CN107066943A - A kind of method for detecting human face and device - Google Patents
A kind of method for detecting human face and device Download PDFInfo
- Publication number
- CN107066943A CN107066943A CN201710127367.3A CN201710127367A CN107066943A CN 107066943 A CN107066943 A CN 107066943A CN 201710127367 A CN201710127367 A CN 201710127367A CN 107066943 A CN107066943 A CN 107066943A
- Authority
- CN
- China
- Prior art keywords
- face
- feature
- candidate
- embedded
- masked
- 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
- G06V40/161—Detection; Localisation; Normalisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Collating Specific Patterns (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of method for detecting human face and device.The present invention method be：1) candidate face is detected from pending image, and extracts the candidate feature of the candidate face；2) each candidate feature is subjected to projective transformation in the conventional external feature space or proximate exterior feature space built in advance, obtains corresponding traditional or be approximately embedded in feature；Wherein, the proximate exterior feature space is to select the dictionary that representative feature is constituted from reference to face characteristics dictionary and non-face characteristics dictionary；3) the embedded feature is verified, determines whether the corresponding candidate face of the embedded feature is face.The human face detection device of the present invention includes candidate block, embedded module and authentication module.The present invention can obtain the higher Face datection performance of precision；To having under circumstance of occlusion, also possesses good Face datection ability.
Description
Technical field
The invention belongs to computer vision and deep learning field, more particularly to a kind of face inspection being directed under obstruction conditions
Survey method and device.
Background technology
Human face detection tech can be applied to camera auto-focusing, man-machine interaction, photo management, city safety monitoring, intelligence
The numerous areas such as driving.Currently, Face datection is in the practical application under the conditions of open environment, due to the generally existing blocked
(as in the case of the crowd is dense), Face datection performance is by serious challenge, thus Face datection performance under obstruction conditions asks
Topic has to be solved.In addition, the Face datection studied under masked obstruction conditions has important practical significance, such as：Video monitoring
In be used to find the regularity of distribution progress weather conditions prediction etc. of a suspect so as to provide warning, by detecting masked face.
Traditional method for detecting human face meets with serious hydraulic performance decline under occlusion, and reason is in detection process, and be blocked portion
The face clue divided is invalid, so as to cause inevitably to introduce noise in characteristic extraction procedure.In a word, it is imperfect and
Inaccurate feature makes the problem of masked Face datection blocked turns into a great challenge.
In recent years, certain methods are also studied in this field, prior art is first to detect face candidate, Ran Houzai
To face candidate class validation.One of which method is examined by training multiple neutral nets to obtain the response of the multiple parts of face
Survey face candidate, then the new neutral net of retraining one carry out face candidate class validation (referring to：S.Yang,
P.Luo,C.C.Loy,and X.Tang.From facial parts responses to face detection:A deep
learning approach.In:IEEE ICCV,2015).Another method then relatively calculates damage by selected section feature
Lose so as to confirm face candidate (referring to：M.Opitz,G.Waltner,G.Poier,H.Possegger,and
H.Bischo.Grid loss:Detecting occluded faces.In ECCV, 2016), this method can preferably be located in
Manage the Face datection problem of partial occlusion situation.The above method is alleviated to a certain extent seriously blocks (such as masked to block) feelings
Face datection problem under condition, but still fail to be fully solved.When face component is blocked, examined by multiple unit responses
The method for detecting human face of face candidate is surveyed, noise or mistake can be introduced in the part of occlusion area, so as to cause face point
Class confirms mistake；When serious shielding, counting loss is compared by selected section feature and confirms that the face of face candidate is examined
Survey method, the loss error that calculating is obtained is larger, so as to cause Face datection to fail.
The content of the invention
To overcome the deficiencies in the prior art, the invention provides a kind of method for detecting human face and device, this method passes through volume
Product neutral net detection candidate face and extraction higher-dimension depth characteristic (i.e. candidate feature), then by being locally linear embedding into progress
Projection Character is not exclusively and inaccurate to eliminate the masked feature brought of blocking, then using multitask convolutional neural networks (i.e.
CNN-V candidate face) is verified, so as to obtain more accurate Face datection performance.Meanwhile, the invention also provides a kind of approximate
The building method in surface space, most like reference face and the maximum ginseng of difference by being found from outside database
Examine non-face, carry out proximate exterior feature space construction, candidate feature is carried out using proximate exterior feature space to be embedded in conversion,
So as to correct candidate feature.The present invention is achieved through the following technical solutions.
A kind of method for detecting human face of the present invention, its step includes：
1) candidate face detection is carried out to image to be detected, obtains candidate face image；
2) candidate feature extraction is carried out to the candidate face image, obtains candidate feature；
3) candidate feature is carried out being embedded in conversion, obtains the embedded feature of tradition or be approximately embedded in feature, the insertion
Feature can recover the noise that face clue and occlusion removal are brought；
4) the embedded feature to the embedded feature of the tradition or approximately, is verified with regression algorithm by classification, is examined
Survey result.
Further, candidate feature carries out being embedded in after conversion by the surface built in advance a space, obtains
The embedded feature of tradition is approximately embedded in feature；Surface space is conventional external feature space or proximate exterior feature space.
Further, what embedded conversion use was traditional is locally linear embedding into method or is quickly approximately locally linear embedding into method
Realize；Traditional method that is locally linear embedding into carries out being embedded in change using conventional external feature space to the candidate feature with noise
Change, obtain the embedded feature of tradition；It is quick to be approximately locally linear embedding into using proximate exterior feature space to the candidate spy with noise
Levy progress and be embedded in conversion, obtain approximately being embedded in feature.
Further, quickly approximately it is locally linear embedding into the building method of proximate exterior feature space in method, including with
Lower step：
A) candidate face detection is carried out to the reference face data set marked and candidate feature is extracted, judge candidate feature
Belong to face characteristic or non-face feature, these candidate features are stored in reference to face characteristics dictionary and with reference to non-face respectively
Characteristics dictionary；
B) candidate face detection is carried out to the masked face data set marked and candidate feature is extracted, judge candidate feature
Belong to masked face characteristic or masked non-face feature, these candidate features are stored in masked face characteristics dictionary and illiteracy respectively
The non-face characteristics dictionary in face；
C) select representative to represent above-mentioned masked face tagged word from above-mentioned reference face characteristics dictionary
The reference face characteristics dictionary of allusion quotation；
D) from above-mentioned with reference to selecting in non-face characteristics dictionary representative to represent above-mentioned masked non-face spy
Levy the non-face characteristics dictionary of reference of dictionary；
E) above-mentioned representative reference face characteristics dictionary and the representative non-face tagged word of reference are merged
Allusion quotation, obtains proximate exterior feature space.
Further, in step a), the face for calculating the corresponding candidate face position of the candidate feature Yu having marked is passed through
Degree of overlapping between position determines, its degree of overlapping with handing over and than measuring, wherein, hand over simultaneously more special than judging candidate more than 0.7
Levy as the feature with reference to face, friendship and refer to non-face feature than judging candidate feature less than 0.3.
Further, in step b), the face for calculating the corresponding candidate face position of the candidate feature Yu having marked is passed through
Degree of overlapping between position determines, its degree of overlapping with handing over and than measuring, wherein, hand over simultaneously more special than judging candidate more than 0.6
Levy as the feature of masked face, friendship and than judging candidate feature for masked non-face feature less than 0.4.
Further, representative reference is selected from reference face characteristics dictionary using greedy algorithm in step c)
Face characteristic dictionary；The greedy algorithm refers to calculate with reference to each loss with reference to face characteristic in face characteristics dictionary, obtains
To by the reference face feature list for losing ascending ascending order arrangement, the reference face characteristic of the list foremost is taken to represent
Masked face characteristic；Wherein described loss refers to each reference face characteristic and the arest neighbors feature of masked face characteristics dictionary
The difference of the distance of distance and the arest neighbors feature of each reference face characteristic and masked non-face characteristics dictionary.
Further, representative ginseng is selected from the non-face characteristics dictionary of reference using greedy algorithm in step d)
Examine non-face characteristics dictionary；The greedy algorithm refers to calculate in the non-face characteristics dictionary of reference each with reference to non-face feature
Loss, obtains, by the non-face feature list of reference for losing ascending ascending order arrangement, taking the reference of the list foremost inhuman
Face feature represents masked non-face feature；Wherein described loss refers to the non-face feature of each reference and masked non-face feature
The distance of the arest neighbors feature of dictionary and each with reference to non-face feature and masked face characteristics dictionary arest neighbors feature away from
It is poor from it.
The invention further relates to a kind of human face detection device, including candidate block, embedded module and authentication module.Candidate block
For carrying out candidate face detection to image to be detected and extracting candidate feature；Embedded module is used to be embedded in candidate feature
Conversion, obtains the embedded feature of tradition or is approximately embedded in feature, and embedded feature can recover face clue and occlusion removal brings
Noise；Authentication module is used for the embedded feature of tradition or is approximately embedded in feature, is verified by classification with regression algorithm, with
To last testing result.Candidate block obtains multiple candidate features, is then built in advance by one in embedded module
Surface space be embedded in after conversion, obtain the embedded feature of tradition or be approximately embedded in feature；Surface space is biography
System surface space or proximate exterior feature space；Embedded conversion is locally linear embedding into method or quick approximate using traditional
It is locally linear embedding into method realization.
The beneficial effects of the present invention are：
For the Face datection problem under obstruction conditions, the Face datection problem under especially serious masked obstruction conditions,
The detection method and device of the present invention have relatively good performance；To the face in the case of unobstructed, face inspection of the invention
Survey method and device and also possess good disposal ability.
Brief description of the drawings
Fig. 1 is a kind of flow chart of method for detecting human face of the invention；
Fig. 2 is apparatus of the present invention candidate block schematic flow sheet；
Fig. 3 is that apparatus of the present invention are embedded in block process schematic diagram；
Fig. 4 is apparatus of the present invention authentication module schematic flow sheet；
Fig. 5 constructs schematic flow sheet for the proximate exterior feature space of the present invention.
Embodiment
To become apparent the such scheme and beneficial effect of the present invention, hereafter by embodiment, and accompanying drawing is coordinated to make
Describe in detail as follows.
The present invention provides a kind of method for detecting human face and device, and the device includes candidate block, embedded module and checking mould
Block；The flow chart of this method is as shown in figure 1, its step includes：
1) image is received.Under described image both can be facial image or serious masked obstruction conditions under obstruction conditions
Facial image or it is unobstructed in the case of facial image or image not comprising face.
2) candidate face is detected by candidate block and extracts the higher-dimension depth characteristic of candidate face, i.e. candidate feature.
In candidate block, candidate face detection is first carried out, then judges whether to detect candidate face, if do not detected
Then terminate to candidate face；Candidate feature extraction is carried out if candidate face is detected, candidate feature is obtained.
Fig. 2 is refer to, the candidate block is mainly comprising two convolutional neural networks：One is small convolutional neural networks
(being referred to as candidate's convolutional neural networks, abbreviation CNN-P), the network is used to realize that candidate face is detected；Another big convolution
Neutral net (is referred to as feature convolutional neural networks, abbreviation CNN-F), for realizing that candidate feature is extracted.First, the figure received
As by candidate's convolutional neural networks, carrying out candidate face detection, then judging whether to detect candidate face, if do not detected
To candidate face, then terminate；If detecting candidate face, candidate face normalized is first carried out, then roll up by feature
Product neutral net carries out candidate feature extraction, obtains candidate feature.
3) candidate feature insertion is carried out by embedded module, obtain the feature after embedded conversion, i.e. tradition insertion feature or
Approximate embedded feature (being referred to as being embedded in feature).
Because masked block can cause face clue to lack and characteristic noise, result in a feature that imperfect and inaccurate.
For the problem, the insertion module in technical solution of the present invention, which is realized, to be recovered face clue from candidate feature and removes noise.
The advantage of embedded resume module be the insertion feature obtained can characterize well it is masked block face, so as to lift detection
Precision.
Fig. 3 is refer to, in embedded module, candidate feature is carried out by the surface built in advance a space
After embedded conversion, obtain the embedded feature of tradition or be approximately embedded in feature.The embedded conversion is main using LLE (Local
Linear Embedding) method realization.LLE is a kind of dimension reduction method for nonlinear data, the low-dimensional data after processing
Original topological relation can be kept, have been widely used for the visualization of the classification of view data and cluster, multidimensional data with
And the field such as bioinformatics.The present invention realizes embedded conversion using traditional LLE methods and quick approximate LLE methods.
4) by authentication module, carry out the embedded feature of tradition or be approximately embedded in signature verification, judge that each tradition is embedded special
Levy or be approximately embedded in whether the corresponding candidate face of feature belongs to real face, if the embedded feature of the tradition or approximate embedded special
Levy corresponding candidate face and belong to real face, then record face information；If the embedded feature of the tradition is approximately embedded in feature
Corresponding candidate face is not belonging to real face, then terminates.
Fig. 4 is refer to, authentication module (referred to as verifies convolutional Neural net by one four layers of full connection convolutional neural networks
Network, abbreviation CNN-V) composition, for carrying out signature verification, that is, differentiate that the embedded feature of the tradition or approximate embedded feature are corresponding
Whether candidate face belongs to real face and corrects corresponding candidate face position and yardstick.If being not belonging to real face,
Ignore the embedded feature of the tradition or the corresponding candidate face of approximate embedded feature；It is if belonging to real face, the tradition is embedding
Enter feature or the corresponding revised candidate face position of approximate embedded feature is added in testing result with yardstick.
The embedded feature of tradition or approximate embedded feature are classified and returned by authentication module, so as to determine candidate
Belong to real face or non-face, and face location is modified with yardstick, so as to obtain the higher Face datection of precision
Performance.
Therefore, a kind of method for detecting human face and device proposed by the present invention have combined candidate's convolution nerve net of candidate block
Network CNN-P, the feature convolutional neural networks CNN-F of candidate block, the checking convolutional neural networks for being embedded in module and authentication module
CNN-V, to reach the purpose of the present invention.
The used method of the embedded conversion of embedded module is detailed below.
1st, traditional LLE methods.
Fig. 3 is refer to, by traditional LLE methods, by the masked candidate feature x blockediIn the tradition constructed in advance
Projective transformation is carried out in surface space, obtains being embedded in feature vi, insertion feature viIt can effectively eliminate and be blocked due to masked
The imperfect and inaccurate problem of the feature brought, ability is blocked with anti-well.Wherein xiSubscript i be used for mark it is different
Candidate feature；viSubscript i be used for mark different insertion features.Embedded feature viReferred to as traditional embedded feature.
The conventional external feature space with reference to face characteristic and with reference to non-face feature by constituting, and it is expressed as tagged word
The form of allusion quotation, i.e. D=[D + ,D-], D here+It is to refer to face characteristics dictionary, D-It is to refer to non-face characteristics dictionary, as a rule
D+And D-Scale has up to a million.
The reference face characteristic and the non-face feature of reference, pass through to build and are realized with reference to candidate characteristic set.Specifically, it is right
The large-scale unobstructed reference face data set S markedn, carry out candidate face detection using candidate block and candidate feature carried
Take.Judge that candidate feature belongs to face characteristic or non-face feature, these candidate features are divided into reference to face characteristic and ginseng
Non-face feature is examined, deposit is with reference to face characteristics dictionary D respectively+With the non-face characteristics dictionary D of reference-.Wherein judge candidate feature
Belong to face characteristic or non-face feature, be the people by calculating the corresponding candidate face position of the candidate feature Yu having marked
Degree of overlapping between face position determines that its degree of overlapping hands over and measured than (Intersection-over-Union, IoU).
Handed in generally conventional method and ratio is judged as face more than 0.5, be judged as less than 0.5 non-face.With conventional method phase
Than handing over and than being judged as referring to face more than 0.7, handing over and than being judged as less than 0.3 with reference to non-face in the present invention so that
The reference face arrived with reference to non-face with having more preferable distinction, it is ensured that has preferably identification energy with reference to candidate feature
Power.
For each candidate feature x with noisei, all from D+And D-Middle chosen distance xiClosest feature set is constituted
The sub- dictionary D of featurei(DiSubscript i be used for mark the sub- dictionary of the corresponding feature of different candidate features), then utilize LLE algorithms
Projective transformation is carried out, it is the embedded feature v of tradition to obtain a new feature representationi, the solution formula of the process is as follows：
Meet vi≥0(1)
2nd, quick approximate LLE methods.
The present invention proposes a kind of quick approximate LLE methods, for each candidate feature x with noisei, using quick near
Projective transformation is carried out like LLE methods, an approximate embedded feature is obtainedThis method solution formula is as follows：
Meet
In above-mentioned formula (2),It is proximate exterior feature space, the space is from reference to face characteristics dictionary D+With it is inhuman
Face characteristics dictionary D-The dictionary of the representative feature composition of middle selection.To each candidate feature xiNo longer needing construction, its is right
The sub- dictionary D of feature answeredi, each candidate feature xiAll using fixed proximate exterior feature spaceProjective transformation is carried out, is obtained
Approximate embedded feature
The construction of proximate exterior feature space in quick approximate LLE methods is detailed below.
The building method of the proximate exterior feature space is by finding most like reference from outside database
Face or the maximum reference of difference are non-face, carry out proximate exterior feature space construction.
Fig. 5 is refer to, the figure is proximate exterior feature spaceThe flow chart of construction,It is from D+And D-Middle selection is most
Representational feature composition, it includes representative reference face characteristics dictionaryIt is inhuman with representative reference
Face characteristics dictionaryIt is expressed asProximate exterior feature space proposed by the present inventionBuilding method is specifically divided into
The following steps：
1) build with reference to face and refer to non-face characteristics dictionary：It is identical with above-mentioned traditional LLE methods, to mark
Good large-scale unobstructed reference face data set Sn, carry out candidate face detection using candidate block and candidate feature extracted.
Face characteristic or non-face feature are belonged to according to candidate feature, these candidate features are stored in reference to face characteristics dictionary respectively
D+With the non-face characteristics dictionary D of reference-.Judge that candidate feature belongs to face characteristic or non-face feature, be by calculating the time
The degree of overlapping between the corresponding candidate face position of feature and the face location marked is selected to determine, its degree of overlapping is handed over and compared
IoU is measured.Handed in generally conventional method and ratio is judged as face more than 0.5, be judged as less than 0.5 non-face.With
Conventional method is compared, and is handed over and than being judged as referring to face more than 0.7, is handed over and than being judged as reference less than 0.3 in the present invention
It is non-face, so as to get reference face with there is more preferable distinction with reference to non-face, it is ensured that have with reference to candidate feature
More preferable identification capability.
2) masked face and masked non-face characteristics dictionary are built：Similar above-mentioned steps 1), it is large-scale masked to what is marked
Human face data collection Sm, carry out candidate face detection using candidate block and candidate feature extracted.Belong to masked according to candidate feature
These candidate features are divided into masked face characteristics dictionary by face characteristic or masked non-face featureWith masked non-face spy
Levy dictionaryHand over and compare because the positioning precision of masked Face datection will be generally less than in unobstructed Face datection, the present invention
Masked face is judged as more than 0.6, is handed over and masked more non-face than being judged as less than 0.4, to select the masked of better quality
Face candidate feature.
3) representative reference face characteristics dictionary is selected From with reference to face characteristics dictionary D+Middle selection, is D+A subset be Representativeness show it when representing masked face with good sign ability simultaneously in generation
Table has separating capacity when masked non-face.So as to,Sparsely representing masked face characteristics dictionaryShi Yingyou minimums
Mistake, while sparsely representing masked non-face characteristics dictionaryThere should be the mistake of maximum.Therefore,By solving following public affairs
Formula (3) is obtained：
Meet
Above-mentioned formula (3) belongs to α in sparse coding processing, formula1And α2It is to utilize respectivelyRepresent some masked face
Feature x1With some masked non-face feature x2The sparse coefficient vector needed.It is 1 to only have an element in sparse coefficient vector,
Other elements are 0.Using the constraints of sparse coefficient vector, sparse coding processing be equivalent to fromMiddle searching arest neighbors.By
InIn each feature come from reference to face characteristics dictionary D+, the optimization problem of formula (3) and classical sparse coding side
Formula is different, is difficult to solve with classical optimized algorithm.So, the present invention proposes a kind of greedy method effectively from reference
Face characteristic dictionary D+It is middle to buildIn the greedy method of proposition, the present invention is calculated with reference to face characteristics dictionary D first+In
Each refer to face characteristicLossThe loss is expressed asWith masked face characteristics dictionaryArest neighbors feature
Distance andWith masked non-face characteristics dictionaryArest neighbors feature distance difference, its pass through below equation (4) realize：
Meet
In above-mentioned formula (4), ρ1And ρ2It is two coefficients of balance, for the distance between balance characteristics, leads in actual treatment
1 often is taken with speed-up computation, each with reference to face characteristicIt is rarely used for representingIn masked face characteristic andIn illiteracy
The non-face feature in face.Pass through counting lossObtain according to the reference face for losing ascending ascending order arrangement
The reference face characteristic that foremost is come in feature list, list is most strong in the ability for representing masked face characteristic aspect, and generation
The ability of the masked non-face feature of table is most weak.In this way, can be by way of iteration, constantly by M before list
A feature pool P is added to reference to face characteristic+In, construct finalIt is preferred that, M is more than or equal to 1 and is less than or equal to
50.Specifically, make initial characteristicses pond for sky i.e.Then walk and use in tCome M candidate before selecting, obtainThen,In feature be used for updateIt is subsequently used for the object function in solution formula (3).
4) the representative non-face characteristics dictionary of reference is selected From with reference to non-face characteristics dictionary D-Middle choosing
Select, be D-A subset beRepresentativeness show it when representing masked non-face with good sign energy
Power has separating capacity when representing masked face simultaneously.So as to,Sparsely representing masked non-face characteristics dictionaryWhen
There should be the mistake of minimum, while sparsely representing masked face characteristics dictionaryMistake maximum Shi Yingyou.Therefore,Energy
It is enough to be obtained by solving following equation (5)：
Meet
Above-mentioned formula (5) belongs to α in sparse coding processing, formula1And α2It is to utilize respectivelyRepresent some masked face
Feature x1With some masked non-face feature x2The sparse coefficient vector needed.It is 1 to only have an element in sparse coefficient vector,
Other elements are 0.Using the constraints of sparse coefficient vector, sparse coding processing be equivalent to fromMiddle searching arest neighbors.By
InIn each feature come from reference to non-face characteristics dictionary D-, the optimization problem and the sparse coding of classics of formula (5)
Mode is different, is difficult to solve with classical optimized algorithm.So, the present invention proposes a kind of greedy method effectively from ginseng
Examine non-face characteristics dictionary D-It is middle to buildIn the greedy method of proposition, the present invention calculates refer to non-face tagged word first
Allusion quotation D-In each refer to non-face featureLossThe loss is expressed asWith masked non-face characteristics dictionaryMost
The distance of neighbour's feature andWith masked face characteristics dictionaryArest neighbors feature distance difference, pass through below equation (6)
Realize：Meet
In above-mentioned formula (6), ρ1And ρ2It is two coefficients of balance, for the distance between balance characteristics, leads in actual treatment
1 often is taken with speed-up computation, each with reference to non-face featureIt is rarely used for representingIn masked face characteristic andIn
Masked non-face feature.Pass through counting lossObtain non-according to the reference for losing ascending ascending order arrangement
The non-face feature of reference that foremost is come in face characteristic list, list is representing the ability of masked non-face characteristic aspect most
By force, the ability for representing masked face characteristic is most weak.In this way, can be by way of iteration, constantly by list
First M is added to a feature pool P with reference to non-face feature-In, construct finalIt is preferred that, M is more than or equal to 1 and small
In equal to 50.Specifically, make initial characteristicses pond for sky i.e.Then walk and use in tCome M candidate before selecting,
ObtainThen,In feature be used for updateIt is subsequently used for the object function in solution formula (5).
5) merge dictionary, obtain proximate exterior feature space
In above-mentioned steps, step 1) and 2) without strict sequencing, can carry out successively or parallel；Step 3) and 4) do not have
There is strict sequencing, can carry out successively or parallel.By above-mentioned steps, proximate exterior feature space is constructedThe proximate exterior feature space is to select most have generation from face characteristic and the non-face feature of reference is largely referred to
The feature composition of table, its selection strategy is by being compared with a large amount of masked face characteristics and masked non-face feature
Arrive, comprising feature can represent masked face characteristic well while can also distinguish masked non-face feature, therefore using near
Like surface spaceCandidate feature be embedded in and projects obtained insertion feature to masked face with good sign
Ability.On the other hand, compared with traditional LLE methods, the proximate exterior of quick approximate LLE method constructs proposed by the present invention is special
Levy spaceThan each candidate feature xiCorresponding local feature space DiIt is big, to each candidate feature xi, carry out projective transformation
Afterwards, the approximate embedded feature obtainedFeature v more embedded than the tradition that traditional LLE methods are obtainediDimension is high, to a certain extent
The characteristic present loss quickly approximately brought is compensate for, so quick approximate LLE method constructs proposed by the present invention is approximate outer
Portion's feature spaceFor in masked Face datection, being had little to no effect to accuracy of detection.
By the relatively corresponding representative reference facial image in approximate surface space and representative
With reference to the example of inhuman face image, it is found that the representative reference facial image of selection includes different outward appearances, pendant
Wear, the colour of skin, expression etc., therefore, it is possible to represent masked face well and while distinguish masked non-face well；The tool of selection
Representational is then texture region, imperfect face, the face containing more background with reference to inhuman face image, therefore, it is possible to fine
Ground represents masked non-face and while distinguishes masked face well.
Implement to be merely illustrative of the technical solution of the present invention rather than be limited above, the ordinary skill people of this area
Member can modify or equivalent substitution to technical scheme, without departing from the spirit and scope of the present invention, this hair
Bright protection domain should be to be defined described in claims.
Claims (10)
1. a kind of method for detecting human face, its step includes：
1) candidate face detection is carried out to image to be detected, obtains candidate face image；
2) candidate feature extraction is carried out to the candidate face image, obtains candidate feature；
3) candidate feature is carried out being embedded in conversion, obtains the embedded feature of tradition or be approximately embedded in feature, the embedded feature
The noise that face clue and occlusion removal are brought can be recovered；
4) the embedded feature to the embedded feature of the tradition or approximately, is verified with regression algorithm by classification, obtains detection knot
Really.
2. the method as described in claim 1, it is characterised in that step 3) described in candidate feature built in advance by one
Surface space be embedded in after conversion, obtain the embedded feature of tradition or be approximately embedded in feature；The surface space
For conventional external feature space or proximate exterior feature space.
3. method as claimed in claim 2, it is characterised in that the embedded conversion is locally linear embedding into method using traditional
Or the quick method that is approximately locally linear embedding into is realized；Traditional is locally linear embedding into method using conventional external feature space to band
The candidate feature of noise carries out being embedded in conversion, obtains the embedded feature of tradition；Quick be approximately locally linear embedding into utilizes proximate exterior
Feature space carries out being embedded in conversion to the candidate feature with noise, obtains approximately being embedded in feature.
4. method as claimed in claim 3, it is characterised in that described to be quickly approximately locally linear embedding into proximate exterior in method
The building method of feature space, comprises the following steps：
A) candidate face detection is carried out to the reference face data set marked and candidate feature is extracted, judge that candidate feature belongs to
Face characteristic or non-face feature, these candidate features are stored in reference to face characteristics dictionary respectively and non-face feature is referred to
Dictionary；
B) candidate face detection is carried out to the masked face data set marked and candidate feature is extracted, judge that candidate feature belongs to
These candidate features are stored in masked face characteristics dictionary and masked non-by masked face characteristic or masked non-face feature respectively
Face characteristic dictionary；
C) select representative to represent above-mentioned masked face characteristics dictionary from above-mentioned reference face characteristics dictionary
With reference to face characteristics dictionary；
D) from above-mentioned with reference to selecting in non-face characteristics dictionary representative to represent above-mentioned masked non-face tagged word
The non-face characteristics dictionary of reference of allusion quotation；
E) merge above-mentioned representative reference face characteristics dictionary and the representative non-face characteristics dictionary of reference, obtain
To proximate exterior feature space.
5. method as claimed in claim 4, it is characterised in that in step a), by calculating the corresponding candidate of the candidate feature
Degree of overlapping between face location and the face location marked determines, its degree of overlapping with handing over and than measuring, wherein, hand over simultaneously
Than judging candidate feature more than 0.7 for the feature with reference to face, hand over and than judging candidate feature for reference to inhuman less than 0.3
The feature of face.
6. method as claimed in claim 4, it is characterised in that step b), by calculating the corresponding candidate of the candidate feature
Degree of overlapping between face position and the face location marked determines, its degree of overlapping with handing over and than measuring, wherein, hand over and compare
Feature of the candidate feature for masked face is judged more than 0.6, is handed over and than judging candidate feature to be masked non-face less than 0.4
Feature.
7. method as claimed in claim 4, it is characterised in that using greedy algorithm from reference to face characteristics dictionary in step c)
The representative reference face characteristics dictionary of middle selection；The greedy algorithm refers to calculate with reference to each in face characteristics dictionary
With reference to the loss of face characteristic, obtain, by the reference face feature list for losing ascending ascending order arrangement, taking before the list most
The reference face characteristic in face represents masked face characteristic；Wherein described loss refers to each reference face characteristic and masked face
The distance of the arest neighbors feature of characteristics dictionary and each arest neighbors feature with reference to face characteristic and masked non-face characteristics dictionary
Distance difference.
8. method as claimed in claim 4, it is characterised in that using greedy algorithm from reference to non-face tagged word in step d)
The non-face characteristics dictionary of representative reference is selected in allusion quotation；The greedy algorithm, which refers to calculate, refers to non-face characteristics dictionary
In each refer to the loss of non-face feature, obtain, by the non-face feature list of reference for losing the arrangement of ascending ascending order, taking
The non-face feature of reference of the list foremost represents masked non-face feature；Wherein described loss refers to each with reference to inhuman
Face feature is special with masked face with reference to non-face feature with the distance of the arest neighbors feature of masked non-face characteristics dictionary and each
Levy the difference of the distance of the arest neighbors feature of dictionary.
9. a kind of human face detection device, including candidate block, embedded module and authentication module；
The candidate block is used to carry out candidate face detection to image to be detected and extracts candidate feature；
The embedded module is used to the candidate feature is carried out to be embedded in conversion, obtains the embedded feature of tradition or approximate embedded special
Levy, the embedded feature can recover the noise that face clue and occlusion removal are brought；
The authentication module is used for the embedded feature of above-mentioned tradition or is approximately embedded in feature, is tested by classification with regression algorithm
Card, to obtain last testing result.
10. device as claimed in claim 9, it is characterised in that the candidate block obtains multiple candidate features, then embedding
Enter in module and to carry out being embedded in after conversion by the surface built in advance space, obtain tradition embedded feature or approximate
Embedded feature；The surface space is conventional external feature space or proximate exterior feature space；The embedded conversion is adopted
Realized with traditional being locally linear embedding into method or be quickly approximately locally linear embedding into method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710127367.3A CN107066943B (en) | 2017-03-06 | 2017-03-06 | A kind of method for detecting human face and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710127367.3A CN107066943B (en) | 2017-03-06 | 2017-03-06 | A kind of method for detecting human face and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107066943A true CN107066943A (en) | 2017-08-18 |
| CN107066943B CN107066943B (en) | 2019-10-25 |
Family
ID=59622056
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710127367.3A Active CN107066943B (en) | 2017-03-06 | 2017-03-06 | A kind of method for detecting human face and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107066943B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108563982A (en) * | 2018-01-05 | 2018-09-21 | 百度在线网络技术（北京）有限公司 | Method and apparatus for detection image |
| CN110363126A (en) * | 2019-07-04 | 2019-10-22 | 杭州视洞科技有限公司 | A kind of plurality of human faces real-time tracking and out of kilter method |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101202845A (en) * | 2007-11-14 | 2008-06-18 | 北京大学 | Method for changing infrared image into visible light image |
| CN101393608A (en) * | 2008-11-04 | 2009-03-25 | 清华大学 | Visual object recognition method and apparatus based on manifold distance analysis |
| CN101493885A (en) * | 2009-02-27 | 2009-07-29 | 中国人民解放军空军工程大学 | Embedded human face characteristic extracting method based on nuclear neighborhood protection |
| US20110216977A1 (en) * | 2008-09-17 | 2011-09-08 | Fujitsu Limited | Image processing apparatus and image processing method |
| CN104978549A (en) * | 2014-04-03 | 2015-10-14 | 北京邮电大学 | Three-dimensional face image feature extraction method and system |
| CN105469063A (en) * | 2015-12-04 | 2016-04-06 | 苏州大学 | Robust human face image principal component feature extraction method and identification apparatus |
| CN106056553A (en) * | 2016-05-31 | 2016-10-26 | 哈尔滨工业大学深圳研究生院 | Image inpainting method based on tight frame feature dictionary |
| US20180122512A1 (en) * | 2014-10-21 | 2018-05-03 | uBiome, Inc. | Method and system for microbiome-derived characterization, diagnostics and therapeutics for cutaneous conditions |
-
2017
- 2017-03-06 CN CN201710127367.3A patent/CN107066943B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101202845A (en) * | 2007-11-14 | 2008-06-18 | 北京大学 | Method for changing infrared image into visible light image |
| US20110216977A1 (en) * | 2008-09-17 | 2011-09-08 | Fujitsu Limited | Image processing apparatus and image processing method |
| CN101393608A (en) * | 2008-11-04 | 2009-03-25 | 清华大学 | Visual object recognition method and apparatus based on manifold distance analysis |
| CN101493885A (en) * | 2009-02-27 | 2009-07-29 | 中国人民解放军空军工程大学 | Embedded human face characteristic extracting method based on nuclear neighborhood protection |
| CN104978549A (en) * | 2014-04-03 | 2015-10-14 | 北京邮电大学 | Three-dimensional face image feature extraction method and system |
| US20180122512A1 (en) * | 2014-10-21 | 2018-05-03 | uBiome, Inc. | Method and system for microbiome-derived characterization, diagnostics and therapeutics for cutaneous conditions |
| CN105469063A (en) * | 2015-12-04 | 2016-04-06 | 苏州大学 | Robust human face image principal component feature extraction method and identification apparatus |
| CN106056553A (en) * | 2016-05-31 | 2016-10-26 | 哈尔滨工业大学深圳研究生院 | Image inpainting method based on tight frame feature dictionary |
Non-Patent Citations (1)
| Title |
|---|
| 严鹏 等: "基于图谱的多标记特征选择算法", 《计算机科学与探》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108563982A (en) * | 2018-01-05 | 2018-09-21 | 百度在线网络技术（北京）有限公司 | Method and apparatus for detection image |
| CN108563982B (en) * | 2018-01-05 | 2020-01-17 | 百度在线网络技术（北京）有限公司 | Method and apparatus for detecting image |
| CN110363126A (en) * | 2019-07-04 | 2019-10-22 | 杭州视洞科技有限公司 | A kind of plurality of human faces real-time tracking and out of kilter method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107066943B (en) | 2019-10-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107330396B (en) | Pedestrian re-identification method based on multi-attribute and multi-strategy fusion learning | |
| CN110349136A (en) | A kind of tampered image detection method based on deep learning | |
| CN108665481A (en) | Multilayer depth characteristic fusion it is adaptive resist block infrared object tracking method | |
| CN103198493B (en) | A kind ofly to merge and the method for tracking target of on-line study based on multiple features self-adaptation | |
| Feng et al. | Robust and efficient algorithms for separating latent overlapped fingerprints | |
| CN106228528B (en) | A kind of multi-focus image fusing method based on decision diagram and rarefaction representation | |
| CN104599286B (en) | A kind of characteristic tracking method and device based on light stream | |
| CN105574534A (en) | Significant object detection method based on sparse subspace clustering and low-order expression | |
| CN107424161B (en) | Coarse-to-fine indoor scene image layout estimation method | |
| Zheng et al. | Attention-based spatial-temporal multi-scale network for face anti-spoofing | |
| CN105205449A (en) | Sign language recognition method based on deep learning | |
| CN108564040B (en) | Fingerprint activity detection method based on deep convolution characteristics | |
| CN104240256A (en) | Image salient detecting method based on layering sparse modeling | |
| CN113095263B (en) | Training method and device for pedestrian re-recognition model under shielding and pedestrian re-recognition method and device under shielding | |
| CN106203284B (en) | Method for detecting human face based on convolutional neural networks and condition random field | |
| CN108776777A (en) | The recognition methods of spatial relationship between a kind of remote sensing image object based on Faster RCNN | |
| CN102270308A (en) | Facial feature location method based on five sense organs related AAM (Active Appearance Model) | |
| Yao et al. | Sensing urban land-use patterns by integrating Google Tensorflow and scene-classification models | |
| CN111950525A (en) | Fine-grained image classification method based on destructive reconstruction learning and GoogLeNet | |
| CN106611158A (en) | Method and equipment for obtaining human body 3D characteristic information | |
| CN103226825B (en) | Based on the method for detecting change of remote sensing image of low-rank sparse model | |
| CN115457277A (en) | Intelligent pavement disease identification and detection method and system | |
| Ren et al. | Mask wearing detection based on YOLOv3 | |
| CN103714340A (en) | Self-adaptation feature extracting method based on image partitioning | |
| CN107066943A (en) | A kind of method for detecting human face and device |
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 |