The content of the invention
The present invention is directed to the problems referred to above, there is provided a kind of face information identifying system, including face detection module, face are matched somebody with somebody
Quasi-mode block, recognition of face confirm module;The face detection module detects the information of face, and passes to face with quasi-mode
Block;Face registration module carries out registration according to face information;Recognition of face is passed to after the completion of registration and confirms that module is carried out really
Recognize.
The recognition methods of the face information identifying system is comprised the following steps:
S1, by face recognition technology Quick Acquisition information;
By processing, S2, information recognize which uses people;
S3, start information process, storage information and pushed information.
Step S1 is specially:The information of human face characteristic point is gathered by face recognition technology, including gathers the information.
Step S2 includes face information collection and processes:
Face information is gathered:
The response diagram of each Feature point correspondence is generated by calculating characteristic point surrounding neighbors texture information;Retouched using overall face
Textural characteristics are stated, by the face texture unrelated with shape, and base on human face characteristic point evolution to standard shape, will be obtained
It is modeled in the pca method face texture unrelated to shape;
Face information process:
Using Cascade CNN, comprising three-level, per grade includes multiple convolutional networks, and the first order provides an initial point position and estimates
Count, on this basis rear two-stage intense adjustment characteristic point position;Instruction of the multitask with brigadier's registration face character related to other
Practicing is carried out simultaneously, and the attribute related to facial feature points includes head pose, expression.
Advantages of the present invention:
The present invention is by recognition of face, automatically quick to recognize, filter out oneself setting information and others be pushed to the letter of oneself
Breath.Ensure that privacy, security and the accuracy of information.
In addition to objects, features and advantages described above, the present invention also has other objects, features and advantages.
Below with reference to figure, the present invention is further detailed explanation.
Specific embodiment
In order that the objects, technical solutions and advantages of the present invention become more apparent, it is below in conjunction with drawings and Examples, right
The present invention is further elaborated.It should be appreciated that specific embodiment described herein is only to explain the present invention, and
It is not used in the restriction present invention.
With reference to Fig. 1, a kind of face information identifying system as shown in Figure 1, including face detection module, face match somebody with somebody quasi-mode
Block, recognition of face confirm module;The face detection module detects the information of face, and passes to face registration module;People
Face registration module carries out registration according to face information;Recognition of face is passed to after the completion of registration and confirms that module is confirmed.
With reference to Fig. 2, as shown in Fig. 2 the recognition methods of the face information identifying system is comprised the following steps:
S1, by face recognition technology Quick Acquisition information;
By processing, S2, information recognize which uses people;
S3, start information process, storage information and pushed information.Information pushing therein refers to the letter for pushing specified someone
Breath, such as:Event is reminded, daily record is reminded etc..
Step S1 is specially:The information of human face characteristic point is gathered by face recognition technology, including gathers the information.
Step S2 includes face information collection and processes:
Face information is gathered:
The response diagram of each Feature point correspondence is generated by calculating characteristic point surrounding neighbors texture information;Retouched using overall face
Textural characteristics are stated, by the face texture unrelated with shape, and base on human face characteristic point evolution to standard shape, will be obtained
It is modeled in the pca method face texture unrelated to shape;
Face information collection includes the collection of below face point information:
The left point 1 of chin profile, left point 2, left point 3, left point 4, left point 5, left point 6, left point 7, left point 8, left point 9, chin profile are right
Point 1, right point 2, right point 3, right point 4, right point 5, right point 6, right point 7, right point 8, right point 9;
Left eye bottom, left eye center, left eye upper left point, left eye upper right point, left eye lower-left point, left eye lower-right most point, pupil of left eye, right eye
Bottom, right eye center, right eye upper left point, right eye upper right point, right eye lower-left point, right eye lower-right most point, pupil of right eye, the left point of left eyebrow, a left side
The right point of eyebrow, left eyebrow midpoint, left eyebrow upper left point, left eyebrow upper right point, left eyebrow lower-left point, left eyebrow lower-right most point, midpoint on left eyebrow, under left eyebrow
Midpoint, right eyebrow upper left point, right eyebrow upper right point, right eyebrow lower-left point, right eyebrow lower-right most point, midpoint on right eyebrow, midpoint under right eyebrow;
The right point of the left point 1 of mouth upper lip, the left point 2 of mouth upper lip, the left point 3 of mouth upper lip, the right point 1 of mouth upper lip, the right point 2 of mouth upper lip, mouth upper lip 3;
The right point of the left point 1 of mouth lower lip, the left point 2 of mouth lower lip, the left point 3 of mouth lower lip, the right point 1 of mouth lower lip, the right point 2 of mouth lower lip, mouth lower lip 3;Nose
Left point 1, left point 2, left point 3, the right point 1 of nose, right point 2, right point 3, the left point of nose, the right point of nose.
CascadeCNN is that the depth convolutional network to classical Violajones methods is realized, be a kind of detection speed compared with
Fast method for detecting human face.Using VGA pictures, 14FPS is reached on CPU, 100FPS is reached on GPU, is reached on FDDB
85.1% recall rate and 87% accuracy rate.
Face information process:
With reference to Fig. 3, as shown in figure 3, adopting Cascade CNN, comprising three-level, per grade includes multiple convolutional networks, and the first order is given
Go out an initial point location estimation, on this basis rear two-stage intense adjustment characteristic point position;Multitask with brigadier registration and its
The training of his related face character is carried out simultaneously, and the attribute related to facial feature points includes head pose, expression.
, different from the image characteristic point on angle point or SIFT feature point ordinary meaning, human face characteristic point is usual for human face characteristic point
It is one group of point by artificial predefined.
Most directly adopted feature descriptor is color, gray scale, face each several part is examined using the difference of the colour of skin
Determine position;It is slightly complicated that various textural characteristics descriptions may be selected, such as it is based on class Haar textural characteristics and Adaboost training cascades
The face registration of grader;
Features above description does not all account for the position relationship between characteristic point, therefore does not possess the rational human face structure of maintenance;
Active shape model (Active Shape Models, ASM) and active appearance models (Active Appearance Model,
AAM two kinds of features of texture and shape (shape)) can be expressed simultaneously;The shape facility of the two is all by points distribution models (Point
Distribution Model, PDM) expressing.
Fig. 4 is the statistical Butut of human face characteristic point in 600 facial images;The textural characteristics of each characteristic point of ASM
Represent respectively, the response diagram of each Feature point correspondence is generated by calculating characteristic point surrounding neighbors texture information
(Response Map);In Fig. 5, AAM describes textural characteristics using overall face, by human face characteristic point evolution is arrived
In standard shape, the face texture unrelated with shape is obtained, and is entered based on the pca method face texture unrelated to shape
Row modeling.
Cascade CNN include three-level, and per grade includes multiple convolutional networks, and the first order provides an initial point position and estimates
Count, on this basis rear two-stage intense adjustment characteristic point position;Instruction of the multitask with brigadier's registration face character related to other
Practicing is carried out simultaneously, and the attribute related to facial feature points includes head pose, and expression, the mouth of such as smiling face are likely to open
, front face characteristic point is then symmetrical;Multitask contributes to lifting feature point detection positioning precision, but different task meeting
There are different convergence rates and difficulty, training difficulty is increased.Educational circles adjusts different tasks there is provided two kinds of solutions at present
Training process:Educational circles adjusts the training process of different tasks there is provided two kinds of solutions at present:Task terminates accurate ahead of time
Then (task-wise early stopping criterion) and dynamic state of parameters controlling mechanism.
Task does sth. in advance stop criterion (task-wise early stopping criterion):This criterion is based on decoding
The change of unreliable variable node number in iterative process, can effectively reduce decoding iteration with low computation complexity
Number, while considerably reducing the hard decision number of times of check equations.
Dynamic state of parameters controlling mechanism:Within class scatter matrix is defined, the deviation of its eigenvalue estimate is weighed using parameter
And variance, to solve small sample problem;Discrete matrix between class is weighted, is allowed edge class to be uniformly distributed, is prevented the overlap of edge class,
To improve discrimination.
The face information identifying system of the present invention can be used for the shared information of a small ranges such as company, family and private information is common
The scene deposited.Such as:Refrigerator content reminiscences etc..
The present invention is by recognition of face, automatically quick to recognize, filter out oneself setting information and others be pushed to oneself
Information.Ensure that privacy, security and the accuracy of information.
The foregoing is only presently preferred embodiments of the present invention, not to limit the present invention, all spirit in the present invention and
Within principle, any modification, equivalent substitution and improvements made etc. should be included within the scope of the present invention.