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 PDF

Info

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
face comparison
decision tree
channel
rgb
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
Application number
CN201810497963.5A
Other languages
Chinese (zh)
Other versions
CN108764106B (en
Inventor
章东平
井长兴
倪佩青
陈奇
杨力
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.)
China Jiliang University
Original Assignee
China Jiliang 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 China Jiliang University filed Critical China Jiliang University
Priority to CN201810497963.5A priority Critical patent/CN108764106B/en
Publication of CN108764106A publication Critical patent/CN108764106A/en
Application granted granted Critical
Publication of CN108764106B publication Critical patent/CN108764106B/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
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Health & Medical Sciences (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Probability & Statistics with Applications (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • General Health & Medical Sciences (AREA)
  • Oral & Maxillofacial Surgery (AREA)
  • Human Computer Interaction (AREA)
  • Multimedia (AREA)
  • Image Analysis (AREA)
  • Image Processing (AREA)
  • Collating Specific Patterns (AREA)

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

multi-scale color image face comparison method based on cascade structure
Technical Field
The invention belongs to the field of computer vision, face comparison and recognition, and particularly relates to a face comparison method aiming at input multi-scale color images.
Background
The biological characteristics are the attributes of the human body, and have strong personal difference and self stability. In recent years, as a face comparison algorithm in a biometric technology, the face comparison algorithm has characteristics of real-time performance, immediacy, convenience and the like, and is increasingly researched by experts and scholars in the field of computer vision. With the development of machine learning algorithms and the wide application of face comparison, the face comparison problem is gradually developing as a research hotspot. Meanwhile, the method plays an increasingly important role in the fields of automatic monitoring, alarming, identity verification, financial activities and the like.
However, the current face comparison is limited to the processing of the gray level image in the aspect of algorithm research, and the research of the face comparison of the color image is neglected. Meanwhile, the comparison of images with different sizes and multiple scales cannot be realized in the face comparison process, the uniform size of the image pair needs to be redefined before the face comparison, image information is lost in the zooming and amplifying process, distortion and the like are caused, computing resources are wasted, the information is not sufficiently used, and the face comparison recognition rate is reduced.
Disclosure of Invention
Aiming at (1) the gray level image research and neglecting the important function of color images in human face comparison in the prior human face comparison technology; (2) in the face of different input size pictures, uniform size needs to be defined again, multi-scale face comparison cannot be achieved, and the like, the invention provides a multi-scale color image face comparison method.
The method for comparing the face of the multi-scale color image based on the cascade structure is characterized by comprising the following steps of:
step 1: aiming at three channels of RGB color images, respectively extracting normalized pixel difference characteristics from the multi-scale face image pair of each channel;
step 2: respectively constructing face comparison decision trees by using the extracted normalized pixel difference characteristics, and training corresponding three-channel face comparison decision tree classifiers;
and step 3: and based on a Softcascade cascade structure, outputting a face comparison probability mean value by combining a three-channel face comparison decision tree to obtain a face comparison comprehensive probability value. And finishing the face comparison probability output of the multi-scale color image face pair.
Further, step 1 comprises the steps of:
step 1.1: the training data sample set D comprises P positive sample pairs and N negative sample pairs with label information in different scales, normalized pixel difference characteristics are respectively extracted from any two points of the multi-scale face image sample pair of each channel aiming at the RGB three-channel color image sample pair of the training data set, and cross-scale feature extraction is realized, wherein the formula is as follows:
wherein A isM×M、BNXn is the RGB color face image of M × M size and N × N size, respectively;r, G, B values of x points of corresponding channel coordinates of the image RGB A;r, G, B values of y points corresponding to channel coordinates of the RGB images B; normalized pixel difference feature is m2×n2A feature vector of a dimension; f(B)、F(B)、F(B)Normalized pixel difference features extracted for RGB for image a and image B.
Further, step 2 comprises the steps of:
step 2.1, training a face comparison decision tree classifier respectively for RGB three channels by using the training data set RGB three-channel normalized pixel difference characteristics f (x, y) obtained in the step 1, assuming the given input normalized pixel difference characteristics f, outputting face comparison similarity probability S by leaf nodes of a decision tree, constructing the split nodes of the decision tree by using the normalized pixel difference characteristics according to a split node function, namely a maximum purity function gamma, and training a double-threshold α by using the input characteristics f through an exhaustion method1，α2according to a double threshold value alpha1，α2Determining left and right of each split node of a decision tree structure
The maximum purity function γ is used to distinguish positive and negative samples of the training sample data set to the maximum extent, and the formula is as follows:
wherein u isP、vNThe number of positive samples P and negative samples N distinguished by a split node in the decision tree.
Assume that in a decision tree of N nodes with a depth of k, there are L leaf nodes. At each leafThe nodes lL may all represent a probability distribution s of image similaritylE.g. S. Outputting an image similarity probability value S by each leaf node under the constraint of a maximum purity function gamma by each node of the decision tree, and assuming that labels are distributed to sigma on the corresponding leaf nodeslThe face comparison similarity probability value of the decision tree for a training sample pair can be expressed as:
wherein,
step 2.2: based on a Softcascade cascade mechanism, the trained decision tree is cascaded into a human face comparison strong classifier, and a decision tree leaf node human face comparison probability mean value is output, wherein the formula is as follows:
wherein, C1、C2、C3The number of decision trees for RGB three-channel training;comparing the probability value of the face output by the ith tree of the RGB channel cascade classifier;and comparing the probability mean value for the RGB three-channel decision tree face.
Further, the step 3 includes the steps of:
step 3.1: according to the RGB three-channel Softcascade classifier decision tree leaf node face comparison probability mean value, the face comparison comprehensive probability value V can be respectively obtainedThe average, maximum or minimum value of (a), respectively, of the following formulae:
wherein,comparing the probability mean value of the face of the RGB three-channel decision tree; l is1、L2、L3And the number of decision trees for RGB three-channel training.
Step 3.2: and respectively training three channels of face comparison decision tree classifiers in a targeted manner to be cascaded into a complete strong classifier, synthesizing face comparison probability mean values output by the three channels of face comparison strong classifiers to obtain a face comparison comprehensive probability value V, and setting a face comparison similarity threshold value tau. And (3) integrating the face comparison comprehensive probability value V and the face comparison similarity threshold value tau, wherein the face comparison similarity is determined as follows:
drawings
FIG. 1 is a general technical flow diagram;
fig. 2 is a diagram of a face comparison decision tree structure of a Softcascade structure.
Detailed Description
The invention will be further explained with reference to the drawings.
As shown in fig. 1, the method for comparing a multi-scale color image face based on a cascade structure of the present invention includes the following steps:
step 1: and (3) extracting normalized pixel difference characteristics of the multi-scale face image pair of each channel aiming at three channels of the color image RGB. The substeps are as follows:
step 1.1: the training data sample set D comprises P positive sample pairs and N negative sample pairs with label information in different scales, normalized pixel difference characteristics are respectively extracted from any two points of the multi-scale face image sample pair of each channel aiming at the RGB three-channel color image sample pair of the training data set, and cross-scale feature extraction is realized, wherein the formula is as follows:
wherein A isM×M、BNXn is the RGB color face image of M × M size and N × N size, respectively;r, G, B values of x points of corresponding channel coordinates of the image RGB A;r, G, B values of y points corresponding to channel coordinates of the RGB images B; normalized pixel difference feature is m2×n2A feature vector of a dimension; f(B)、F(B)、F(B)Normalized pixel difference features extracted for RGB for image a and image B.
Step 2: and respectively constructing a face comparison decision tree by using the extracted normalized pixel difference characteristics, and training a corresponding three-channel face comparison decision tree classifier. The substeps are as follows:
step 2.1, training a face comparison decision tree classifier respectively for RGB three channels by using the training data set RGB three-channel normalized pixel difference characteristics f (x, y) obtained in the step 1, assuming the given input normalized pixel difference characteristics f, outputting face comparison similarity probability S by leaf nodes of a decision tree, constructing the split nodes of the decision tree by using the normalized pixel difference characteristics according to a split node function, namely a maximum purity function gamma, and training a double-threshold α by using the input characteristics f through an exhaustion method1，α2according to a double threshold value alpha1，α2Determining the left and right route distinction of each split node of the decision tree structure, wherein the distinction mode is as follows:
the maximum purity function γ is used to distinguish positive and negative samples of the training sample data set to the maximum extent, and the formula is as follows:
wherein u isP、vNThe number of positive samples P and negative samples N distinguished by a split node in the decision tree.
Assume that in a decision tree of N nodes with a depth of k, there are L leaf nodes. Probability distribution s that can represent image similarity at each leaf node L ∈ LlE.g. S. Outputting an image similarity probability value S by each leaf node under the constraint of a maximum purity function gamma by each node of the decision tree, and assuming that labels are distributed to sigma on the corresponding leaf nodeslThe face comparison similarity probability value of the decision tree for a training sample pair can be expressed as:
wherein,
step 2.2: based on a Softcascade cascade mechanism, the trained decision tree is cascaded into a human face comparison strong classifier, and a decision tree leaf node human face comparison probability mean value is output, wherein the formula is as follows:
wherein, C1、C2、C3The number of decision trees for RGB three-channel training;comparing the probability value of the face output by the ith tree of the RGB channel cascade classifier;and comparing the probability mean value for the RGB three-channel decision tree face.
And step 3: and based on a Softcascade cascade structure, outputting a face comparison probability mean value by combining a three-channel face comparison decision tree to obtain a face comparison comprehensive probability value. And finishing the face comparison probability output of the multi-scale color image face pair. The substeps are as follows:
step 3.1: according to the RGB three-channel Softcascade classifier decision tree leaf node face comparison probability mean value, the face comparison comprehensive probability value V can be respectively obtainedThe average, maximum or minimum value of (a), respectively, of the following formulae:
wherein,comparing the probability mean value of the face of the RGB three-channel decision tree; l is1、L2、L3And the number of decision trees for RGB three-channel training.
Step 3.2: and respectively training three channels of face comparison decision tree classifiers in a targeted manner to be cascaded into a complete strong classifier, synthesizing face comparison probability mean values output by the three channels of face comparison strong classifiers to obtain a face comparison comprehensive probability value V, and setting a face comparison similarity threshold value tau. And (3) integrating the face comparison comprehensive probability value V and the face comparison similarity threshold value tau, wherein the face comparison similarity is determined as follows:

Claims (4)

1. The multi-scale color image face comparison method based on the cascade structure comprises the following steps:
step 1: aiming at three channels of RGB color images, respectively extracting normalized pixel difference characteristics from the multi-scale face image pair of each channel;
step 2: respectively constructing face comparison decision trees by using the extracted normalized pixel difference characteristics, and training corresponding three-channel face comparison decision tree classifiers;
and step 3: and based on a Softcascade cascade structure, outputting a face comparison probability mean value by combining a three-channel face comparison decision tree to obtain a face comparison comprehensive probability value. And finishing the face comparison probability output of the multi-scale color image face pair.
2. The method for comparing multi-scale color image faces based on cascade structure as claimed in claim 1, wherein: the specific steps in the step 1 are as follows:
step 1.1: the training data sample set D comprises P positive sample pairs and N negative sample pairs with label information in different scales, normalized pixel difference characteristics are respectively extracted from any two points of the multi-scale face image sample pair of each channel aiming at the RGB three-channel color image sample pair of the training data set, and cross-scale feature extraction is realized, wherein the formula is as follows:
wherein A isM×M、BNXn is the RGB color face image of M × M size and N × N size, respectively;
r, G, B values of x points of corresponding channel coordinates of the image RGB A;
r, G, B values of y points corresponding to channel coordinates of the RGB images B; normalized pixel difference feature is m2×n2A feature vector of a dimension; f(B)、F(B)、F(B)Normalization for RGB extraction for image A and image BThe pixel difference feature is quantized.
3. The method for comparing multi-scale color image faces based on cascade structure as claimed in claim 1, wherein: the step 2 comprises the following specific steps:
step 2.1, training a face comparison decision tree classifier respectively for RGB three channels by using the training data set RGB three-channel normalized pixel difference characteristics f (x, y) obtained in the step 1, assuming the given input normalized pixel difference characteristics f, outputting face comparison similarity probability S by leaf nodes of a decision tree, constructing the split nodes of the decision tree by using the normalized pixel difference characteristics according to a split node function, namely a maximum purity function gamma, and training a double-threshold α by using the input characteristics f through an exhaustion method1，α2according to a double threshold value alpha1，α2Determining the left and right route distinction of each split node of the decision tree structure, wherein the distinction mode is as follows:
the maximum purity function γ is used to distinguish positive and negative samples of the training sample data set to the maximum extent, and the formula is as follows:
wherein u isP、vNThe number of positive samples P and negative samples N which are distinguished by the split node in the decision tree;
assume that in a decision tree of N nodes with a depth of k, there are L leaf nodes. The probability distribution s that can represent the similarity of images at each leaf node lLlE.g. S. Outputting an image similarity probability value S by each leaf node under the constraint of a maximum purity function gamma by each node of the decision tree, and assuming that labels are distributed to sigma on the corresponding leaf nodeslThe face comparison similarity probability value of the decision tree for a training sample pair can be expressed as:
wherein,
∑l∈Lσl(f|γ,h)＝1；
step 2.2: based on a Softcascade cascade mechanism, the trained decision tree is cascaded into a human face comparison strong classifier, and a decision tree leaf node human face comparison probability mean value is output, wherein the formula is as follows:
wherein, C1、C2、C3The number of decision trees for RGB three-channel training;comparing the probability value of the face output by the ith tree of the RGB channel cascade classifier;and comparing the probability mean value for the RGB three-channel decision tree face.
4. The method for comparing multi-scale color image faces based on cascade structure as claimed in claim 1, wherein: the specific steps in the step 3 are as follows:
step 3.1: according to the RGB three-channel Softcascade classifier decision tree leaf node face comparison probability mean value, the face comparison comprehensive probability value V can be respectively obtainedThe average, maximum or minimum value of (a), respectively, of the following formulae:
wherein,comparing the probability mean value of the face of the RGB three-channel decision tree; l is1、L2、L3And the number of decision trees for RGB three-channel training.
Step 3.2: and respectively training three channels of face comparison decision tree classifiers in a targeted manner to be cascaded into a complete strong classifier, synthesizing face comparison probability mean values output by the three channels of face comparison strong classifiers to obtain a face comparison comprehensive probability value V, and setting a face comparison similarity threshold value tau. And (3) integrating the face comparison comprehensive probability value V and the face comparison similarity threshold value tau, wherein the face comparison similarity is determined as follows:
CN201810497963.5A 2018-05-22 2018-05-22 Multi-scale color image face comparison method based on cascade structure Active CN108764106B (en)

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)

* Cited by examiner, † Cited by third party
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)

* Cited by examiner, † Cited by third party
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

Patent Citations (4)

* Cited by examiner, † Cited by third party
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)

* Cited by examiner, † Cited by third party
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)

* Cited by examiner, † Cited by third party
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
Cheng et al. Exploiting effective facial patches for robust gender recognition
CN107330397B (en) Pedestrian re-identification method based on large-interval relative distance measurement learning
WO2020114118A1 (en) Facial attribute identification method and device, storage medium and processor
Wu et al. Face detection in color images using AdaBoost algorithm based on skin color information
CN106778796B (en) Human body action recognition method and system based on hybrid cooperative training
Islam et al. Performance of SVM, CNN, and ANN with BoW, HOG, and image pixels in face recognition
US8023701B2 (en) Method, apparatus, and program for human figure region extraction
CN106156777B (en) Text picture detection method and device
US20220292394A1 (en) Multi-scale deep supervision based reverse attention model
CN107316059B (en) Learner gesture recognition method
US10210424B2 (en) Method and system for preprocessing images
US20100111375A1 (en) Method for Determining Atributes of Faces in Images
CN107045621A (en) Facial expression recognizing method based on LBP and LDA
CN107944398A (en) Based on depth characteristic association list diagram image set face identification method, device and medium
CN110414431B (en) Face recognition method and system based on elastic context relation loss function
Sahbi et al. Coarse to fine face detection based on skin color adaption
Gaston et al. Matching larger image areas for unconstrained face identification
CN108764106B (en) Multi-scale color image face comparison method based on cascade structure
CN106709442B (en) Face recognition method
Adeyanju et al. Development of an american sign language recognition system using canny edge and histogram of oriented gradient
Lee et al. Local and global feature extraction for face recognition
Mandal et al. Deep residual network with subclass discriminant analysis for crowd behavior recognition
Hanselmann et al. Deep fisher faces
Zeng et al. Local discriminant training and global optimization for convolutional neural network based handwritten Chinese character recognition
Vivekanandam et al. Face recognition from video frames using hidden markov model classification model based on modified random feature extraction

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