Normalization pixel difference is characterized as m2×n2The feature vector of dimension;F(B)、F(B)、F(B)
For the normalization pixel difference feature extracted for the RGB of image A and image B.
Step 2:Using the normalization pixel difference feature of extraction, face alignment decision tree, the corresponding threeway of training are built respectively
Road face alignment decision tree classifier.Sub-step is as follows:
Step 2.1:The training dataset RGB triple channels normalization pixel difference feature f (x, y) obtained by step 1, difference needle
Face alignment decision tree classifier is trained to RGB triple channels.Assuming that given input normalizes pixel difference feature f, decision leaf
Child node exports face alignment similarity probability S.According to split vertexes function, that is, maximum value purity function gamma, pixel will be normalized
The split vertexes of poor feature construction decision tree, input feature vector f train dual threshold α by the method for exhaustion1, α2, according to dual threshold α1, α2
It determines that the left and right route of each split vertexes of decision tree structure is distinguished, it is as follows to distinguish mode:
Maximum value purity function gamma enables training sample data collection distinguish positive negative sample to the greatest extent, and formula is as follows:
Wherein, uP、vNTo be split off the number of positive sample P and negative sample N that node is distinguished in decision tree.
It is assumed that in the decision tree for N number of node that depth is κ, wherein there is L leaf node.In each leafy node l ∈ L
It can indicate the probability distribution s of image similarityl∈S.Each node of decision tree is under the constraint of maximum value purity function gamma
By each leaf node output image similarity probability value S, it is assumed that label is distributed as σ on corresponding leaf nodel, decision
Tree can be expressed as the face alignment similarity probability value of a training sample pair:
Wherein,
Step 2.2:Mechanism is cascaded based on Softcascade, trained decision tree is cascaded into face alignment strong classifier,
And decision tree leaf node face alignment mathematical expectation of probability is exported, formula is as follows:
Wherein, C1、C2、C3For the decision tree number of RGB triple channels training;For RGB channel cascade point
The face alignment probability value of i-th of tree of class device output;It is equal for RGB triple channel decision tree face alignment probability
Value.
Step 3:Based on Softcascade cascade structures, joint triple channel face alignment decision tree output face alignment is general
Rate mean value obtains face alignment combined chance value.Complete multiple dimensioned, colour image human face pair face alignment probability output.Son
Steps are as follows:
Step 3.1:It is general according to the Softcascade cascade classifier decision tree leaf node face alignments of RGB triple channels
Rate mean value, can respectively obtain face alignment combined chance value V isMean value, maximum value or minimum value,
Formula difference is as follows:
Wherein,For the face alignment mathematical expectation of probability of RGB triple channel decision trees;L1、L2、L3For RGB tri-
The decision tree number of channel training.
Step 3.2:By targetedly training the respective face alignment decision tree classifier of triple channel and being cascaded into one
Complete strong classifier, it is comprehensive that the face alignment mathematical expectation of probability that comprehensive triple channel face alignment strong classifier exports obtains face alignment
Close probability value V, setting face alignment similarity threshold τ.Comprehensive face alignment combined chance value V and face alignment similarity threshold
τ, the judgement of face alignment similarity are as follows:
Claims (4)
1. the multiple dimensioned colour image human face comparison method based on cascade structure, including steps are as follows:
Step 1:For coloured image RGB triple channels, the multiple dimensioned facial image in each channel normalizes extraction respectively
Pixel difference feature;
Step 2:Using the normalization pixel difference feature of extraction, face alignment decision tree, the corresponding threeway Taoist of training are built respectively
Face compares decision tree classifier;
Step 3:Based on Softcascade cascade structures, joint triple channel face alignment decision tree output face alignment probability is equal
Value, obtains face alignment combined chance value.Complete multiple dimensioned, colour image human face pair face alignment probability output.
2. the multiple dimensioned colour image human face comparison method according to claim 1 based on cascade structure, it is characterised in that:
It is as follows in above-mentioned steps 1:
Step 1.1:Training data sample set D includes that the P positive sample pair with label information of different scale size is born with N number of
Sample pair, for training dataset RGB Three Channel Color image patterns pair, respectively to the multiple dimensioned facial image in each channel
Any two points extraction normalization pixel difference feature of sample pair, realizes that across scale feature extraction, formula are as follows:
Wherein, AM×M、BN× N is the RGB color facial image for being respectively M × M sizes and N × N sizes;
For R, G, B value that A image RGB respective channel coordinates are x points;
For R, G, B value that B image RGB corresponding channel coordinates are y points;Normalization pixel difference is characterized as m2×
n2The feature vector of dimension;F(B)、F(B)、F(B)For the normalization pixel difference feature extracted for the RGB of image A and image B.
3. the multiple dimensioned colour image human face comparison method according to claim 1 based on cascade structure, it is characterised in that:
It is as follows in above-mentioned steps 2:
Step 2.1:The training dataset RGB triple channels normalization pixel difference feature f (x, y) obtained by step 1, is directed to respectively
RGB triple channels train face alignment decision tree classifier.Assuming that given input normalizes pixel difference feature f, decision leaf
Node exports face alignment similarity probability S.According to split vertexes function, that is, maximum value purity function gamma, pixel difference will be normalized
The split vertexes of feature construction decision tree, input feature vector f train dual threshold α by the method for exhaustion1, α2, according to dual threshold α1, α2Really
The left and right route for determining each split vertexes of decision tree structure is distinguished, and it is as follows to distinguish mode:
Maximum value purity function gamma enables training sample data collection distinguish positive negative sample to the greatest extent, and formula is as follows:
Wherein, uP、vNTo be split off the number of positive sample P and negative sample N that node is distinguished in decision tree;
It is assumed that in the decision tree for N number of node that depth is κ, wherein there is L leaf node.Each leafy node lL can
Indicate the probability distribution s of image similarityl∈S.Each node of decision tree will be each under the constraint of maximum value purity function gamma
A leaf node output image similarity probability value S, it is assumed that label is distributed as σ on corresponding leaf nodel, decision tree pair one
The face alignment similarity probability value of a training sample pair can be expressed as:
Wherein,
∑l∈Lσl(f | γ, h)=1;
Step 2.2:Mechanism is cascaded based on Softcascade, trained decision tree is cascaded into face alignment strong classifier, and defeated
Go out decision tree leaf node face alignment mathematical expectation of probability, formula is as follows:
Wherein, C1、C2、C3For the decision tree number of RGB triple channels training;For RGB channel cascade classifier
The face alignment probability value of i-th of tree output;For RGB triple channel decision tree face alignment mathematical expectation of probability.
4. the multiple dimensioned colour image human face comparison method according to claim 1 based on cascade structure, it is characterised in that:
It is as follows in above-mentioned steps 3:
Step 3.1:It is equal according to the Softcascade cascade classifier decision tree leaf node face alignment probability of RGB triple channels
Value, can respectively obtain face alignment combined chance value V isMean value, maximum value or minimum value, formula
It is as follows respectively:
Wherein,For the face alignment mathematical expectation of probability of RGB triple channel decision trees;L1、L2、L3For RGB triple channels
Trained decision tree number.
Step 3.2:By targetedly training the respective face alignment decision tree classifier of triple channel and being cascaded into one completely
Strong classifier, it is general that the face alignment mathematical expectation of probability of comprehensive triple channel face alignment strong classifier output obtains face alignment synthesis
Rate value V, setting face alignment similarity threshold τ.Comprehensive face alignment combined chance value V and face alignment similarity threshold τ,
The judgement of face alignment similarity is as follows:
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810497963.5A CN108764106B (en) | 2018-05-22 | 2018-05-22 | Multi-scale color image face comparison method based on cascade structure |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810497963.5A CN108764106B (en) | 2018-05-22 | 2018-05-22 | Multi-scale color image face comparison method based on cascade structure |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108764106A true CN108764106A (en) | 2018-11-06 |
CN108764106B CN108764106B (en) | 2021-12-21 |
Family
ID=64004809
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810497963.5A Active CN108764106B (en) | 2018-05-22 | 2018-05-22 | Multi-scale color image face comparison method based on cascade structure |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108764106B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20210374386A1 (en) * | 2017-03-24 | 2021-12-02 | Stripe, Inc. | Entity recognition from an image |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20080240504A1 (en) * | 2007-03-29 | 2008-10-02 | Hewlett-Packard Development Company, L.P. | Integrating Object Detectors |
CN104966085A (en) * | 2015-06-16 | 2015-10-07 | 北京师范大学 | Remote sensing image region-of-interest detection method based on multi-significant-feature fusion |
CN106599918A (en) * | 2016-12-13 | 2017-04-26 | 开易(深圳)科技有限公司 | Vehicle tracking method and system |
CN107506702A (en) * | 2017-08-08 | 2017-12-22 | 江西高创保安服务技术有限公司 | Human face recognition model training and test system and method based on multi-angle |
-
2018
- 2018-05-22 CN CN201810497963.5A patent/CN108764106B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20080240504A1 (en) * | 2007-03-29 | 2008-10-02 | Hewlett-Packard Development Company, L.P. | Integrating Object Detectors |
CN104966085A (en) * | 2015-06-16 | 2015-10-07 | 北京师范大学 | Remote sensing image region-of-interest detection method based on multi-significant-feature fusion |
CN106599918A (en) * | 2016-12-13 | 2017-04-26 | 开易(深圳)科技有限公司 | Vehicle tracking method and system |
CN107506702A (en) * | 2017-08-08 | 2017-12-22 | 江西高创保安服务技术有限公司 | Human face recognition model training and test system and method based on multi-angle |
Non-Patent Citations (1)
Title |
---|
MENGLIN JIANG ET AL.: "Video Copy Detection Using a Soft Cascade of Multimodal Features", 《2012 IEEE INTERNATIONAL CONFERENCE ON MULTIMEDIA AND EXPO》 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20210374386A1 (en) * | 2017-03-24 | 2021-12-02 | Stripe, Inc. | Entity recognition from an image |
US11727053B2 (en) * | 2017-03-24 | 2023-08-15 | Stripe, Inc. | Entity recognition from an image |
Also Published As
Publication number | Publication date |
---|---|
CN108764106B (en) | 2021-12-21 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2020114118A1 (en) | Facial attribute identification method and device, storage medium and processor | |
US8682029B2 (en) | Rule-based segmentation for objects with frontal view in color images | |
Cao et al. | Gender recognition from body | |
US8385609B2 (en) | Image segmentation | |
CN108268859A (en) | A kind of facial expression recognizing method based on deep learning | |
Islam et al. | Performance of SVM, CNN, and ANN with BoW, HOG, and image pixels in face recognition | |
CN105069400A (en) | Face image gender recognition system based on stack type sparse self-coding | |
CN109657612B (en) | Quality sorting system based on facial image features and application method thereof | |
CN108229458A (en) | A kind of intelligent flame recognition methods based on motion detection and multi-feature extraction | |
CN106485222A (en) | A kind of method for detecting human face being layered based on the colour of skin | |
CN110348322A (en) | Human face in-vivo detection method and equipment based on multi-feature fusion | |
Chen et al. | Skin color modeling for face detection and segmentation: a review and a new approach | |
Indi et al. | Detection of malpractice in e-exams by head pose and gaze estimation | |
CN103440510A (en) | Method for positioning characteristic points in facial image | |
Yao et al. | Face detection and location based on skin chrominance and lip chrominance transformation from color images | |
CN108509920A (en) | The face identification method of the multichannel combined feature selecting study of more patch based on CNN | |
CN106845328A (en) | A kind of Intelligent human-face recognition methods and system based on dual camera | |
CN109190561A (en) | Face identification method and system in a kind of video playing | |
CN108985200A (en) | A kind of In vivo detection algorithm of the non-formula based on terminal device | |
CN115527269B (en) | Intelligent human body posture image recognition method and system | |
Shah et al. | Multi-camera handoff for person re-identification | |
CN108764106A (en) | Multiple dimensioned colour image human face comparison method based on cascade structure | |
CN108446639A (en) | Low-power consumption augmented reality equipment | |
KR102617756B1 (en) | Apparatus and Method for Tracking Missing Person based on Attribute | |
CN106548130A (en) | A kind of video image is extracted and recognition methods and system |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |