CN107316029B - A kind of living body verification method and equipment - Google Patents
A kind of living body verification method and equipment Download PDFInfo
- Publication number
- CN107316029B CN107316029B CN201710533353.1A CN201710533353A CN107316029B CN 107316029 B CN107316029 B CN 107316029B CN 201710533353 A CN201710533353 A CN 201710533353A CN 107316029 B CN107316029 B CN 107316029B
- Authority
- CN
- China
- Prior art keywords
- characteristic point
- parameter
- point
- image
- matching
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
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
- G06V40/168—Feature extraction; Face representation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/46—Extracting features or characteristics from the video content, e.g. video fingerprints, representative shots or key frames
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Human Computer Interaction (AREA)
- Image Analysis (AREA)
Abstract
The embodiment of the invention discloses a kind of living body verification method and equipment.The method includes:Image data is obtained based on action command, parses described image data, identifies the region for characterizing face position in described image data;The variation of face position in the multiple image for being included based on described image data, tracks the region;Extract the textural characteristics in the region;Parameter based on the textural characteristics computational representation posture;Movement is determined based on the parameter;When movement movement corresponding with action command matching, determine that living body is verified.
Description
Technical field
The present invention relates to face recognition technologies, and in particular to a kind of living body verification method and equipment.
Background technique
Currently, more and more authentication systems authenticate user identity using face recognition technology.Especially with
Mobile terminal client terminal it is universal, more and more faces verifying systems replace traditional password authentification to become mainstream.But with
The extensive use of face recognition technology, there are various methods and pretend to be living body faces with by authentication, such as pass through photograph
The duplicity such as piece, video are identified to pass through authentication;In order to take precautions against increasingly multiplicity, increasingly have fraudulent attacker
Formula, the living body faces recognizer in authentication system is also more complicated, and the application capacity of generation also can be increasing.And it moves
The limited storage space of client terminals is to accommodate biggish living body verification algorithm;Meanwhile the processing capacity of processor
The time on the weak side for making data processing is longer, this significantly limits the application of living body faces verifying system on mobile terminals.
Summary of the invention
To solve existing technical problem, the embodiment of the present invention provides a kind of living body verification method and equipment.
In order to achieve the above objectives, the technical solution of the embodiment of the present invention is realized in:
The embodiment of the invention provides a kind of living body verification method, the method includes:
Image data is obtained based on action command, described image data is parsed, identifies in described image data and characterize people
The region of face position;
The variation of face position in the multiple image for being included based on described image data, tracks the region;
Extract the textural characteristics in the region;Parameter based on the textural characteristics computational representation posture;Based on described
Parameter determines movement;
When movement movement corresponding with action command matching, determine that living body is verified.
In above scheme, the variation of face position in the multiple image for being included based on described image data,
The region is tracked, including:
The first frame image and the second frame image in the multiple image are extracted, identifies in the first frame image and characterizes people
The first area of face position obtains corresponding first coordinate range in the first area;
Obtain initial coordinate range corresponding with first coordinate range in the second frame image;
Calculate the corresponding offset parameter of the initial coordinate range;
The second coordinate range is obtained based on the initial coordinate range and corresponding offset parameter, described second is recorded and sits
Mark range is the region of the characterization face position after tracking;
Wherein, the offset parameter characterizes offset journey of second coordinate range relative to first coordinate range
Degree.
It is described to obtain initial coordinate model corresponding with first coordinate range in the second frame image in above scheme
It encloses, including:
First group of N number of characteristic point is chosen in the corresponding first area of first coordinate range by preset step-length, is obtained
First coordinate of fisrt feature point in first group of N number of characteristic point;N is positive integer;Wherein, the fisrt feature point is described
Any feature point in first group of N number of characteristic point;
Obtain second group of N number of characteristic point in the second frame image, second feature point in second group of N number of characteristic point
The second coordinate it is identical as the first coordinate of corresponding fisrt feature point in first group of N number of characteristic point;
Initial coordinate range is determined based on the second coordinate of each characteristic point in second group of N number of characteristic point.
It is described to calculate the corresponding offset parameter of the initial coordinate range in above scheme, including:
Each characteristic point is calculated in second group of N number of characteristic point relative to corresponding in first group of N number of characteristic point
First offset parameter of characteristic point;The identical second feature point of first offset parameter characterization coordinate and first spy
Difference degree between sign point;
Multiple matching characteristics are determined based on corresponding first offset parameter of characteristic point each in second group of N number of characteristic point
Point calculates the second offset parameter of each matching characteristic point in the multiple matching characteristic point;The second offset parameter characterization
The matching characteristic point in second group of N number of characteristic point, between third feature point corresponding with the matching characteristic point
Degrees of offset;
And calculate the third offset parameter of each matching characteristic point in the multiple matching characteristic point;The third is inclined
Shifting parameter characterize the matching characteristic point in first group of N number of characteristic point, with the matching characteristic point the corresponding 4th
Degrees of offset between characteristic point;
The corresponding offset of the initial coordinate range is determined according to second offset parameter and the third offset parameter
Parameter.
In above scheme, the third offset parameter for calculating each matching characteristic point in the multiple matching characteristic point,
Including:
Multiple groups fisrt feature point pair is extracted from multiple matching characteristic points that the second frame image is included respectively;It is described
Fisrt feature point to include the first matching characteristic point and the second matching characteristic point, and from the first frame image, with it is described
Extract multiple groups second feature point pair in the corresponding source characteristic point of multiple matching characteristic points, the second feature point is to including first
Source characteristic point and the second source characteristic point;Wherein, the first matching characteristic point and the second matching characteristic point are the multiple
Any two characteristic point in matching characteristic point;
The first distance between the first matching characteristic point and the second matching characteristic point is calculated, and described in calculating
Second distance between first source characteristic point and second source characteristic point;
Obtain the relative parameter between the first distance and the second distance;The relative parameter is denoted as described
The third offset parameter of 1 matching characteristic point and the second matching characteristic point.
It is described that the initial coordinate is determined according to second offset parameter and the third offset parameter in above scheme
The corresponding offset parameter of range, including:
Multiple second offset parameters are handled by the first default processing rule, obtain particular offset parameter;
And handle multiple relative parameters by the second default processing rule, obtain specific relative parameter;
Using the particular offset degree and the specific relative parameter as the corresponding offset ginseng of the initial coordinate range
Number.
It is described based on corresponding first offset parameter of characteristic point each in second group of N number of characteristic point in above scheme
Determine multiple matching characteristic points, including:
Each feature is determined based on corresponding first offset parameter of characteristic point each in second group of N number of characteristic point
The corresponding target feature point of point, obtains the third coordinate of the target feature point;
Determine initial characteristics point corresponding with the target feature point in the first frame image;The initial characteristics point
4-coordinate it is identical as the third coordinate of the target feature point;
The initial characteristics point is obtained relative to the 4th offset parameter between the target feature point;
When the 4th offset parameter reaches preset threshold, determine that the target feature point is matching characteristic point;
When the 4th offset parameter is not up to the preset threshold, determine that the target feature point is not matching characteristic
Point.
In above scheme, the textural characteristics extracted in the region;Based on the textural characteristics computational representation posture
Parameter;Movement is determined based on the parameter, including:
Extract the first textural characteristics and/or the second textural characteristics in the region;
Based on the first parameter of the first posture of the first textural characteristics computational representation, and/or, it is based on second texture
Feature calculation characterizes the second parameter of the second posture;
The first movement is determined based on first parameter, and/or, the second movement is determined based on second parameter.
In above scheme, first textural characteristics extracted in the region, including:
Characteristic point in the region is handled according to the default processing rule of third, obtains in the region and characterizes often
First procedure parameter of the difference degree of a characteristic point characteristic point adjacent with the characteristic point;
Analyze the first textural characteristics that first procedure parameter obtains characterization face texture edge;
Correspondingly, second textural characteristics extracted in the region, including:
Extract the first part region in the region;
Characteristic point in the first part region is handled according to the 4th default processing rule, obtains described first
The second procedure parameter of the difference degree of each characteristic point characteristic point adjacent with the characteristic point is characterized in partial region;
Analyze the second textural characteristics that second procedure parameter obtains characterization eye texture edge.
In above scheme, first parameter based on the first posture of the first textural characteristics computational representation, including:
First textural characteristics are inputted into preconfigured first disaggregated model, obtain the first ginseng of the first posture of characterization
Number.
In above scheme, second parameter based on the second posture of the second textural characteristics computational representation, including:
Second textural characteristics are inputted into preconfigured second disaggregated model, obtain the second ginseng of the second posture of characterization
Number.
It is described that first movement is determined based on first parameter in above scheme, including:
Based on corresponding multiple first parameters of the multiple image, first part's image in the multiple image is judged
Whether corresponding first parameter is all satisfied second part image corresponding first in first threshold range and the multiple image
Whether parameter is not satisfied the first threshold range;
When corresponding first parameter of first part's image is all satisfied first threshold range and described in the multiple image
When the first threshold range is not satisfied in corresponding first parameter of second part image in multiple image, first ginseng is determined
Number corresponds to the first movement.
It is described that second movement is determined based on second parameter in above scheme, including:
Based on corresponding multiple second parameters of the multiple image, Part III image in the multiple image is judged
Whether corresponding second parameter is all satisfied Part IV image corresponding second in second threshold range and the multiple image
Whether parameter is not satisfied the second threshold range;
When corresponding second parameter of Part III image is all satisfied second threshold range and described in the multiple image
When the second threshold range is not satisfied in corresponding second parameter of Part IV image in multiple image, second ginseng is determined
Number corresponds to the second movement.
The embodiment of the invention also provides a kind of living bodies to verify equipment, and the equipment includes:Detection unit, tracking cell,
Feature extraction unit, computing unit, movement judging unit and authentication unit;Wherein,
The detection unit parses described image data, identifies described for obtaining image data based on action command
The region of face position is characterized in image data;
The tracking cell, the multiple image that the described image data for being identified based on the detection unit are included
The variation of middle face position, tracks the region;
The feature extraction unit, the textural characteristics in the region for extracting the tracking cell tracking;
The computing unit, the textural characteristics computational representation posture for being extracted based on the feature extraction unit
Parameter;
The movement judging unit, the parameter for being obtained based on the computing unit determine movement;
The authentication unit, for determining that living body is tested when movement movement corresponding with action command matching
Card passes through.
In above scheme, the tracking cell, for extracting first frame image and the second frame figure in the multiple image
Picture identifies the first area for characterizing face position in the first frame image, obtains the first area corresponding first
Coordinate range;Obtain initial coordinate range corresponding with first coordinate range in the second frame image;It calculates described first
The corresponding offset parameter of beginning coordinate range;The second coordinate model is obtained based on the initial coordinate range and corresponding offset parameter
It encloses, records the region that second coordinate range is the characterization face position after tracking;Wherein, the offset parameter characterization
Degrees of offset of second coordinate range relative to first coordinate range.
In above scheme, the tracking cell, for pressing preset step-length in corresponding firstth area of first coordinate range
First group of N number of characteristic point is chosen in domain, obtains the first coordinate of fisrt feature point in first group of N number of characteristic point;N is positive whole
Number;Wherein, the fisrt feature point is any feature point in first group of N number of characteristic point;Obtain the second frame image
In second group of N number of characteristic point, in second group of N number of characteristic point the second coordinate of second feature point and described first group it is N number of
The first coordinate of corresponding fisrt feature point is identical in characteristic point;Based on each characteristic point in second group of N number of characteristic point
The second coordinate determine initial coordinate range.
In above scheme, the tracking cell, for calculate in second group of N number of characteristic point each characteristic point relative to
First offset parameter of the individual features point in first group of N number of characteristic point;The first offset parameter characterization coordinate is identical
The second feature point and fisrt feature point between difference degree;Based on each in second group of N number of characteristic point
Corresponding first offset parameter of characteristic point determines multiple matching characteristic points, and it is special to calculate each matching in the multiple matching characteristic point
Levy the second offset parameter of point;In the second offset parameter characterization matching characteristic point and second group of N number of characteristic point,
Degrees of offset between third feature point corresponding with the matching characteristic point;And calculate the multiple matching characteristic point
In each matching characteristic point third offset parameter;The third offset parameter characterizes the matching characteristic point and described first group
Degrees of offset in N number of characteristic point, between fourth feature point corresponding with the matching characteristic point;According to second offset
Parameter and the third offset parameter determine the corresponding offset parameter of the initial coordinate range.
In above scheme, the tracking cell, multiple matching characteristics for being included from the second frame image respectively
Multiple groups fisrt feature point pair is extracted in point;The fisrt feature point to include the first matching characteristic point and the second matching characteristic point,
And from the first frame image, in source characteristic point corresponding with the multiple matching characteristic point extract multiple groups second feature
Point pair, the second feature point is to including the first source characteristic point and the second source characteristic point;Wherein, the first matching characteristic point and
The second matching characteristic point is any two characteristic point in the multiple matching characteristic point;Calculate first matching characteristic
First distance between point and the second matching characteristic point, and calculate first source characteristic point and second source feature
Second distance between point;Obtain the relative parameter between the first distance and the second distance;By the relative parameter
It is denoted as the third offset parameter of the first matching characteristic point and the second matching characteristic point.
In above scheme, the tracking cell, for carrying out multiple second offset parameters by the first default processing rule
Processing obtains particular offset parameter;And handle multiple relative parameters by the second default processing rule, it obtains specific
Relative parameter;Using the particular offset degree and the specific relative parameter as the corresponding offset ginseng of the initial coordinate range
Number.
In above scheme, the tracking cell, for corresponding based on each characteristic point in second group of N number of characteristic point
First offset parameter determines the corresponding target feature point of each characteristic point, obtains the third coordinate of the target feature point;
Determine initial characteristics point corresponding with the target feature point in the first frame image;The 4th of the initial characteristics point sits
It marks identical as the third coordinate of the target feature point;The initial characteristics point is obtained relative between the target feature point
4th offset parameter;When the 4th offset parameter reaches preset threshold, determine that the target feature point is matching characteristic point;
When the 4th offset parameter is not up to the preset threshold, determining the target feature point not is matching characteristic point.
In above scheme, the feature extraction unit, for extracting the first textural characteristics and/or second in the region
Textural characteristics;
The computing unit, for the first parameter based on the first posture of the first textural characteristics computational representation, and/
Or, the second parameter based on the second posture of the second textural characteristics computational representation;
The movement judging unit, for determining the first movement based on first parameter, and/or, it is based on described second
Parameter determines the second movement.
In above scheme, the feature extraction unit, for being handled according to third is default the characteristic point in the region
Rule is handled, and the difference degree that each characteristic point characteristic point adjacent with the characteristic point is characterized in the region is obtained
First procedure parameter;Analyze the first textural characteristics that first procedure parameter obtains characterization face texture edge;It is also used to take out
Take the first part region in the region;To the characteristic point in the first part region according to the 4th default processing rule into
Row processing, obtains the difference degree that each characteristic point characteristic point adjacent with the characteristic point is characterized in the first part region
The second procedure parameter;Analyze the second textural characteristics that second procedure parameter obtains characterization eye texture edge.
In above scheme, the computing unit, for first textural characteristics input preconfigured first to be classified
Model obtains the first parameter of the first posture of characterization;And/or for second textural characteristics to be inputted preconfigured the
Two disaggregated models obtain the second parameter of the second posture of characterization.
In above scheme, the movement judging unit, for based on corresponding multiple first ginsengs of the multiple image
Number, judges whether corresponding first parameter of first part's image is all satisfied first threshold range and institute in the multiple image
State whether corresponding first parameter of second part image in multiple image is not satisfied the first threshold range;When the multiframe
Corresponding first parameter of first part's image is all satisfied second part in first threshold range and the multiple image in image
When the first threshold range is not satisfied in corresponding first parameter of image, determine that first parameter corresponds to the first movement;
And/or for being based on corresponding multiple second parameters of the multiple image, judge Part III figure in the multiple image
As whether corresponding second parameter is all satisfied Part IV image in second threshold range and the multiple image corresponding
Whether two parameters are not satisfied the second threshold range;When corresponding second parameter of Part III image in the multiple image
It is all satisfied corresponding second parameter of Part IV image in second threshold range and the multiple image and is not satisfied described
When two threshold ranges, determine that second parameter corresponds to the second movement.
Living body verification method provided in an embodiment of the present invention and equipment, the living body verification method include:Referred to based on movement
It enables and obtains image data, parse described image data, identify the region for characterizing face position in described image data;Base
The variation of face position in the multiple image that described image data are included, tracks the region;Extract the region
In textural characteristics;Parameter based on the textural characteristics computational representation posture;Movement is determined based on the parameter;When described dynamic
When making movement corresponding with action command matching, determine that living body is verified.Using the technical solution of the embodiment of the present invention,
It is instructed by active output action and determines in image data whether to include the side acted accordingly according to the image data of acquisition
Living body when formula determines whether, therefore, it is determined that living body verifies whether to pass through;The technical solution realization of the embodiment of the present invention passes through region
The mode of tracking substitutes recognition of face detection, only needs when corresponding to parameter according to textural characteristics calculating posture in scheme preparatory
The computation model of configuration, and the computation model capacity is minimum, therefore, living body proof scheme provided in an embodiment of the present invention is carried
Algorithm file it is smaller, meet the memory space requirements of mobile terminal significantly;And data processing amount substantially reduces, and also meets
The demand of the processing capacity of mobile terminal.
Detailed description of the invention
Fig. 1 is the overall procedure schematic diagram of the living body verification method of the embodiment of the present invention;
Fig. 2 is the details flow diagram of the living body verification method of the embodiment of the present invention;
Fig. 3 is the flow diagram of the living body verification method of the embodiment of the present invention;
Fig. 4 a to Fig. 4 e is respectively the application display schematic diagram in the living body verification method of the embodiment of the present invention;
Fig. 5 is that the living body of the embodiment of the present invention verifies the composed structure schematic diagram of equipment;
Fig. 6 is that the living body of the embodiment of the present invention verifies composed structure schematic diagram of the equipment as hardware.
Specific embodiment
With reference to the accompanying drawing and specific embodiment the present invention is described in further detail.
Before the living body verification method to the embodiment of the present invention is described in detail, first to the work of the embodiment of the present invention
The general realisation of body proof scheme is illustrated.Fig. 1 is that the overall procedure of the living body verification method of the embodiment of the present invention shows
It is intended to;As shown in Figure 1, the living body verification method of the embodiment of the present invention may include following several stages:
Stage 1:Input video stream namely living body verifying equipment obtain image data.Before input video stream, need first
Output action instruction, so that identifying object (object that can be understood as user or the verifying of pending living body) is according to the movement
Instruction execution corresponding actions;Image data is acquired based on the corresponding actions that identifying object executes.
Stage 2:Living body verifying equipment carries out Face datection and tracks, and determines face place according to the human face data detected
Region, and the region where the tracking face;In the exportable mark in the image output unit side of living body verifying equipment
The face frame of face region.Recognition of face is carried out just for the first frame image in image data in the application and is detected,
Effect is the region obtained where face, it is understood that for the position for obtaining " face frame ";Use track algorithm to people rear
Face region is tracked.Wherein, the specific implementation process of face tracking can refer to step in subsequent detailed description embodiment
101 and the corresponding description of step 102 shown in.
Stage 3:In vivo detection, after testing result is shown to be living body, into the stage 4:Image data is sent to backstage
Carry out further face verification;After testing result shows not to be living body, the In vivo detection stage is reentered.Wherein, living body
The specific implementation process of detection can refer in subsequent detailed description embodiment shown in step 103 and the corresponding description of step 106.
Based on Fig. 1, Fig. 2 is the details flow diagram of the living body verification method of the embodiment of the present invention, mainly to work in Fig. 2
The implementation that physical examination is surveyed is refined, and specifically can refer to shown in Fig. 2, particularly may be divided into texture feature extraction, ginseng for the stage 3
Number calculates and movement judges three processes;Wherein, the textural characteristics can specifically include the of characterization face texture feature
One textural characteristics, and/or, the second textural characteristics of eye textural characteristics are characterized, parameter calculated can be understood as face and be in
The score of existing posture, and/or, the score of posture is presented in eye;The movement judged specifically can be understood as rotary head and act, and/
Or, blink acts, concrete implementation mode be can refer to described in subsequent detailed description embodiment.That is, the present invention is implemented
Whether the head portrait (face in other words) that the technical solution of example determines that image is included from the image of acquisition can be based on a specified
It acts and is acted accordingly, therefore, it is determined that whether being living body faces.Based on this, acts of determination is dynamic for rotary head in the stage 3
After work and/or blink movement, in the stage 4, the rotary head is acted and/or blink movement is corresponding with initial action command
Required movement matched, if matching is consistent, be shown to be living body, execute the stage 4:Image data is sent to backstage to carry out
Face verification;If face tracking fails, perhaps movement judgement time-out or the rotary head movement and/or blink movement with it is initial
The corresponding required movement of action command mismatch, then show non-living body.
The embodiment of the invention provides a kind of living body verification methods.Fig. 3 is the living body verification method of the embodiment of the present invention
Flow diagram;As shown in figure 3, the living body verification method of the embodiment of the present invention includes:
Step 101:Image data is obtained based on action command, described image data is parsed, identifies described image data
The region of middle characterization face position.
Step 102:The variation of face position in the multiple image for being included based on described image data, described in tracking
Region.
Step 103:Extract the textural characteristics in the region;Parameter based on the textural characteristics computational representation posture.
Step 104:Movement is determined based on the parameter.
Step 105:When movement movement corresponding with action command matching, determine that living body is verified.
The living body verification method of the embodiment of the present invention is applied in living body verifying equipment.The living body verifying equipment specifically may be used
To be the electronic equipment with image acquisition units, to obtain image data by described image acquisition unit;The electronics is set
It is standby specifically to can be the mobile devices such as mobile phone, tablet computer, it is also possible to personal computer, configured with the access control system (door
Access control system is specially the system for carrying out control to exit and entrance) access control equipment etc.;Wherein, described image acquisition unit has
Body can be the camera of setting on an electronic device.On the other hand, the living body verifying equipment can also be defeated with audio
The electronic equipment of unit out, to be instructed by the audio output unit output action.The technical solution of the embodiment of the present invention is
It is instructed by active output action, further acquisition includes the image data of user's head portrait, and the image data based on acquisition is sentenced
Determine whether user executes corresponding movement therefore, it is determined that living body verifies whether to pass through.
In the present embodiment, living body verifies equipment, and (in the following embodiment of the present invention, the living body verifying equipment is referred to as
Equipment) after obtaining image data by image acquisition units, parse described image data;Wherein, image data obtained
Including multiple image.The region for identifying characterization face position in described image data, specially:It identifies described
The region of face position is characterized in the first frame image of image data, to characterize face institute in the first frame image
On the basis of the region of position, based on the variation of face position in the multiple image, the region is tracked.
It is as an implementation, described to be wrapped based on described image data for the tracking mode of face region
The variation of face position in the multiple image contained tracks the region, including:Extract the first frame in the multiple image
Image and the second frame image identify the first area for characterizing face position in the first frame image, obtain described first
Corresponding first coordinate range in region;Obtain initial coordinate model corresponding with first coordinate range in the second frame image
It encloses;Calculate the corresponding offset parameter of the initial coordinate range;Based on the initial coordinate range and corresponding offset parameter
The second coordinate range is obtained, the region that second coordinate range is the characterization face position after tracking is recorded;Wherein, institute
It states offset parameter and characterizes degrees of offset of second coordinate range relative to first coordinate range.
In the present embodiment, in order to reduce Face datection load, while splicing video attack is avoided, the embodiment of the present invention uses
Face tracking mode substitutes Face datection.Specifically, equipment carries out Face datection for the first frame image of described image data,
The specific mode that facial feature points detection can be used identifies the face in the first frame image, and then determines the first frame
The first area of face position is characterized in image;Wherein, the first area can be indicated by the first coordinate range.Having
In body application process, the first area is known in the image-display units of equipment by the face collimation mark of display, such as Fig. 4 a
It is shown;The face collimation mark knows corresponding coordinate range and matches with first coordinate range.Further, described first
It is tracked on the basis of the first area in frame image.Wherein, the first frame image can be for for a target person
First frame image in the image data of object acquisition;The present embodiment be former two field pictures treatment process for be illustrated,
It certainly, during actual image real time transfer, is handled for multiple image data, the multiple image data
Treatment process can refer to the treatment process that front cross frame image data is directed to described in the present embodiment.
It is described to obtain initial coordinate model corresponding with first coordinate range in the second frame image in the present embodiment
It encloses, including:First group of N number of characteristic point is chosen in the corresponding first area of first coordinate range by preset step-length, is obtained
First coordinate of fisrt feature point in first group of N number of characteristic point;N is positive integer;Wherein, the fisrt feature point is described
Any feature point in first group of N number of characteristic point;Obtain second group of N number of characteristic point in the second frame image, described second
Second coordinate of second feature point and corresponding fisrt feature point in first group of N number of characteristic point in the N number of characteristic point of group
First coordinate is identical;Initial coordinate range is determined based on the second coordinate of each characteristic point in second group of N number of characteristic point.
Specifically, first group of N number of characteristic point is chosen by preset step-length for the first area in the first frame image,
Obtain the coordinate of each characteristic point in first group of N number of characteristic point.For example, the first area meets m × n, wherein m is small
In the length/width for being equal to the first frame image;Then n is less than or equal to the width/height of the first frame image;It then can be according to
Step-length is that m × n/ (10 × 10) uniformly choose 100 characteristic points in the first area.Wherein, the step-length can be not limited to
The above-mentioned step-length enumerated, the step-length can be according to configuring at equal intervals, and the step-length can also be configured according to unequal interval, can specifically be pressed
It is configured according to actual demand.Further, according to each characteristic point in first group of N number of characteristic point in the first frame image
Coordinate, obtain the second frame image in second group of N number of characteristic point so that each characteristic point in second group of N number of characteristic point
Coordinate it is identical as the coordinate of corresponding characteristic point in first group of N number of characteristic point.For example, if suitable according to identical arrangement
Sequence simultaneously identifies, i.e., the characteristic point in described first group of N number of characteristic point and second group of N number of characteristic point according to from top to bottom,
Sequence from left to right is ranked up, then the coordinate and described first of first characteristic point in second group of N number of characteristic point
The coordinate of first characteristic point in the N number of characteristic point of group is identical, the seat of second characteristic point in second group of N number of characteristic point
Mark it is identical as the coordinate of second characteristic point in first group of N number of characteristic point, and so on, then second group of N number of spy
The coordinate (i.e. the second coordinate) of any feature point (i.e. second feature point) is corresponding with first group of N number of characteristic point in sign point
Fisrt feature point coordinate (i.e. the first coordinate) it is identical.Further, based on each feature in second group of N number of characteristic point
Second coordinate of point determines a coordinate range, and the coordinate range is determined as the initial coordinate range.
It is described to calculate the corresponding offset parameter of the initial coordinate range in the present embodiment, including:Calculate described second group
First offset parameter of each characteristic point relative to the individual features point in first group of N number of characteristic point in N number of characteristic point;Institute
State the difference degree between the first offset parameter characterization identical second feature point of coordinate and fisrt feature point;It is based on
Corresponding first offset parameter of each characteristic point determines multiple matching characteristic points in second group of N number of characteristic point, described in calculating
Second offset parameter of each matching characteristic point in multiple matching characteristic points;Second offset parameter characterizes the matching characteristic
It puts and the degrees of offset in second group of N number of characteristic point, between third feature point corresponding with the matching characteristic point;With
And calculate the third offset parameter of each matching characteristic point in the multiple matching characteristic point;The third offset parameter characterization
The matching characteristic point in first group of N number of characteristic point, between fourth feature point corresponding with the matching characteristic point
Degrees of offset;Determine that the initial coordinate range is corresponding partially according to second offset parameter and the third offset parameter
Shifting parameter.
Wherein, described to calculate in second group of N number of characteristic point before the first offset parameter of each characteristic point, the side
Method further includes:It is handled for the first frame image and the second frame image, obtains the first frame image respectively
Second gradient image of first gradient image and the second frame image.Specifically, for the first frame image and described
Two frame images establish the first pyramid diagram picture of the first frame image respectively, establish the second gold medal word of the second frame image
Tower image;The first gradient image of the first pyramid diagram picture is calculated, and calculates the second of the second pyramid diagram picture
Gradient image;Wherein, the pyramid diagram picture for establishing image (such as establishes the first pyramid diagram picture of the first frame image, again
Such as second pyramid diagram picture for establishing the second frame image etc.) it can be understood as downscaled images pari passu, to obtain more
The result of robustness.In the specific implementation process, image can be smoothed first, then to the image after smoothing processing into
Line sampling, to can get the pyramid diagram picture of size reduction.In the present embodiment, the first gradient image and second ladder
Each characteristic point gradient is represented by degree image:
Gx (y, x)=Img (y, x+1)-Img (y, x-1) (1)
Gy (y, x)=Img (y+1, x)-Img (y-1, x) (2)
Wherein, Gx (y, x) indicates the gradient of characteristic point (y, x) in x-axis;Gy (y, x) indicates characteristic point (y, x) in y-axis
On gradient;Img (y, x) indicates the display parameters of characteristic point (y, x);Correspondingly, Img (y, x+1) indicates characteristic point (y, x+1)
Display parameters;Img (y, x-1) indicates the display parameters of characteristic point (y, x+1);Img (y+1, x) indicates characteristic point (y+1, x)
Display parameters;Img (y-1, x) indicates the display parameters of characteristic point (y-1, x);As an implementation, the display ginseng
Number specifically can be gray value, it is of course also possible to be other display parameters.
Further, in second group of N number of characteristic point each characteristic point relative in first group of N number of characteristic point
The gradient for the individual features point that first offset parameter of individual features point can be obtained based on expression formula (1) and expression formula (2) calculates
It obtains;Wherein, in second group of N number of characteristic point each characteristic point relative to the corresponding spy in first group of N number of characteristic point
Sign point refers to:Characteristic point 2 in second group of N number of characteristic point is relative to the characteristic point in first group of N number of characteristic point
1, characteristic point 2 is identical with coordinate of the characteristic point 1 in the first frame image in the coordinate in the second frame image;Namely institute
It states the first offset parameter and is characterized a little 2 opposite degrees of offset relative to characteristic point 1.Specifically, for second group of N number of spy
Each characteristic point in sign point, analyzes neighbour of each characteristic point relative to the individual features point in first group of N number of characteristic point
Characteristic of field obtains difference parameter (diff) and gradient parameter (grad), is obtained based on the difference parameter and gradient parameter calculating
Obtain the first offset parameter;Wherein, the difference parameter characterizes the characteristic point in second group of N number of characteristic point relative to first group
The difference degree of individual features point in N number of characteristic point.Wherein, the difference parameter and the gradient parameter specifically can by with
Lower expression formula indicates:
Diff=Img1 (y1, x1)-Img2 (y2, x2) (3)
Gx=Gx (y1, x1)+Gx (y2, x2) (4)
Gy=Gy (y1, x1)+Gy (y2, x2) (5)
Wherein, Img1 (y1, x1) indicate in first group of N number of characteristic point, the display parameters of characteristic point (y1, x1);Img2
(y2, x2) indicate in second group of N number of characteristic point, the display parameters of characteristic point (y2, x2);Wherein, the display parameters specifically may be used
To be gray value, it is of course also possible to be other display parameters.Gx indicates gradient in the direction of the x axis;Gy is indicated in y-axis direction
On gradient;Gx (y1, x1) indicates the gradient of characteristic point (y1, x1) in the direction of the x axis in first frame image;Gx (y2, x2) table
Show in the second frame image the gradient of corresponding characteristic point (y2, x2) in the direction of the x axis, characteristic point with characteristic point (y1, x1)
(y1, x1) is identical as the coordinate of characteristic point (y2, x2) in the second needle image in the coordinate in first frame image;Correspondingly, Gy
(y1, x1) indicates the gradient of characteristic point (y1, x1) in the y-axis direction in first frame image;Gy (y2, x2) indicates the second frame image
In with characteristic point (y1, x1) gradient of corresponding characteristic point (y2, x2) in the y-axis direction.
Further, the first offset parameter of individual features point is calculated according to the difference parameter and the gradient parameter,
First offset parameter can be indicated by following formula:
(dy, dx)=f (Gxx, Gxy, Gyy, Diff) (6)
Wherein it is determined that in first frame image characteristic point 1 gradient parameter, be denoted as Gx1 and Gy1;And determine the second frame figure
The gradient parameter of characteristic point 2 corresponding with characteristic point 1, is denoted as Gx2 and Gy2 as in;Then based on (Gx1, Gy1) and (Gx2,
Gy2 matrix operation is carried out) to integrate the gradient of the characteristic point 1 and the characteristic point 2;Then in expression formula (6), Gxx indicates special
The operation result of sign 1 gradient of the gradient and characteristic point 2 in x-axis in x-axis of point;Gyy indicates the ladder of characteristic point 1 on the y axis
The operation result of degree and the gradient of characteristic point 2 on the y axis;Gxy indicates gradient and characteristic point 2 of the characteristic point 1 in x-axis in y-axis
On gradient operation result, or indicate the operation knot of the gradient of gradient and characteristic point 2 in x-axis on the y axis of characteristic point 1
Fruit.F () indicates certain operations rule.
During specific implementation, based on expression formula (6) internal successive ignition, it can get in the first frame image
Matching characteristic point (y0, x0) of one characteristic point (y1, x1) on the second frame image, the matching characteristic point meet following table
Up to formula:
(y0, x0)=(y2+dy, x2+dx) (7)
But since the position of the face in image is in variation, the portion in only described first group of N number of characteristic point
Point characteristic point can find matching characteristic point in the second frame image.
As an implementation, it is described based on characteristic point corresponding first each in second group of N number of characteristic point partially
Shifting parameter determines multiple matching characteristic points, including:Partially based on characteristic point each in second group of N number of characteristic point corresponding first
Shifting parameter determines the corresponding target feature point of each characteristic point, obtains the third coordinate of the target feature point;Determine institute
State initial characteristics point corresponding with the target feature point in first frame image;The 4-coordinate of the initial characteristics point and institute
The third coordinate for stating target feature point is identical;It is inclined relative to the 4th between the target feature point to obtain the initial characteristics point
Shifting parameter;When the 4th offset parameter reaches preset threshold, determine that the target feature point is matching characteristic point;When described
When 4th offset parameter is not up to the preset threshold, determining the target feature point not is matching characteristic point.
Specifically, in present embodiment, in obtaining the second frame image by adopting the above technical scheme, as first frame image
After the corresponding matching characteristic point (y0, x0) of middle characteristic point (y1, x1), obtained in first frame image using identical technical solution,
As the corresponding matching characteristic point (y3, x3) of characteristic point (y0, x0) in the second frame image, obtain matching characteristic point (y3, x3) with
The offset parameter of characteristic point (y1, x1), the offset parameter are the 4th offset parameter;When the 4th offset parameter reaches
When to preset threshold, show that the deviation of bi-directional matching characteristic point is smaller, successful match;Correspondingly, working as the 4th offset parameter
Not up to preset threshold when, show that the deviation of bi-directional matching characteristic point is larger, it fails to match.Two-way is used in present embodiment
The matching characteristic point that the mode matched obtains, greatly strengthens the robustness of matching characteristic point.
In the present embodiment, equipment is based on corresponding first offset parameter of characteristic point each in second group of N number of characteristic point
It determines multiple matching characteristic points, calculates the second offset parameter of each matching characteristic point in the multiple matching characteristic point, it is described
Second offset parameter characterize the matching characteristic point in second group of N number of characteristic point, it is corresponding with the matching characteristic point
Third feature point between displacement bias degree;It is also understood that second offset parameter characterizes the multiple matching
Each matching characteristic o'clock is relative to the position in first group of N number of characteristic point, between characteristic point corresponding with matching characteristic point in point
Move degrees of offset.For example, in characteristic point N number of for first group, characteristic point (y1, x1), in second group of N number of characteristic point with characteristic point
The identical characteristic point (y2, x2) of (y1, x1) coordinate, calculates the matching characteristic point (y0, x0) of acquisition;Then second offset parameter
Show the relative displacement degrees of offset between matching characteristic point (y0, x0) and characteristic point (y2, x2), it is understood that partially for second
Shifting parameter shows the relative displacement degrees of offset between matching characteristic point (y0, x0) and characteristic point (y1, x1).
It is as an implementation, described to calculate each matching characteristic in the multiple matching characteristic point in the present embodiment
The third offset parameter of point, including:Multiple groups the are extracted from multiple matching characteristic points that the second frame image is included respectively
One characteristic point pair;The fisrt feature point is to including the first matching characteristic point and the second matching characteristic point, and from described first
In frame image, in source characteristic point corresponding with the multiple matching characteristic point extract multiple groups second feature point pair, described second
Characteristic point is to including the first source characteristic point and the second source characteristic point;Wherein, the first matching characteristic point and second matching
Characteristic point is any two characteristic point in the multiple matching characteristic point;Calculate the first matching characteristic point and described second
Second between first distance between matching characteristic point, and calculating first source characteristic point and second source characteristic point
Distance;Obtain the relative parameter between the first distance and the second distance;The relative parameter is denoted as described first
The third offset parameter of matching characteristic point and the second matching characteristic point.
Specifically, equipment is directed to the multiple matching characteristic points and the first frame figure that the second frame image is included respectively
As in, source corresponding with matching characteristic point characteristic point extracted two-by-two;Two in the second frame image extracted
With two source features in characteristic point (the i.e. described first matching characteristic point and the second matching characteristic point) and first frame image
Point (the i.e. described first source characteristic point and second source characteristic point) corresponds respectively.It is special then to calculate separately first matching
First distance between sign point and the second matching characteristic point, and calculate first source characteristic point and second source spy
Second distance between sign point;Wherein, the first distance and the second distance can be Euclidean distance, it is of course also possible to
It is other distances that can characterize the relative positional relationship between corresponding two characteristic points.Further, obtain described first away from
From the relative parameter between the second distance;The relative parameter is denoted as the first matching characteristic point and second described
Third offset parameter with characteristic point;The zoom degree of third offset parameter characterization face region.As a kind of reality
Mode is applied, the relative parameter is specifically as follows the ratio of the first distance and the second distance;Certainly, the opposite ginseng
Number can also be obtained by characterizing other processing modes of relativeness between the first distance and the second distance.
It is as an implementation, described according to second offset parameter and third offset ginseng in the present embodiment
Number determines the corresponding offset parameter of the initial coordinate range, including:By multiple second offset parameters by the first default processing rule
It is then handled, obtains particular offset parameter;And handle multiple relative parameters by the second default processing rule, it obtains
Obtain specific relative parameter;The particular offset degree and the specific relative parameter is corresponding as the initial coordinate range
Offset parameter.
In the present embodiment, each matching characteristic point in the second frame image can be calculated and obtain corresponding the
Two offset parameters, ideally, the corresponding second offset ginseng of all matching characteristic points of the acquisition in the second frame image
Number is equal, but since various factors (such as deviation in the movement of face location, data handling procedure etc.) will cause mistake
Difference, so that corresponding second offset parameter of all matching characteristic points of the acquisition in the second frame image may equal not phases
Deng.In order to make processing result with more robustness, all matching characteristic points pair of the equipment to the acquisition in the second frame image
The second offset parameter answered is ranked up, and selects the intermediate value of N number of second offset parameter as particular offset parameter.It is understood that
For the degrees of offset of the particular offset parameter characterization face frame.On the other hand, in the ideal case, the second frame image
In the corresponding relative parameter of any two matching characteristic point be equal, but due to various factors (such as face location movement,
Deviation in data handling procedure etc.) it will cause error, so that any two in the second frame image match spy
The corresponding relative parameter of sign point may be unequal.In order to make processing result with more robustness, equipment is to the second frame figure
The corresponding relative parameter of any two matching characteristic point as in is ranked up, and selects the intermediate value of multiple relative parameters as specific
Relative parameter.It is to be understood that the zoom degree of the specific relative parameter characterization face frame.
Further, in this embodiment equipment is using the particular offset degree and the specific relative parameter as described in
The corresponding offset parameter of initial coordinate range;To obtain second based on the initial coordinate range and corresponding offset parameter
Coordinate range records the region that second coordinate range is the characterization face position after tracking;It specifically can refer to Fig. 4 b
It is shown, wherein solid box is using the tracking result after the technical solution track human faces of the embodiment of the present invention;And dotted line frame is to adopt
With the directly detected result of face detection scheme;Compared to the technical solution of Face datection characteristic point, using of the invention real
The living body faces tracking scheme for applying example can obtain the variation of more quasi- good description face, specifically can be the change of face location
Change.As an implementation, KLT (Kanade Lucas Tomasi) can be used in the face tracking scheme of the embodiment of the present invention
Tracker is realized, KLT tracker is certainly not limited to, and the living body faces of the embodiment of the present invention can be achieved in other fast iterative algorithms
Tracking;The present embodiment directlys adopt the uniform selected characteristic point in region where characterizing face, and track human faces frame, greatly promotes
Face tracking speed also avoids characteristic point very few the problem of leading to tracking failure.Also, the mode of bi-directional matching is also significantly
Improve the robustness of face tracking.
In the embodiment of the present invention, the textural characteristics extracted in the region;Based on the textural characteristics computational representation
The parameter of posture;Movement is determined based on the parameter, including:Extract the first textural characteristics and/or the second line in the region
Manage feature;Based on the first parameter of the first posture of the first textural characteristics computational representation, and/or, it is based on second texture
Feature calculation characterizes the second parameter of the second posture;The first movement is determined based on first parameter, and/or, based on described the
Two parameters determine the second movement.
For the acquisition modes of the first textural characteristics, as an implementation, first extracted in the region
Textural characteristics, including:Characteristic point in the region is handled according to the default processing rule of third, is obtained in the region
Characterize the first procedure parameter of the difference degree of each characteristic point characteristic point adjacent with the characteristic point;Analyze first mistake
First textural characteristics at journey gain of parameter characterization face texture edge;Correspondingly, being directed to the acquisition modes of the second textural characteristics, institute
The second textural characteristics extracted in the region are stated, including:Extract the first part region in the region;To described first
Characteristic point in subregion is handled according to the 4th default processing rule, is obtained in the first part region and is characterized each spy
Second procedure parameter of the difference degree of the sign point characteristic point adjacent with the characteristic point;Second procedure parameter is analyzed to obtain
Characterize second textural characteristics at eye texture edge.
In the present embodiment, equipment extracts different textural characteristics for different movements, is denoted as first in the present embodiment
Textural characteristics and the second textural characteristics.Classify in the present embodiment mainly for two kinds of movements, it is contemplated that rotary head is acted and blinked
Eye movement work is larger for the attitudes vibration of face, therefore, classifies in the present embodiment for rotary head movement and blink movement.It can
To be interpreted as, first textural characteristics characterize face texture edge, and second textural characteristics characterize eye texture edge.
Specifically, being directed to the first textural characteristics, treatment process includes:To the second frame image, characterization people is extracted
The region (region can be the second area after tracking) of face position, to the region according to the default processing rule of third
Then handled.As an implementation, described that the region is handled according to the default processing rule of third, including:
The region is reduced, such as is contracted to (64,64) range, by the image after diminution according to binary conversion treatment mode or three
Value processing mode is handled.In binary conversion treatment mode as an example, then equipment can be for second in the second frame image
Region carries out local binary patterns (LBP, Local Binary Patterns) processing;Such as:The second frame figure is extracted first
With the matched procedural image of the second area as in, gray proces are carried out to the procedural image, obtain the procedural image
Gray level image, further determine that relatively grey between each characteristic point and eight adjacent characteristic points in the gray level image
Degree relationship multiplies the gray level image of three characteristic point matrix, institute in the gray scale of each characteristic point such as Fig. 4 c for three as illustrated in fig. 4 c
Show;The gray value of each characteristic point is subjected to numeralization expression, specifically can refer to shown in Fig. 4 d.Further, by adjacent eight
The gray scale of characteristic point is compared with the gray scale of central feature point, if the gray scale of adjacent characteristic point is greater than the ash of central feature point
Degree, then be denoted as 1 for the value of the adjacent characteristic point;Conversely, if the gray scale of adjacent characteristic point is less than or equal to the ash of central feature point
Degree, then be denoted as 0 for the value of the adjacent characteristic point, for details, reference can be made to shown in Fig. 4 e.Further, by the value string of adjacent characteristic point
Connection obtains 8 strings of binary characters, and the string of binary characters can be understood as the gray value for being distributed in (0,255).Having
It in body implementation process, can refer to shown in Fig. 4 e, if using first, upper left corner characteristic point as initiation feature point, according to side clockwise
To arrangement, then 8 character strings obtained are 10001111.Thus it can get in the procedural image each characteristic point (in i.e.
Heart characteristic point) corresponding string of binary characters.Above-mentioned treatment process can specifically be realized by following formula:
LBP=[code0, code1 ... ..., code7] (8)
Code (m, n)=Img (y+m, x+n)>Img (y, x)?1:0 (9)
Further, in order to remove redundancy, count in the corresponding string of binary characters of each characteristic point, 0 and 1 variation be less than
2 string of binary characters;For example, character string is in 10001111, first and second 0 and 1 change 1 time, the 4th and the
It five 0 and 1 variation 1 time, amounts to variation twice, is unsatisfactory for the condition of " 0 and 1 variation is less than 2 ".In another example character string is
In 00001111, only changed 1 time by the 4th and the 5th 0 and 1, meets the condition of " 0 and 1 variation is less than 2 ".Then, it will unite
String of binary characters after meter is mapped in (0,58) range, and the data after mapping can be used as LBP data;It can also subtract significantly in this way
Few data processing amount.
Wherein, in expression formula (8) LBP indicate fisrt feature point in the region display parameters and adjacent characteristic point
The relativeness of display parameters;The fisrt feature point is any feature point in the region;Code0, code1 ... ...,
Code7 respectively indicates the display parameters of the adjacent characteristic point of the fisrt feature point;Expression formula (9) is indicated characteristic point (y+m, x
+ n) gray value and the gray value of characteristic point (y, x) be compared, if the gray value of characteristic point (y+m, x+n) is greater than characteristic point
The string of binary characters code (m, n) of characteristic point (m, n) is then denoted as 1, is otherwise denoted as 0 by the gray value of (y, x).
Further, the LBP numerical value that (0,58) is distributed in certain image-region is subjected to statistic histogram processing.Because
The histogram has 59 dimensions, so available one 59 vector tieed up, the first texture as the region after statistics
Feature.
For the second textural characteristics, since the region of eye is smaller, blink movement is also very fast, therefore, is extracting the second line
When managing feature, the extraction of eye region is carried out to region first, that is, extracts the first part region in the region, it is described
First part region can be the upper half area in the region, for example, extracting the upper half/one third in the region
Region, as the first part region.It can refer to the place of aforementioned first textural characteristics to the first part region after extraction
Reason mode is handled, and the second textural characteristics are obtained.As another embodiment, equipment can be directed to the first part region
It carries out three value modes of part (LTP, Local Ternary Pattern) to handle, treatment process is approximate with LBP processing mode, area
It is not, in LBP treatment process, when indicating the relativeness of gray value of characteristic point and adjacent characteristic point, passes through 0 and 1
It is marked.In the LTP treatment process of present embodiment, in the opposite pass for indicating the gray value of characteristic point and adjacent characteristic point
It when being, is marked by 0,1 and -1, specific treatment process can be realized by following formula:
Code (m, n)=Img (y+m, x+n)>Img(y,x)+Fuzzy?1:0 (10)
Code (m, n)=Img (y+m, x+n)<Img(y,x)-Fuzzy?-1:0 (11)
Fuzzy=ratio × Img (y, x) (12)
Wherein, Img (y, x) indicates the gray value of characteristic point (y, x);The characteristic point (y, x) can be three spies for multiplying three
Levy the central feature point in dot matrix;Img (y+m, x+n) indicates the gray value of characteristic point (y+m, x+n);Characteristic point (y+m, x+
It n) is specially the characteristic point adjacent with characteristic point (y, x);Ratio indicates scale parameter, can be pre-configured with;Fuzzy indicates feature
The gray value of point (y, x) and the product of ratio;The gray value of characteristic point (y, x) is bigger, and the value of Fuzzy is also maximum.Expression formula
(10) meaning indicates to be compared the sum of the gray value of the gray value of characteristic point (y+m, x+n) and characteristic point (y, x) and Fuzzy
Compared with, if gray value and Fuzzy the sum of of the gray value of characteristic point (y+m, x+n) greater than characteristic point (y, x), then by characteristic point (m,
N) character string code (m, n) is denoted as 1, is otherwise denoted as 0;The meaning of expression formula (11) is indicated the ash of characteristic point (y+m, x+n)
The difference of the gray value and Fuzzy of angle value and characteristic point (y, x) is compared, if the gray value of characteristic point (y+m, x+n) is greater than spy
The gray value of point (y, x) and the difference of Fuzzy are levied, then the character string code (m, n) of characteristic point (m, n) is denoted as -1, is otherwise denoted as
0。
In the present embodiment, first parameter based on the first posture of the first textural characteristics computational representation, including:It will
First procedure parameter inputs preconfigured first disaggregated model, obtains the first parameter of the first posture of characterization.
Specifically, equipment acquires great amount of samples data in advance, (sample data specifically be can be using above-mentioned processing side
The first textural characteristics that formula obtains) and corresponding posture class indication, classify to the sample data and corresponding posture
Mark carries out machine learning training, obtains the first disaggregated model.After obtaining first textural characteristics, by first texture
Feature inputs first disaggregated model, obtains corresponding first posture of first textural characteristics, first posture is for example
Indicate front or side;First posture can be indicated by the first parameter.In practical applications, first parameter can be passed through
Numerical values recited show that first posture levels off to front or side, such as first parameter is bigger, shows described first
Posture more levels off to front;Correspondingly, if first parameter is smaller, show that first posture more levels off to side.
In the present embodiment, second parameter based on the second posture of the second textural characteristics computational representation, including:It will
Second textural characteristics input preconfigured second disaggregated model, obtain the second parameter of the second posture of characterization.
Specifically, equipment acquires great amount of samples data in advance, (sample data specifically be can be using above-mentioned processing side
The second textural characteristics that formula obtains) and corresponding posture class indication, classify to the sample data and corresponding posture
Mark carries out machine learning training, obtains the second disaggregated model.After obtaining second textural characteristics, by second texture
Feature inputs second disaggregated model, obtains corresponding second posture of second textural characteristics, second posture is for example
It indicates to open eyes or close one's eyes, second posture can be indicated by the second parameter.In practical applications, second parameter can be passed through
Numerical values recited show that second posture levels off to eye opening or eye closing, such as second parameter is bigger, shows described second
Posture more levels off to eye opening;Correspondingly, if second parameter is smaller, show that second posture more levels off to eye closing.
It is described that first movement is determined based on first parameter in the present embodiment, including:Distinguished based on the multiple image
Corresponding multiple first parameters judge whether corresponding first parameter of first part's image is all satisfied first in the multiple image
Whether corresponding first parameter of second part image is not satisfied the first threshold in threshold range and the multiple image
Range;When corresponding first parameter of first part's image is all satisfied first threshold range and described more in the multiple image
When the first threshold range is not satisfied in corresponding first parameter of second part image in frame image, first parameter is determined
Corresponding to the first movement.
Specifically, it is aforementioned obtain characterization the first posture the first parameter when, due to it is aforementioned be for two field pictures carry out
The first parameter obtained is handled, and in the specific application process, it is for multiple image included in the image data obtained
Carry out the first parameter of processing acquisition, it can be understood as, first parameter can get for each frame image, then be directed to multiframe
Image, which can correspond to, obtains multiple first parameters.Based on this, argument sequence can get for the multiple first parameter.Described in analysis
Argument sequence, if numerical value change from low to high, can determine face by flanks transform for front;Correspondingly, if numerical value change by
It is high to Low, then it can determine that face is changed into side by front, it is certainly, opposite, if making the by other data processing methods
One parameter is bigger, shows that face more levels off to front, and the first parameter is smaller, shows that face more levels off to side, then joins in analysis
During Number Sequence, if numerical value change from low to high, can determine that face is changed into side by front;Correspondingly, if numerical value becomes
Change from high to low, then can determine face by flanks transform for front.Based on this, the variation of numerical value in the argument sequence can be passed through
So that it is determined that corresponding first movement.
In the specific implementation process, for the multiple image of acquisition, the X frame image before present frame is chosen, by the X
Frame image uniform cutting is Y sections;Wherein, X and Y is positive integer, and Y is less than X;For every section of image, due in every section of image
Including at least two field pictures, then at least two first parameters can be obtained in every section of image;In at least two first parameter
First parameter of the median as every section of image is chosen, to can get corresponding Y the first parameter of Y sections of images.
Y the first parameter strings are obtained into obtain argument sequence, judge whether the Parameters variation in the argument sequence meets preset rules, example
Such as, if Parameters variation determines corresponding first movement from high to low or from low to high, based on Parameters variation.Another real
It applies in mode, multiple first parameters corresponding for the multiple image determine front portion image packet in the multiple image
The face contained is in front, and the face that rear portion image includes is in side, or determines previous in the multiple image
The face that parts of images includes is in side, and the face that rear portion image includes is in front, then can determine that the first parameter
Corresponding first movement.For example, judging in X frame image, corresponding first parameter of preceding one third image of the X frame image is equal
Meet first threshold range, then show include face direct picture and the X frame image rear one third image it is corresponding
The first parameter first threshold range is not satisfied, then show to include face side image, then can determine in the X frame image
Middle face can determine that rotary head acts by obverting as side.
It is described that second movement is determined based on second parameter in the present embodiment, including:Distinguished based on the multiple image
Corresponding multiple second parameters judge whether corresponding second parameter of Part III image is all satisfied second in the multiple image
Whether corresponding second parameter of Part IV image is not satisfied the second threshold in threshold range and the multiple image
Range;When corresponding second parameter of Part III image is all satisfied second threshold range and described more in the multiple image
When the second threshold range is not satisfied in corresponding second parameter of Part IV image in frame image, second parameter is determined
Corresponding to the second movement.
Specifically, the method for determination acted with first is similarly, in aforementioned the second parameter for obtaining the second posture of characterization, by
In it is aforementioned be the second parameter that processing acquisition is carried out for two field pictures, be for the figure obtained and in the specific application process
The multiple image as included in data carries out the second parameter of processing acquisition, it can be understood as, each frame image can be obtained
Second parameter is obtained, then can be corresponded to for multiple image and obtain multiple second parameters.Based on this, for the multiple second ginseng
The available argument sequence of number.The argument sequence is analyzed, if numerical value change from low to high, can determine that human eye is changed by opening eyes
It closes one's eyes;Correspondingly, if numerical value change from high to low, can determine that human eye is changed into eye opening by closing one's eyes, certainly, opposite, if logical
Crossing other data processing methods makes the second parameter bigger, shows that human eye more levels off to eye opening, the second parameter is smaller, shows face
More level off to eye closing, then during analyzing argument sequence, if numerical value change from low to high, can determine that human eye is changed by closing one's eyes
To open eyes;Correspondingly, if numerical value change from high to low, can determine that human eye is changed into eye closing by opening eyes.Wherein, dynamic due to blinking
Make comparatively fast, the argument sequence can be the second parameter comprising there are two, have described two second parameter energy of discrete state
Enough characteristics presented from low to high or from high to low.Based on this, can by the variation of numerical value in the argument sequence so that it is determined that
Corresponding second movement.
In the specific implementation process, for the multiple image of acquisition, the X frame image before present frame is chosen, by the X
Frame image uniform cutting is Y sections;Wherein, X and Y is positive integer, and Y is less than X;For every section of image, due in every section of image
Including at least two field pictures, then at least two second parameters can be obtained in every section of image;In at least two second parameter
Second parameter of the median as every section of image is chosen, to can get corresponding Y the second parameter of Y sections of images.
Y the second parameter strings are obtained into obtain argument sequence, judge whether the Parameters variation in the argument sequence meets preset rules, example
Such as, if Parameters variation determines corresponding second movement from high to low or from low to high, based on Parameters variation.Another real
It applies in mode, multiple second parameters corresponding for the multiple image determine front portion image packet in the multiple image
The human eye contained is in eyes-open state, and the human eye that rear portion image includes is in closed-eye state, or determines the multiframe figure
The human eye that front portion image includes as in is in closed-eye state, and the human eye that rear portion image includes is in eyes-open state,
It then can determine that corresponding second movement of the second parameter.For example, judging in X frame image, the preceding one third image of the X frame image
Corresponding second parameter is all satisfied second threshold range, then shows that the human eye that image includes is in eyes-open state and the X frame
Second threshold range is not satisfied in corresponding second parameter of rear one third image of image, then shows at human eye that image includes
In closed-eye state, then it can determine the blink movement in the X frame image.
In the present embodiment, since eye closing and the eye opening movement of human eye are excessively rapid, then preconfigured second threshold range
Different from the first threshold range.
In the present embodiment, equipment is mentioned for textural characteristics (including the first textural characteristics and the second textural characteristics) in region
It takes, the textural characteristics are not limited to LBP/LTP feature, are also possible to that other textural characteristics of image change, example can be described
Such as histograms of oriented gradients (HOG, Histogram of Oriented Gradient) feature.In equipment for sample characteristics into
The process of row machine learning training, can be carried out by support vector machines (SVM, Support Vector Machine) learning model
Training classification is not limited to SVM training mode classification, the mode classifications such as other neural networks, linear projection can also be real certainly
It is existing.
Living body verification method described in the present embodiment is not limited to apply in mobile terminal, can also be applied to any occasion
Living body faces verifying, including but not limited to personal computer (PC), server or embedded device etc..Since the present invention is implemented
Example technical solution realize region track by way of substitutes recognition of face detection, blink judgement in without to eyes into
Row positioning;In addition preconfigured computation model only is needed when corresponding to parameter according to textural characteristics calculating posture in scheme
(such as SVM model), and the computation model capacity is minimum (model file can be reduced to 160k), it, can during specific implementation
SVM parameter is directly written in code, it might even be possible to it is interpreted as not needing computation model, it is therefore, provided in an embodiment of the present invention
The algorithm file that living body proof scheme is carried is smaller, meets the memory space requirements of mobile terminal significantly;And data processing
Amount substantially reduces, and also meets the demand of the processing capacity of mobile terminal.
In the present embodiment, after obtaining the first movement and/or the second movement through the above way, first movement is judged
And/or second movement whether movement corresponding with the action command of output matching;If matching is consistent, show described image number
Face in is living body faces rather than preparatory filmed picture or the corresponding face of video.And in the prior art, it emits
Filling living body faces (which can be described as attacking) in a manner of verifying by living body mainly includes:1, normal photo;2, include
The photo of certain class movement;It 3, include the video of certain class movement;4, the video spliced according to action sequence.This makes research staff
It is proposed that different algorithm optimization movements judges performance to cope with different attacks.Existing living body faces verification technique is based on essence
True facial modeling, in mobile terminal side, fast and accurately computation model file required for positioning feature point is excessive,
Processing speed is also not fast enough.By taking the location algorithm of mainstream as an example, display shape recurrence (ESR) algorithm process time can achieve 10
Millisecond, computation model file can reach tens, and active appearance models (AAM) are although the model file of algorithm is only several hundred
K, but the processing time of every frame data is difficult to real-time processing.And what the technical solution of the embodiment of the present invention was tracked by region
Mode substitutes recognition of face detection, positions in blink judgement without to eyes;In addition only according to line in scheme
Reason feature calculation posture needs preconfigured computation model (such as SVM model) when corresponding to parameter, and the computation model capacity
SVM parameter can be directly written in code, or even can by minimum (model file can be reduced to 160k) during specific implementation
To be interpreted as not needing computation model, therefore, the algorithm file that living body proof scheme provided in an embodiment of the present invention is carried compared with
It is small, the memory space requirements of mobile terminal are met significantly;And data processing amount substantially reduces, and also meets the place of mobile terminal
The demand of reason ability.
In order to verify the effectiveness of living body proof scheme provided in an embodiment of the present invention, respectively to different movements, and string
Linkage has carried out the test of attack and true man's experience, specifically can be as shown in table 1.As shown in table 1, attempt to use mobile phone in attacker
When photo, photograph print or video pass through the verifying of authentication system, above-mentioned three kinds of attacks are zero entirely through rate;
It wherein, is zero for the number of pass times of blink movement and rotary head movement in above-mentioned three kinds of attack patterns, so as to table
The bright living body proof scheme using the embodiment of the present invention can effectively identify the void such as cell phone pictures, photograph print or video
The attack of false identity greatly improves the accuracy rate of authentication.
Table 1
Table 2 is actually detected situation signal of the embodiment of the present invention under different testing conditions.As shown in table 2, it detects
Condition may include light condition and test object whether wearing spectacles, and whether the glasses worn have the various items such as frame
Part.Wherein, light condition may include ordinary ray condition, highlighting bar part and dim light condition etc.;Wherein, detection can be passed through
To light intensity parameter decision light condition be to belong to ordinary ray, strong light or half-light line;Can be pre-configured with first threshold and
Second threshold, the first threshold are greater than second threshold;When the light intensity parameter detected is greater than the first threshold, light is determined
Lines part is highlighting bar part;When the light intensity parameter detected is less than the second threshold, determine that light condition is half-light line
Condition;When the light intensity parameter detected is between the first threshold and the second threshold, determine that light condition is normal
Light.On the other hand, whether wearing spectacles and glasses have frame and will affect mentioning for eye textural characteristics test object
Take and blink movement judgement.As can be seen from Table 2, by taking first kind ordinary ray glasses as an example, 23/25 shows 25 times
By 23 times in test, show in the verifying of living body true man, even if user's wearing spectacles, the living body of its living body true man is verified
Percent of pass it is also higher, be not in the unacceptable situation of multiple authentication, the experience of user will not be adversely affected;Its
In, detection averagely needs 1 time, and it is 1 second (s) that detection, which averagely needs the time, every time, and the detection of blink movement averagely needs 2 times, often
Being averaged for secondary detection needs the time for 2s;The detection of rotary head movement averagely needs 2 times, and being averaged of detecting every time needs the time to be
2s。
Table 2
The embodiment of the invention also provides a kind of living bodies to verify equipment.Fig. 5 is that the living body of the embodiment of the present invention verifies equipment
Composed structure schematic diagram;As shown in figure 5, the equipment includes:Detection unit 31, tracking cell 32, feature extraction unit 33,
Computing unit 34, movement judging unit 35 and authentication unit 36;Wherein,
The detection unit 31 parses described image data, identifies institute for obtaining image data based on action command
State the region that face position is characterized in image data;
The tracking cell 32, the multiframe that the described image data for being identified based on the detection unit 31 are included
The variation of face position in image, tracks the region;
The feature extraction unit 33, for extracting the textural characteristics in the region that the tracking cell 32 tracks;
The computing unit 34, the textural characteristics computational representation appearance for being extracted based on the feature extraction unit 33
The parameter of state;
The movement judging unit 35, the parameter for being obtained based on the computing unit 34 determine movement;
The authentication unit 36, for determining living body when movement movement corresponding with action command matching
It is verified.
The tracking cell 32 identifies institute for extracting first frame image and the second frame image in the multiple image
The first area for characterizing face position in first frame image is stated, corresponding first coordinate range in the first area is obtained;
Obtain initial coordinate range corresponding with first coordinate range in the second frame image;Calculate the initial coordinate range
Corresponding offset parameter;The second coordinate range is obtained based on the initial coordinate range and corresponding offset parameter, records institute
State the region that the second coordinate range is the characterization face position after tracking;Wherein, the offset parameter characterization described second
Degrees of offset of the coordinate range relative to first coordinate range.
Further include image acquisition units in the detection unit 31 in the embodiment of the present invention, is acquired by described image single
Member obtains image data;Image data obtained includes multiple image.The detection unit 31 identifies described image data
The region of middle characterization face position, specially:It identifies where characterizing face in the first frame image of described image data
The region of position, thus on the basis of characterizing the region of face position in the first frame image, the tracking cell
32 variations based on face position in the multiple image, track the region.
For the tracking mode of face region, as an implementation, the tracking cell 32, for by default
Step-length chooses first group of N number of characteristic point in the corresponding first area of first coordinate range, obtains first group of N number of spy
First coordinate of fisrt feature point in sign point;N is positive integer;Wherein, the fisrt feature point is first group of N number of characteristic point
In any feature point;Second group of N number of characteristic point in the second frame image is obtained, in second group of N number of characteristic point
Second coordinate of two characteristic points is identical as the first coordinate of corresponding fisrt feature point in first group of N number of characteristic point;Base
The second coordinate of each characteristic point determines initial coordinate range in second group of N number of characteristic point.
In the present embodiment, in order to reduce Face datection load, while splicing video attack is avoided, the embodiment of the present invention uses
Face tracking mode substitutes Face datection.Specifically, the detection unit 31 for described image data first frame image into
Row Face datection, the mode that facial feature points detection specifically can be used identify face in the first frame image, and then really
The first area of face position is characterized in the fixed first frame image;Wherein, the first area can pass through the first coordinate
Range Representation.In the specific application process, the first area passes through the face frame of display in the image-display units of equipment
Mark, such as shown in Fig. 4 a;The face collimation mark knows corresponding coordinate range and matches with first coordinate range.Further
Ground, the first area of the tracking cell 32 in first frame image on the basis of, are tracked.Wherein, described
One frame image can be the first frame image in the image data acquired for a target person;The present embodiment is with front cross frame figure
Be illustrated for the treatment process of picture, certainly, during actual image real time transfer, be for multiple image data into
Row processing, the treatment process of the multiple image data can refer to the place that front cross frame image data is directed to described in the present embodiment
Reason process.
In the present embodiment, the tracking cell 32, for pressing preset step-length in first coordinate range corresponding first
First group of N number of characteristic point is chosen in region, obtains the first coordinate of fisrt feature point in first group of N number of characteristic point;N is positive
Integer;Wherein, the fisrt feature point is any feature point in first group of N number of characteristic point;Obtain the second frame figure
Second group of N number of characteristic point as in, the second coordinate and first group of N of second feature point in second group of N number of characteristic point
The first coordinate of corresponding fisrt feature point is identical in a characteristic point;Based on each feature in second group of N number of characteristic point
Second coordinate of point determines initial coordinate range.
Specifically, for the first area in the first frame image, the tracking cell 32 chooses the by preset step-length
One group of N number of characteristic point obtains the coordinate of each characteristic point in first group of N number of characteristic point.For example, the first area meets
M × n, wherein m is less than or equal to the length/width of the first frame image;Then n be less than or equal to the first frame image width/
Length;It can be then that m × n/ (10 × 10) uniformly choose 100 characteristic points in the first area according to step-length.Wherein, described
Step-length can be not limited to the above-mentioned step-length enumerated, and the step-length can be according to configuring at equal intervals, and the step-length can also be according to unequal interval
Configuration, can specifically be configured according to actual demand.Further, the tracking cell 32 is according in the first frame image
The coordinate of each characteristic point in first group of N number of characteristic point obtains second group of N number of characteristic point in the second frame image, so that described
The coordinate of the coordinate of each characteristic point and corresponding characteristic point in first group of N number of characteristic point in second group of N number of characteristic point
It is identical.For example, if putting in order and identifying, i.e., described first group of N number of characteristic point and second group of N number of feature according to identical
Characteristic point in point is ranked up according to sequence from top to bottom, from left to right, then in second group of N number of characteristic point
The coordinate of one characteristic point is identical as the coordinate of first characteristic point in first group of N number of characteristic point, and described second group N number of
The coordinate of second characteristic point in characteristic point is identical as the coordinate of second characteristic point in first group of N number of characteristic point,
And so on, then in second group of N number of characteristic point the coordinate (i.e. the second coordinate) of any feature point (i.e. second feature point) with
The coordinate (i.e. the first coordinate) of corresponding fisrt feature point is identical in first group of N number of characteristic point.Further, it is based on institute
The second coordinate for stating each characteristic point in second group of N number of characteristic point determines a coordinate range, and the coordinate range is determined as institute
State initial coordinate range.
In the present embodiment, the tracking cell 32 is opposite for calculating each characteristic point in second group of N number of characteristic point
First offset parameter of the individual features point in first group of N number of characteristic point;First offset parameter characterizes coordinate phase
Difference degree between the same second feature point and fisrt feature point;Based on every in second group of N number of characteristic point
Corresponding first offset parameter of a characteristic point determines multiple matching characteristic points, calculates each matching in the multiple matching characteristic point
Second offset parameter of characteristic point;Second offset parameter characterizes the matching characteristic point and second group of N number of characteristic point
In, the degrees of offset between third feature point corresponding with the matching characteristic point;And calculate the multiple matching characteristic
The third offset parameter of each matching characteristic point in point;The third offset parameter characterizes the matching characteristic point and described first
It organizes in N number of characteristic point, the degrees of offset between fourth feature point corresponding with the matching characteristic point;Partially according to described second
Shifting parameter and the third offset parameter determine the corresponding offset parameter of the initial coordinate range.
Wherein, the tracking cell 32, first for each characteristic point in calculating second group of N number of characteristic point are inclined
It before shifting parameter, is handled for the first frame image and the second frame image, obtains the first frame image respectively
First gradient image and the second frame image the second gradient image.Specifically, for the first frame image and described
Second frame image establishes the first pyramid diagram picture of the first frame image respectively, establishes the second gold medal of the second frame image
Word tower image;It calculates the first gradient image of the first pyramid diagram picture, and calculates the of the second pyramid diagram picture
Two gradient images;Wherein, establish image pyramid diagram picture (such as establish the first pyramid diagram picture of the first frame image,
In another example establishing second pyramid diagram picture of the second frame image etc.) it can be understood as downscaled images pari passu, to obtain
The result of more robustness.In the specific implementation process, image can be smoothed first, then to the image after smoothing processing
It is sampled, to can get the pyramid diagram picture of size reduction.In the present embodiment, the first gradient image and described second
Each characteristic point gradient is represented by gradient image:
Gx (y, x)=Img (y, x+1)-Img (y, x-1) (1)
Gy (y, x)=Img (y+1, x)-Img (y-1, x) (2)
Wherein, Gx (y, x) indicates the gradient of characteristic point (y, x) in x-axis;Gy (y, x) indicates characteristic point (y, x) in y-axis
On gradient;Img (y, x) indicates the display parameters of characteristic point (y, x);Correspondingly, Img (y, x+1) indicates characteristic point (y, x+1)
Display parameters;Img (y, x-1) indicates the display parameters of characteristic point (y, x+1);Img (y+1, x) indicates characteristic point (y+1, x)
Display parameters;Img (y-1, x) indicates the display parameters of characteristic point (y-1, x);As an implementation, the display ginseng
Number specifically can be gray value, it is of course also possible to be other display parameters.
Further, in second group of N number of characteristic point each characteristic point relative in first group of N number of characteristic point
The gradient for the individual features point that first offset parameter of individual features point can be obtained based on expression formula (1) and expression formula (2) calculates
It obtains;Wherein, in second group of N number of characteristic point each characteristic point relative to the corresponding spy in first group of N number of characteristic point
Sign point refers to:Characteristic point 2 in second group of N number of characteristic point is relative to the characteristic point in first group of N number of characteristic point
1, characteristic point 2 is identical with coordinate of the characteristic point 1 in the first frame image in the coordinate in the second frame image;Namely institute
It states the first offset parameter and is characterized a little 2 opposite degrees of offset relative to characteristic point 1.Specifically, for second group of N number of spy
Each characteristic point in sign point, analyzes neighbour of each characteristic point relative to the individual features point in first group of N number of characteristic point
Characteristic of field obtains difference parameter (diff) and gradient parameter (grad), is obtained based on the difference parameter and gradient parameter calculating
Obtain the first offset parameter;Wherein, the difference parameter characterizes the characteristic point in second group of N number of characteristic point relative to first group
The difference degree of individual features point in N number of characteristic point.Wherein, the difference parameter and the gradient parameter specifically can by with
Lower expression formula indicates:
Diff=Img1 (y1, x1)-Img2 (y2, x2) (3)
Gx=Gx (y1, x1)+Gx (y2, x2) (4)
Gy=Gy (y1, x1)+Gy (y2, x2) (5)
Wherein, Img1 (y1, x1) indicate in first group of N number of characteristic point, the display parameters of characteristic point (y1, x1);Img2
(y2, x2) indicate in second group of N number of characteristic point, the display parameters of characteristic point (y2, x2);Wherein, the display parameters specifically may be used
To be gray value, it is of course also possible to be other display parameters.Gx indicates gradient in the direction of the x axis;Gy is indicated in y-axis direction
On gradient;Gx (y1, x1) indicates the gradient of characteristic point (y1, x1) in the direction of the x axis in first frame image;Gx (y2, x2) table
Show in the second frame image the gradient of corresponding characteristic point (y2, x2) in the direction of the x axis, characteristic point with characteristic point (y1, x1)
(y1, x1) is identical as the coordinate of characteristic point (y2, x2) in the second needle image in the coordinate in first frame image;Correspondingly, Gy
(y1, x1) indicates the gradient of characteristic point (y1, x1) in the y-axis direction in first frame image;Gy (y2, x2) indicates the second frame image
In with characteristic point (y1, x1) gradient of corresponding characteristic point (y2, x2) in the y-axis direction.
Further, the first offset parameter of individual features point is calculated according to the difference parameter and the gradient parameter,
First offset parameter can be indicated by following formula:
(dy, dx)=f (Gxx, Gxy, Gyy, Diff) (6)
Wherein it is determined that in first frame image characteristic point 1 gradient parameter, be denoted as Gx1 and Gy1;And determine the second frame figure
The gradient parameter of characteristic point 2 corresponding with characteristic point 1, is denoted as Gx2 and Gy2 as in;Then based on (Gx1, Gy1) and (Gx2,
Gy2 matrix operation is carried out) to integrate the gradient of the characteristic point 1 and the characteristic point 2;Then in expression formula (6), Gxx indicates special
The operation result of sign 1 gradient of the gradient and characteristic point 2 in x-axis in x-axis of point;Gyy indicates the ladder of characteristic point 1 on the y axis
The operation result of degree and the gradient of characteristic point 2 on the y axis;Gxy indicates gradient and characteristic point 2 of the characteristic point 1 in x-axis in y-axis
On gradient operation result, or indicate the operation knot of the gradient of gradient and characteristic point 2 in x-axis on the y axis of characteristic point 1
Fruit.F () indicates certain operations rule.
During specific implementation, based on expression formula (6) internal successive ignition, it can get in the first frame image
Matching characteristic point (y0, x0) of one characteristic point (y1, x1) on the second frame image, the matching characteristic point meet following table
Up to formula:
(y0, x0)=(y2+dy, x2+dx) (7)
But since the position of the face in image is in variation, the portion in only described first group of N number of characteristic point
Point characteristic point can find matching characteristic point in the second frame image.
As an implementation, the tracking cell 32, for based on each feature in second group of N number of characteristic point
Corresponding first offset parameter of point determines the corresponding target feature point of each characteristic point, obtains the of the target feature point
Three coordinates;Determine initial characteristics point corresponding with the target feature point in the first frame image;The initial characteristics point
4-coordinate it is identical as the third coordinate of the target feature point;The initial characteristics point is obtained relative to the target signature
The 4th offset parameter between point;When the 4th offset parameter is not up to preset threshold, determine that the target feature point is
Matching characteristic point;When the 4th offset parameter reaches the preset threshold, determining the target feature point not is that matching is special
Sign point.
Specifically, in present embodiment, the tracking cell 32 in obtaining the second frame image by adopting the above technical scheme,
After the corresponding matching characteristic point (y0, x0) of characteristic point (y1, x1) in first frame image, obtained using identical technical solution
In first frame image, as the corresponding matching characteristic point (y3, x3) of characteristic point (y0, x0) in the second frame image, acquisition matching spy
The offset parameter of point (y3, x3) and characteristic point (y1, x1) are levied, the offset parameter is the 4th offset parameter;When described
When 4th offset parameter reaches preset threshold, show that the deviation of bi-directional matching characteristic point is smaller, successful match;Correspondingly, working as institute
When stating the 4th offset parameter and being not up to preset threshold, show that the deviation of bi-directional matching characteristic point is larger, it fails to match.This embodiment party
The matching characteristic point obtained by the way of bi-directional matching in formula, greatly strengthens the robustness of matching characteristic point.
In the present embodiment, the tracking cell 32 is based on each characteristic point corresponding the in second group of N number of characteristic point
One offset parameter determines multiple matching characteristic points, calculates the second offset of each matching characteristic point in the multiple matching characteristic point
Parameter, second offset parameter characterizes in the matching characteristic point and second group of N number of characteristic point and the matching characteristic
Displacement bias degree between the corresponding third feature point of point;It is also understood that described in the second offset parameter characterization
In multiple match points each matching characteristic o'clock relative in first group of N number of characteristic point, characteristic point corresponding with matching characteristic point
Between displacement bias degree.For example, in characteristic point N number of for first group, characteristic point (y1, x1), in second group of N number of characteristic point
The identical characteristic point (y2, x2) with characteristic point (y1, x1) coordinate, calculates the matching characteristic point (y0, x0) of acquisition;Then described second
Offset parameter shows the relative displacement degrees of offset between matching characteristic point (y0, x0) and characteristic point (y2, x2), it is understood that
Show the relative displacement degrees of offset between matching characteristic point (y0, x0) and characteristic point (y1, x1) for the second offset parameter.
In the present embodiment, as an implementation, the tracking cell 32, for respectively from the second frame image institute
Multiple groups fisrt feature point pair is extracted in the multiple matching characteristic points for including;The fisrt feature point is to including the first matching characteristic point
With the second matching characteristic point, and from the first frame image, source characteristic point corresponding with the multiple matching characteristic point
Middle extraction multiple groups second feature point pair, the second feature point is to including the first source characteristic point and the second source characteristic point;Wherein, institute
It states the first matching characteristic point and the second matching characteristic point is any two characteristic point in the multiple matching characteristic point;Meter
The first distance between the first matching characteristic point and the second matching characteristic point is calculated, and calculates first source feature
Second distance between point and second source characteristic point;Obtain the opposite ginseng between the first distance and the second distance
Number;The relative parameter is denoted as to the third offset parameter of the first matching characteristic point and the second matching characteristic point.
Specifically, the tracking cell 32 is directed to the multiple matching characteristic points and institute that the second frame image is included respectively
It states in first frame image, source corresponding with matching characteristic point characteristic point is extracted two-by-two;The the second frame image extracted
In two matching characteristic points (the i.e. described first matching characteristic point and the second matching characteristic point) and first frame image in
Two source characteristic points (the i.e. described first source characteristic point and second source characteristic point) correspond respectively.It then calculates separately described
First distance between first matching characteristic point and the second matching characteristic point, and calculate first source characteristic point and institute
State the second distance between the second source characteristic point;Wherein, the first distance and the second distance can be Euclidean distance, when
So, it is also possible to that other distances of the relative positional relationship between corresponding two characteristic points can be characterized.Further, institute is obtained
State the relative parameter between first distance and the second distance;The relative parameter is denoted as the first matching characteristic point and institute
State the third offset parameter of the second matching characteristic point;The zoom degree of third offset parameter characterization face region.Make
For a kind of embodiment, the relative parameter is specifically as follows the ratio of the first distance and the second distance;Certainly, institute
Stating relative parameter can also be obtained by characterizing other processing modes of relativeness between the first distance and the second distance
?.
In the present embodiment, as an implementation, the tracking cell 32, for multiple second offset parameters to be pressed the
One default processing rule is handled, and particular offset parameter is obtained;And by multiple relative parameters by the second default processing rule
It is handled, obtains specific relative parameter;Using the particular offset degree and the specific relative parameter as the initial seat
Mark the corresponding offset parameter of range.
Corresponding second is obtained partially specifically, can calculate for each matching characteristic point in the second frame image
Shifting parameter, ideally, corresponding second offset parameter of all matching characteristic points of the acquisition in the second frame image are equal
It is equal, but since various factors (such as deviation in the movement of face location, data handling procedure etc.) will cause error, from
And make corresponding second offset parameter of all matching characteristic points of the acquisition in the second frame image may be unequal.For
Make processing result with more robustness, all matching characteristics of the tracking cell 32 to the acquisition in the second frame image
Corresponding second offset parameter of point is ranked up, and selects the intermediate value of N number of second offset parameter as particular offset parameter.It can manage
Xie Wei, the degrees of offset of the particular offset parameter characterization face frame.On the other hand, in the ideal case, the second frame figure
The corresponding relative parameter of any two matching characteristic point as in is equal, but due to various factors (such as the shifting of face location
Dynamic, deviation in data handling procedure etc.) it will cause error, so that any two in the second frame image match
The corresponding relative parameter of characteristic point may be unequal.In order to make processing result with more robustness, the tracking cell 32 is right
The corresponding relative parameter of any two matching characteristic point in the second frame image is ranked up, and selects multiple relative parameters
Intermediate value is as specific relative parameter.It is to be understood that the zoom degree of the specific relative parameter characterization face frame.
Further, in this embodiment the tracking cell 32 is by the particular offset degree and the specific opposite ginseng
Number is as the corresponding offset parameter of the initial coordinate range;To be joined based on the initial coordinate range and corresponding offset
Number obtains the second coordinate range, records the region that second coordinate range is the characterization face position after tracking;Specifically
It can refer to shown in Fig. 4 b, wherein solid box is using the tracking result after the technical solution track human faces of the embodiment of the present invention;And
Dotted line frame is using the directly detected result of Face datection scheme;Compared to the technical solution of Face datection characteristic point, adopt
The variation that more quasi- good description face can be obtained with the living body faces tracking scheme of the embodiment of the present invention, specifically can be face
The variation of position.As an implementation, the realization of KLT tracker can be used in the face tracking scheme of the embodiment of the present invention, when
It so is not limited to KLT tracker, the living body faces tracking of the embodiment of the present invention can be achieved in other fast iterative algorithms;The present embodiment
The uniform selected characteristic point in region where characterizing face, and track human faces frame are directlyed adopt, face tracking speed is greatly improved
Degree also avoids characteristic point very few the problem of leading to tracking failure.Also, the mode of bi-directional matching also greatly improve face with
The robustness of track.
In the present embodiment, the feature extraction unit 33, for extracting the first textural characteristics and/or in the region
Two textural characteristics;
The computing unit 34, for the first parameter based on the first posture of the first textural characteristics computational representation, and/
Or, the second parameter based on the second posture of the second textural characteristics computational representation;
The movement judging unit 35, for determining the first movement based on first parameter, and/or, based on described the
Two parameters determine the second movement.
As an implementation, the feature extraction unit 33, for the characteristic point in the region according to third
Default processing rule is handled, and the difference that each characteristic point characteristic point adjacent with the characteristic point is characterized in the region is obtained
First procedure parameter of off course degree;Analyze the first textural characteristics that first procedure parameter obtains characterization face texture edge;
It is also used to extract the first part region in the region;To the characteristic point in the first part region according to the 4th default place
Reason rule is handled, and is obtained and is characterized each characteristic point characteristic point adjacent with the characteristic point in the first part region
Second procedure parameter of difference degree;Analyze the second texture spy that second procedure parameter obtains characterization eye texture edge
Sign.
In the present embodiment, the feature extraction unit 33 extracts different textural characteristics for different movements, in this reality
It applies and is denoted as the first textural characteristics and the second textural characteristics in example.Classify in the present embodiment mainly for two kinds of movements, considers
It is larger for the attitudes vibration of face to rotary head movement and blink movement, therefore, acts and blink for rotary head in the present embodiment
Movement is classified.It is to be understood that first textural characteristics characterize face texture edge, the second textural characteristics characterization
Eye texture edge.
Specifically, the feature extraction unit 33 is directed to the first textural characteristics, treatment process includes:To second frame
Image extracts the region (region can be the second area after tracking) of characterization face position, to the region
It is handled according to the default processing rule of third.As an implementation, described that the region is handled according to third is default
Rule is handled, including:The region is reduced, such as is contracted to (64,64) range, by the image after diminution according to
Binary conversion treatment mode or three-valued processing mode are handled.In binary conversion treatment mode as an example, then equipment can be for described
Second area in second frame image carries out LBP processing;Such as:Extract first in the second frame image with the second area
Matched procedural image carries out gray proces to the procedural image, obtains the gray level image of the procedural image, further really
The versus grayscale relationship between each characteristic point and eight adjacent characteristic points in the fixed gray level image, as illustrated in fig. 4 c,
Multiply the gray level image of three characteristic point matrix for three, the gray scale of each characteristic point is for example shown in Fig. 4 c;By each characteristic point
Gray value carries out numeralization expression, specifically can refer to shown in Fig. 4 d.Further, by the gray scale of adjacent eight characteristic points and center
The gray scale of characteristic point is compared, will be described adjacent if the gray scale of adjacent characteristic point is more than or equal to the gray scale of central feature point
The value of characteristic point is denoted as 1;Conversely, if the gray scale of adjacent characteristic point is less than the gray scale of central feature point, by the adjacent feature
The value of point is denoted as 0, and for details, reference can be made to shown in Fig. 4 e.Further, the value of adjacent characteristic point is connected to obtain 8 binary words
Symbol string, the string of binary characters can be understood as the gray value for being distributed in (0,255).In the specific implementation process, it can refer to
Shown in Fig. 4 e, if being arranged using first, upper left corner characteristic point as initiation feature point according to clockwise direction, then 8 obtained
Character string be 10001111.Thus can get in the procedural image each characteristic point (i.e. central feature point) corresponding two into
Character string processed.
Further, in order to remove redundancy, count in the corresponding string of binary characters of each characteristic point, 0 and 1 variation be less than
2 string of binary characters;For example, character string is in 10001111, first and second 0 and 1 change 1 time, the 4th and the
It five 0 and 1 variation 1 time, amounts to variation twice, is unsatisfactory for the condition of " 0 and 1 variation is less than 2 ".In another example character string is
In 00001111, only changed 1 time by the 4th and the 5th 0 and 1, meets the condition of " 0 and 1 variation is less than 2 ".Then, it will unite
String of binary characters after meter is mapped in (0,58) range, and the data after mapping can be used as LBP data;It can also subtract significantly in this way
Few data processing amount.Above-mentioned treatment process can specifically be realized by following formula:
LBP=[code0, code1 ... ..., code7] (8)
Code (m, n)=Img (y+m, x+n)>Img (y, x)?1:0 (9)
Wherein, in expression formula (8) LBP indicate fisrt feature point in the region display parameters and adjacent characteristic point
The relativeness of display parameters;The fisrt feature point is any feature point in the region;Code0, code1 ... ...,
Code7 respectively indicates the display parameters of the adjacent characteristic point of the fisrt feature point;Expression formula (9) is indicated characteristic point (y+m, x
+ n) gray value and the gray value of characteristic point (y, x) be compared, if the gray value of characteristic point (y+m, x+n) is greater than characteristic point
The string of binary characters code (m, n) of characteristic point (m, n) is then denoted as 1, is otherwise denoted as 0 by the gray value of (y, x).
Further, the LBP numerical value that (0,58) is distributed in certain image-region is subjected to statistic histogram processing.Because
The histogram has 59 dimensions, so available one 59 vector tieed up, the first texture as the region after statistics
Feature.
For the second textural characteristics, since the region of eye is smaller, blink movement is also very fast, therefore, is extracting the second line
When managing feature, the extraction of eye region is carried out to region first, that is, extracts the first part region in the region, it is described
First part region can be the upper half area in the region, for example, extracting the upper half/one third in the region
Region, as the first part region.It can refer to the place of aforementioned first textural characteristics to the first part region after extraction
Reason mode is handled, and the second textural characteristics are obtained.As another embodiment, equipment can be directed to the first part region
Carry out LTP processing, treatment process is approximate with LBP processing mode, and difference is, in LBP treatment process, expression characteristic point and
When the relativeness of the gray value of adjacent characteristic point, it is marked by 0 and 1.In the LTP treatment process of present embodiment,
When indicating the relativeness of gray value of characteristic point and adjacent characteristic point, it is marked by 0,1 and -1, it is specific processed
Journey can be realized by following formula:
Code (m, n)=Img (y+m, x+n)>Img(y,x)+Fuzzy?1:0 (10)
Code (m, n)=Img (y+m, x+n)<Img(y,x)-Fuzzy?-1:0 (11)
Fuzzy=ratio × Img (y, x) (12)
Wherein, Img (y, x) indicates the gray value of characteristic point (y, x);The characteristic point (y, x) can be three spies for multiplying three
Levy the central feature point in dot matrix;Img (y+m, x+n) indicates the gray value of characteristic point (y+m, x+n);Characteristic point (y+m, x+
It n) is specially the characteristic point adjacent with characteristic point (y, x);Ratio indicates scale parameter, can be pre-configured with;Fuzzy indicates feature
The gray value of point (y, x) and the product of ratio;The gray value of characteristic point (y, x) is bigger, and the value of Fuzzy is also maximum.Expression formula
(10) meaning indicates to be compared the sum of the gray value of the gray value of characteristic point (y+m, x+n) and characteristic point (y, x) and Fuzzy
Compared with, if gray value and Fuzzy the sum of of the gray value of characteristic point (y+m, x+n) greater than characteristic point (y, x), then by characteristic point (m,
N) character string code (m, n) is denoted as 1, is otherwise denoted as 0;The meaning of expression formula (11) is indicated the ash of characteristic point (y+m, x+n)
The difference of the gray value and Fuzzy of angle value and characteristic point (y, x) is compared, if the gray value of characteristic point (y+m, x+n) is greater than spy
The gray value of point (y, x) and the difference of Fuzzy are levied, then the character string code (m, n) of characteristic point (m, n) is denoted as -1, is otherwise denoted as
0。
In the present embodiment, as an implementation, the computing unit 34, for inputting first textural characteristics
Preconfigured first disaggregated model obtains the first parameter of the first posture of characterization;And/or for second texture is special
Sign inputs preconfigured second disaggregated model, obtains the second parameter of the second posture of characterization.
Specifically, acquiring great amount of samples data in equipment in advance, (sample data specifically be can be using above-mentioned processing
The first textural characteristics that mode obtains) and corresponding posture class indication, to the sample data and corresponding posture point
Class mark carries out machine learning training, obtains the first disaggregated model.After obtaining first textural characteristics, by first line
It manages feature and inputs first disaggregated model, obtain corresponding first posture of first textural characteristics, the first posture example
Such as indicate front or side;First posture can be indicated by the first parameter.In practical applications, can join by described first
Several numerical values reciteds shows that first posture levels off to front or side, such as first parameter is bigger, shows described the
One posture more levels off to front;Correspondingly, if first parameter is smaller, show that first posture more levels off to side.Separately
On the one hand, equipment acquire in advance great amount of samples data (sample data specifically can be using above-mentioned processing mode obtain
Second textural characteristics) and corresponding posture class indication, the sample data and corresponding posture class indication are carried out
Machine learning training, obtains the second disaggregated model.After obtaining second textural characteristics, second textural characteristics are inputted
Second disaggregated model, obtains corresponding second posture of second textural characteristics, and second posture for example indicates to open eyes
Or close one's eyes, second posture can be indicated by the second parameter.It in practical applications, can be big by the numerical value of second parameter
It is small to show that second posture levels off to eye opening or eye closing, such as second parameter is bigger, shows that second posture more becomes
It is bordering on eye opening;Correspondingly, if second parameter is smaller, show that second posture more levels off to eye closing.
In the present embodiment, as an implementation, the movement judging unit 35, for based on the multiple image point
Not corresponding multiple first parameters judge whether corresponding first parameter of first part's image is all satisfied in the multiple image
Whether corresponding first parameter of second part image is not satisfied first threshold in one threshold range and the multiple image
It is worth range;When corresponding first parameter of first part's image is all satisfied first threshold range and described in the multiple image
When the first threshold range is not satisfied in corresponding first parameter of second part image in multiple image, first ginseng is determined
Number corresponds to the first movement;And/or for being based on corresponding multiple second parameters of the multiple image, judge described more
Whether corresponding second parameter of Part III image is all satisfied in second threshold range and the multiple image in frame image
Whether corresponding second parameter of four parts of images is not satisfied the second threshold range;When Part III in the multiple image
Corresponding second parameter of image is all satisfied Part IV image corresponding second in second threshold range and the multiple image
When the second threshold range is not satisfied in parameter, determine that second parameter corresponds to the second movement.
Specifically, it is aforementioned obtain characterization the first posture the first parameter when, due to it is aforementioned be for two field pictures carry out
The first parameter obtained is handled, and in the specific application process, it is for multiple image included in the image data obtained
Carry out the first parameter of processing acquisition, it can be understood as, first parameter can get for each frame image, then be directed to multiframe
Image, which can correspond to, obtains multiple first parameters.Based on this, argument sequence can get for the multiple first parameter.Described in analysis
Argument sequence, if numerical value change from low to high, can determine face by flanks transform for front;Correspondingly, if numerical value change by
It is high to Low, then it can determine that face is changed into side by front, it is certainly, opposite, if making the by other data processing methods
One parameter is bigger, shows that face more levels off to front, and the first parameter is smaller, shows that face more levels off to side, then joins in analysis
During Number Sequence, if numerical value change from low to high, can determine that face is changed into side by front;Correspondingly, if numerical value becomes
Change from high to low, then can determine face by flanks transform for front.Based on this, the variation of numerical value in the argument sequence can be passed through
So that it is determined that corresponding first movement.
In the specific implementation process, for the multiple image of acquisition, the X frame image before present frame is chosen, by the X
Frame image uniform cutting is Y sections;Wherein, X and Y is positive integer, and Y is less than X;For every section of image, due in every section of image
Including at least two field pictures, then at least two first parameters can be obtained in every section of image;In at least two first parameter
First parameter of the median as every section of image is chosen, to can get corresponding Y the first parameter of Y sections of images.
Y the first parameter strings are obtained into obtain argument sequence, judge whether the Parameters variation in the argument sequence meets preset rules, example
Such as, if Parameters variation determines corresponding first movement from high to low or from low to high, based on Parameters variation.Another real
It applies in mode, multiple first parameters corresponding for the multiple image determine front portion image packet in the multiple image
The face contained is in front, and the face that rear portion image includes is in side, or determines previous in the multiple image
The face that parts of images includes is in side, and the face that rear portion image includes is in front, then can determine that the first parameter
Corresponding first movement.For example, judging in X frame image, corresponding first parameter of preceding one third image of the X frame image is equal
Meet first threshold range, then show include face direct picture and the X frame image rear one third image it is corresponding
The first parameter first threshold range is not satisfied, then show to include face side image, then can determine in the X frame image
Middle face can determine that rotary head acts by obverting as side.
In the present embodiment, detection unit 31, tracking cell 32, feature extraction unit 33 in living body verifying equipment,
Computing unit 34, movement judging unit 35 and authentication unit 36, in practical applications can be by the central processing in the equipment
Device (CPU, Central Processing Unit), digital signal processor (DSP, Digital Signal Processor),
Micro-control unit (MCU, Microcontroller Unit) or programmable gate array (FPGA, Field-Programmable
Gate Array) it realizes.
The embodiment of the invention also provides a kind of living bodies to verify equipment, and living body verifies equipment as one example of hardware entities
As shown in Figure 6.The equipment includes processor 61, storage medium 62, camera 65 and at least one external communication interface 63;
The processor 61, storage medium 62, camera 65 and external communication interface 63 are connected by bus 64.
The living body verification method of the embodiment of the present invention can be integrated in institute by the algorithms library form of algorithm and arbitrary format
It states in living body verifying equipment;It can specifically be integrated in the client that can be run in the living body verifying equipment.In practical applications,
Algorithm can be packaged together with client, when user activates client, i.e. unlatching living body authentication function, client call algorithm
Library, and start camera, the image data acquired by camera is acted as source data according to the source data of acquisition
Determine.
It need to be noted that be:Above is referred to the description of equipment item, be with above method description it is similar, with method
Beneficial effect description, does not repeat them here.For undisclosed technical detail in present device embodiment, the method for the present invention is please referred to
The description of embodiment.
It should be understood by those skilled in the art that, the embodiment of the present invention can provide as method, equipment or computer program
Product.Therefore, the shape of hardware embodiment, software implementation or embodiment combining software and hardware aspects can be used in the present invention
Formula.Moreover, the present invention, which can be used, can use storage in the computer that one or more wherein includes computer usable program code
The form for the computer program product implemented on medium (including but not limited to magnetic disk storage and optical memory etc.).
The present invention be referring to according to the method for the embodiment of the present invention, the flow chart of equipment and computer program product and/or
Block diagram describes.It should be understood that each process that can be realized by computer program instructions in flowchart and/or the block diagram and/or
The combination of process and/or box in box and flowchart and/or the block diagram.It can provide these computer program instructions to arrive
General purpose computer, special purpose computer, Embedded Processor or other programmable data processing devices processor to generate one
Machine, so that being generated by the instruction that computer or the processor of other programmable data processing devices execute for realizing flowing
The device for the function of being specified in journey figure one process or multiple processes and/or block diagrams one box or multiple boxes.
These computer program instructions, which may also be stored in, is able to guide computer or other programmable data processing devices with spy
Determine in the computer-readable memory that mode works, so that it includes referring to that instruction stored in the computer readable memory, which generates,
Enable the manufacture of device, the command device realize in one box of one or more flows of the flowchart and/or block diagram or
The function of being specified in multiple boxes.
These computer program instructions also can be loaded onto a computer or other programmable data processing device, so that counting
Series of operation steps are executed on calculation machine or other programmable devices to generate computer implemented processing, thus in computer or
The instruction executed on other programmable devices is provided for realizing in one or more flows of the flowchart and/or block diagram one
The step of function of being specified in a box or multiple boxes.
The foregoing is only a preferred embodiment of the present invention, is not intended to limit the scope of the present invention.
Claims (20)
1. a kind of living body verification method, which is characterized in that the method includes:
Image data is obtained based on action command, parses described image data, identifies characterization face institute in described image data
In the region of position;
The variation of face position in the multiple image for being included based on described image data is uniformly chosen in this region
Characteristic point tracks the region using preset track algorithm;
Characteristic point in the region is handled according to the default processing rule of third, obtains and characterizes each spy in the region
First procedure parameter of the difference degree of the sign point characteristic point adjacent with the characteristic point;First procedure parameter is analyzed to obtain
Characterize first textural characteristics at face texture edge;And/or
Extract the first part region in the region;To the characteristic point in the first part region according to the 4th default processing
Rule is handled, and the difference that each characteristic point characteristic point adjacent with the characteristic point is characterized in the first part region is obtained
Second procedure parameter of off course degree;Analyze the second textural characteristics that second procedure parameter obtains characterization eye texture edge;
Based on the first parameter of the first posture of the first textural characteristics computational representation, and/or, it is based on second textural characteristics
Second parameter of the second posture of computational representation;
The first movement is determined based on first parameter, and/or, the second movement is determined based on second parameter;
When first movement and/or the second movement movement corresponding with action command matching, determine that living body verifying is logical
It crosses.
2. the method according to claim 1, wherein the multiple image for being included based on described image data
The variation of middle face position tracks the region, including:
The first frame image and the second frame image in the multiple image are extracted, identifies characterization face institute in the first frame image
In the first area of position, corresponding first coordinate range in the first area is obtained;
Obtain initial coordinate range corresponding with first coordinate range in the second frame image;
Calculate the corresponding offset parameter of the initial coordinate range;
The second coordinate range is obtained based on the initial coordinate range and corresponding offset parameter, records the second coordinate model
The region of characterization face position after enclosing for tracking;
Wherein, the offset parameter characterizes degrees of offset of second coordinate range relative to first coordinate range.
3. according to the method described in claim 2, it is characterized in that, described obtain in the second frame image is sat with described first
The corresponding initial coordinate range of range is marked, including:
First group of N number of characteristic point is chosen in the corresponding first area of first coordinate range by preset step-length, described in acquisition
First coordinate of fisrt feature point in first group of N number of characteristic point;N is positive integer;Wherein, the fisrt feature point is described first
Any feature point in the N number of characteristic point of group;
Second group of N number of characteristic point in the second frame image is obtained, the of second feature point in second group of N number of characteristic point
Two coordinates are identical as the first coordinate of corresponding fisrt feature point in first group of N number of characteristic point;
Initial coordinate range is determined based on the second coordinate of each characteristic point in second group of N number of characteristic point.
4. according to the method described in claim 3, it is characterized in that, described calculate the corresponding offset ginseng of the initial coordinate range
Number, including:
Each characteristic point is calculated in second group of N number of characteristic point relative to the individual features in first group of N number of characteristic point
First offset parameter of point;The identical second feature point of the first offset parameter characterization coordinate and the fisrt feature point
Between difference degree;
Multiple matching characteristic points are determined based on corresponding first offset parameter of characteristic point each in second group of N number of characteristic point,
Calculate the second offset parameter of each matching characteristic point in the multiple matching characteristic point;Described in the second offset parameter characterization
Matching characteristic point with it is inclined in second group of N number of characteristic point, between third feature point corresponding with the matching characteristic point
Shifting degree;
And calculate the third offset parameter of each matching characteristic point in the multiple matching characteristic point;The third offset ginseng
Number characterize the matching characteristic points in first group of N number of characteristic point, fourth feature corresponding with the matching characteristic point
Degrees of offset between point;
The corresponding offset parameter of the initial coordinate range is determined according to second offset parameter and the third offset parameter.
5. according to the method described in claim 4, it is characterized in that, described calculate each matching in the multiple matching characteristic point
The third offset parameter of characteristic point, including:
Multiple groups fisrt feature point pair is extracted from multiple matching characteristic points that the second frame image is included respectively;Described first
Characteristic point to include the first matching characteristic point and the second matching characteristic point, and from the first frame image, with it is the multiple
Multiple groups second feature point pair is extracted in the corresponding source characteristic point of matching characteristic point, the second feature point is to special including the first source
Sign point and the second source characteristic point;Wherein, the first matching characteristic point and the second matching characteristic point are the multiple matching
Any two characteristic point in characteristic point;
The first distance between the first matching characteristic point and the second matching characteristic point is calculated, and calculates described first
Second distance between source characteristic point and second source characteristic point;
Obtain the relative parameter between the first distance and the second distance;The relative parameter is denoted as described first
Third offset parameter with characteristic point and the second matching characteristic point.
6. according to the method described in claim 5, it is characterized in that, described inclined according to second offset parameter and the third
Shifting parameter determines the corresponding offset parameter of the initial coordinate range, including:
Multiple second offset parameters are handled by the first default processing rule, obtain particular offset parameter;
And handle multiple relative parameters by the second default processing rule, obtain specific relative parameter;
Using the particular offset parameter and the specific relative parameter as the corresponding offset parameter of the initial coordinate range.
7. according to the method described in claim 4, it is characterized in that, described based on each spy in second group of N number of characteristic point
Corresponding first offset parameter of sign point determines multiple matching characteristic points, including:
Each characteristic point pair is determined based on corresponding first offset parameter of characteristic point each in second group of N number of characteristic point
The target feature point answered obtains the third coordinate of the target feature point;
Determine initial characteristics point corresponding with the target feature point in the first frame image;The of the initial characteristics point
4-coordinate is identical as the third coordinate of the target feature point;
The initial characteristics point is obtained relative to the 4th offset parameter between the target feature point;
When the 4th offset parameter reaches preset threshold, determine that the target feature point is matching characteristic point;
When the 4th offset parameter is not up to the preset threshold, determining the target feature point not is matching characteristic point.
8. the method according to claim 1, wherein described be based on the first textural characteristics computational representation first
First parameter of posture, including:
First textural characteristics are inputted into preconfigured first disaggregated model, obtain the first parameter of the first posture of characterization.
9. the method according to claim 1, wherein described be based on the second textural characteristics computational representation second
Second parameter of posture, including:
Second textural characteristics are inputted into preconfigured second disaggregated model, obtain the second parameter of the second posture of characterization.
10. being wrapped the method according to claim 1, wherein described determine the first movement based on first parameter
It includes:
Based on corresponding multiple first parameters of the multiple image, judge that first part's image is corresponding in the multiple image
The first parameter whether be all satisfied corresponding first parameter of second part image in first threshold range and the multiple image
Whether the first threshold range is not satisfied;
When corresponding first parameter of first part's image is all satisfied first threshold range and the multiframe in the multiple image
When the first threshold range is not satisfied in corresponding first parameter of second part image in image, first parameter pair is determined
It should be in the first movement.
11. being wrapped the method according to claim 1, wherein described determine the second movement based on second parameter
It includes:
Based on corresponding multiple second parameters of the multiple image, judge that Part III image is corresponding in the multiple image
The second parameter whether be all satisfied corresponding second parameter of Part IV image in second threshold range and the multiple image
Whether the second threshold range is not satisfied;
When corresponding second parameter of Part III image is all satisfied second threshold range and the multiframe in the multiple image
When the second threshold range is not satisfied in corresponding second parameter of Part IV image in image, second parameter pair is determined
It should be in the second movement.
12. a kind of living body verifies equipment, which is characterized in that the equipment includes:Detection unit, tracking cell, feature extraction list
Member, computing unit, movement judging unit and authentication unit;Wherein,
The detection unit parses described image data, identifies described image for obtaining image data based on action command
The region of face position is characterized in data;
The tracking cell, people in the multiple image that the described image data for being identified based on the detection unit are included
The variation of face position, uniform selected characteristic point, tracks the region using preset track algorithm in this region;
The feature extraction unit is obtained for being handled according to the default processing rule of third the characteristic point in the region
Obtain the first procedure parameter that the difference degree of each characteristic point characteristic point adjacent with the characteristic point is characterized in the region;Point
Analyse the first textural characteristics that first procedure parameter obtains characterization face texture edge;It is also used to extract in the region
A part of region;Characteristic point in the first part region is handled according to the 4th default processing rule, described in acquisition
The second procedure parameter of the difference degree of each characteristic point characteristic point adjacent with the characteristic point is characterized in first part region;
Analyze the second textural characteristics that second procedure parameter obtains characterization eye texture edge;
The computing unit, for the first parameter based on the first posture of the first textural characteristics computational representation, and/or, base
In the second parameter of the second posture of the second textural characteristics computational representation;
The movement judging unit, for determining the first movement based on first parameter, and/or, it is based on second parameter
Determine the second movement;
The authentication unit, for when first movement and/or the second movement movement matching corresponding with the action command
When, determine that living body is verified.
13. equipment according to claim 12, which is characterized in that the tracking cell, for extracting the multiple image
In first frame image and the second frame image, identify in the first frame image characterize face position first area, obtain
Obtain corresponding first coordinate range in the first area;It obtains corresponding with first coordinate range in the second frame image
Initial coordinate range;Calculate the corresponding offset parameter of the initial coordinate range;Based on the initial coordinate range and correspondence
Offset parameter obtain the second coordinate range, record second coordinate range be tracking after characterize face position area
Domain;Wherein, the offset parameter characterizes degrees of offset of second coordinate range relative to first coordinate range.
14. equipment according to claim 13, which is characterized in that the tracking cell, for pressing preset step-length described
First group of N number of characteristic point is chosen in the corresponding first area of first coordinate range, is obtained first in first group of N number of characteristic point
First coordinate of characteristic point;N is positive integer;Wherein, the fisrt feature point is any spy in first group of N number of characteristic point
Sign point;Second group of N number of characteristic point in the second frame image is obtained, second feature point in second group of N number of characteristic point
Second coordinate is identical as the first coordinate of corresponding fisrt feature point in first group of N number of characteristic point;Based on described second
The second coordinate of each characteristic point determines initial coordinate range in the N number of characteristic point of group.
15. equipment according to claim 14, which is characterized in that the tracking cell, it is N number of for calculating described second group
First offset parameter of each characteristic point relative to the individual features point in first group of N number of characteristic point in characteristic point;It is described
First offset parameter characterizes the difference degree between the identical second feature point of coordinate and fisrt feature point;Based on institute
It states corresponding first offset parameter of each characteristic point in second group of N number of characteristic point and determines multiple matching characteristic points, calculate described more
Second offset parameter of each matching characteristic point in a matching characteristic point;Second offset parameter characterizes the matching characteristic point
With the degrees of offset in second group of N number of characteristic point, between third feature point corresponding with the matching characteristic point;With
And calculate the third offset parameter of each matching characteristic point in the multiple matching characteristic point;The third offset parameter characterization
The matching characteristic point in first group of N number of characteristic point, between fourth feature point corresponding with the matching characteristic point
Degrees of offset;Determine that the initial coordinate range is corresponding partially according to second offset parameter and the third offset parameter
Shifting parameter.
16. equipment according to claim 15, which is characterized in that the tracking cell, for respectively from second frame
Multiple groups fisrt feature point pair is extracted in multiple matching characteristic points that image is included;The fisrt feature point is to including the first matching
Characteristic point and the second matching characteristic point, and from the first frame image, source corresponding with the multiple matching characteristic point
Multiple groups second feature point pair is extracted in characteristic point, the second feature point is to including the first source characteristic point and the second source characteristic point;
Wherein, the first matching characteristic point and the second matching characteristic point are that any two in the multiple matching characteristic point are special
Sign point;It calculates the first distance between the first matching characteristic point and the second matching characteristic point, and calculates described the
Second distance between one source characteristic point and second source characteristic point;It obtains between the first distance and the second distance
Relative parameter;The relative parameter is denoted as to the third offset of the first matching characteristic point and the second matching characteristic point
Parameter.
17. equipment according to claim 16, which is characterized in that the tracking cell, for multiple second offsets to be joined
Number is handled by the first default processing rule, obtains particular offset parameter;And by multiple relative parameters by the second default place
Reason rule is handled, and specific relative parameter is obtained;Using the particular offset parameter and the specific relative parameter as described in
The corresponding offset parameter of initial coordinate range.
18. equipment according to claim 15, which is characterized in that the tracking cell, for N number of based on described second group
Corresponding first offset parameter of each characteristic point determines the corresponding target feature point of each characteristic point in characteristic point, obtains institute
State the third coordinate of target feature point;Determine initial characteristics corresponding with the target feature point in the first frame image
Point;The 4-coordinate of the initial characteristics point is identical as the third coordinate of the target feature point;Obtain the initial characteristics point
Relative to the 4th offset parameter between the target feature point;When the 4th offset parameter reaches preset threshold, determine
The target feature point is matching characteristic point;When the 4th offset parameter is not up to the preset threshold, the mesh is determined
Marking characteristic point is not matching characteristic point.
19. equipment according to claim 18, which is characterized in that the computing unit, for first texture is special
Sign inputs preconfigured first disaggregated model, obtains the first parameter of the first posture of characterization;And/or it is used for described second
Textural characteristics input preconfigured second disaggregated model, obtain the second parameter of the second posture of characterization.
20. equipment according to claim 18, which is characterized in that the movement judging unit, for being based on the multiframe
Corresponding multiple first parameters of image judge whether corresponding first parameter of first part's image is equal in the multiple image
Meet corresponding first parameter of second part image in first threshold range and the multiple image whether be not satisfied it is described
First threshold range;When corresponding first parameter of first part's image is all satisfied first threshold range, simultaneously in the multiple image
And when the first threshold range is not satisfied in corresponding first parameter of second part image in the multiple image, described in determination
First parameter corresponds to the first movement;And/or for being based on corresponding multiple second parameters of the multiple image, judgement
Whether corresponding second parameter of Part III image is all satisfied second threshold range and the multiframe figure in the multiple image
Whether corresponding second parameter of Part IV image is not satisfied the second threshold range as in;When in the multiple image
It is corresponding that corresponding second parameter of three parts image is all satisfied Part IV image in second threshold range and the multiple image
The second parameter when the second threshold range is not satisfied, determine that second parameter corresponds to the second movement.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710533353.1A CN107316029B (en) | 2017-07-03 | 2017-07-03 | A kind of living body verification method and equipment |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710533353.1A CN107316029B (en) | 2017-07-03 | 2017-07-03 | A kind of living body verification method and equipment |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107316029A CN107316029A (en) | 2017-11-03 |
CN107316029B true CN107316029B (en) | 2018-11-23 |
Family
ID=60181058
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710533353.1A Active CN107316029B (en) | 2017-07-03 | 2017-07-03 | A kind of living body verification method and equipment |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107316029B (en) |
Families Citing this family (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108021892B (en) * | 2017-12-06 | 2021-11-19 | 上海师范大学 | Human face living body detection method based on extremely short video |
KR102374747B1 (en) * | 2017-12-15 | 2022-03-15 | 삼성전자주식회사 | Method and device to recognize object |
CN108304708A (en) * | 2018-01-31 | 2018-07-20 | 广东欧珀移动通信有限公司 | Mobile terminal, face unlocking method and related product |
CN110688878B (en) * | 2018-07-06 | 2021-05-04 | 北京三快在线科技有限公司 | Living body identification detection method, living body identification detection device, living body identification detection medium, and electronic device |
CN109886080A (en) * | 2018-12-29 | 2019-06-14 | 深圳云天励飞技术有限公司 | Human face in-vivo detection method, device, electronic equipment and readable storage medium storing program for executing |
CN110276313B (en) * | 2019-06-25 | 2022-04-22 | 杭州网易智企科技有限公司 | Identity authentication method, identity authentication device, medium and computing equipment |
CN111860394A (en) * | 2020-07-28 | 2020-10-30 | 成都新希望金融信息有限公司 | Gesture estimation and gesture detection-based action living body recognition method |
CN112699857A (en) * | 2021-03-24 | 2021-04-23 | 北京远鉴信息技术有限公司 | Living body verification method and device based on human face posture and electronic equipment |
CN116152936A (en) * | 2023-02-17 | 2023-05-23 | 深圳市永腾翼科技有限公司 | Face identity authentication system with interactive living body detection and method thereof |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102402691A (en) * | 2010-09-08 | 2012-04-04 | 中国科学院自动化研究所 | Method for tracking gestures and actions of human face |
CN105868733A (en) * | 2016-04-21 | 2016-08-17 | 腾讯科技(深圳)有限公司 | Face in-vivo validation method and device |
CN106599772A (en) * | 2016-10-31 | 2017-04-26 | 北京旷视科技有限公司 | Living body authentication method, identity authentication method and device |
-
2017
- 2017-07-03 CN CN201710533353.1A patent/CN107316029B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102402691A (en) * | 2010-09-08 | 2012-04-04 | 中国科学院自动化研究所 | Method for tracking gestures and actions of human face |
CN105868733A (en) * | 2016-04-21 | 2016-08-17 | 腾讯科技(深圳)有限公司 | Face in-vivo validation method and device |
CN106599772A (en) * | 2016-10-31 | 2017-04-26 | 北京旷视科技有限公司 | Living body authentication method, identity authentication method and device |
Also Published As
Publication number | Publication date |
---|---|
CN107316029A (en) | 2017-11-03 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107316029B (en) | A kind of living body verification method and equipment | |
CN111340008B (en) | Method and system for generation of counterpatch, training of detection model and defense of counterpatch | |
CN106897658B (en) | Method and device for identifying human face living body | |
JP5010905B2 (en) | Face recognition device | |
CN108229330A (en) | Face fusion recognition methods and device, electronic equipment and storage medium | |
CN106022317A (en) | Face identification method and apparatus | |
CN110222573B (en) | Face recognition method, device, computer equipment and storage medium | |
US20230033052A1 (en) | Method, apparatus, device, and storage medium for training image processing model | |
WO2016084072A1 (en) | Anti-spoofing system and methods useful in conjunction therewith | |
CN107194361A (en) | Two-dimentional pose detection method and device | |
CN107844742B (en) | Facial image glasses minimizing technology, device and storage medium | |
CN105335719A (en) | Living body detection method and device | |
CN109034095A (en) | A kind of face alignment detection method, apparatus and storage medium | |
CN111414858B (en) | Face recognition method, target image determining device and electronic system | |
CN110428399A (en) | Method, apparatus, equipment and storage medium for detection image | |
Smith-Creasey et al. | Continuous face authentication scheme for mobile devices with tracking and liveness detection | |
CN107463865A (en) | Face datection model training method, method for detecting human face and device | |
CN108875474A (en) | Assess the method, apparatus and computer storage medium of face recognition algorithms | |
US11620854B2 (en) | Evaluating the security of a facial recognition system using light projections | |
KR20200029659A (en) | Method and apparatus for face recognition | |
WO2020195732A1 (en) | Image processing device, image processing method, and recording medium in which program is stored | |
CN109871845A (en) | Certificate image extracting method and terminal device | |
CN111382791B (en) | Deep learning task processing method, image recognition task processing method and device | |
CN111784658A (en) | Quality analysis method and system for face image | |
CN107369086A (en) | A kind of identity card stamp system and method |
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 |