CN106295502B - A kind of method for detecting human face and device - Google Patents

A kind of method for detecting human face and device Download PDF

Info

Publication number
CN106295502B
CN106295502B CN201610590134.2A CN201610590134A CN106295502B CN 106295502 B CN106295502 B CN 106295502B CN 201610590134 A CN201610590134 A CN 201610590134A CN 106295502 B CN106295502 B CN 106295502B
Authority
CN
China
Prior art keywords
image
candidate
layer
convolutional network
depth convolutional
Prior art date
Application number
CN201610590134.2A
Other languages
Chinese (zh)
Other versions
CN106295502A (en
Inventor
陈书楷
王辉能
Original Assignee
厦门中控智慧信息技术有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 厦门中控智慧信息技术有限公司 filed Critical 厦门中控智慧信息技术有限公司
Priority to CN201610590134.2A priority Critical patent/CN106295502B/en
Publication of CN106295502A publication Critical patent/CN106295502A/en
Application granted granted Critical
Publication of CN106295502B publication Critical patent/CN106295502B/en

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06KRECOGNITION OF DATA; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
    • G06K9/00Methods or arrangements for reading or recognising printed or written characters or for recognising patterns, e.g. fingerprints
    • G06K9/00221Acquiring or recognising human faces, facial parts, facial sketches, facial expressions
    • G06K9/00288Classification, e.g. identification

Abstract

The embodiment of the invention discloses a kind of method for detecting human face and devices, can handle the image of arbitrary size, and can detect positive face and side face simultaneously, improve detection speed.The method comprise the steps that obtaining candidate face image by the first depth convolutional network, the first depth convolutional network is the full convolutional network for initial survey;The candidate face image is calculated by the second depth convolutional network, obtains the reliability values of the candidate face image, the second depth convolutional network is the depth convolutional network for verification;If the reliability values of the candidate face image are greater than preset threshold, it is determined as final facial image.

Description

A kind of method for detecting human face and device

Technical field

The present invention relates to image procossing and field of face identification more particularly to a kind of method for detecting human face and device.

Background technique

Method for detecting human face is the basis of recognition of face, detects facial image for identification from image like clockwork It is extremely important.In some scenes, computer auto-detection is needed to go out face for being identified, this requires Face datection sides Method can detect face image and side face image simultaneously, in addition, the image for arbitrary size carries out handling also proposed wanting It asks.

Currently, method for detecting human face can only detect positive face, or side face can only be detected, and generally pass through fixed size Image is handled.

It clearly for the needing while detecting positive face and side face of the task, can only detect respectively, thereby result in detection speed mistake Slowly, in addition the image of arbitrary size can not be handled.

Summary of the invention

The embodiment of the invention provides a kind of method for detecting human face and device, can be to the image of arbitrary size at Reason, and positive face and side face can be detected simultaneously, improve detection speed.

In view of this, first aspect present invention provides a kind of method for detecting human face, comprising:

Candidate face image is obtained by the first depth convolutional network, the first depth convolutional network is for initial survey Full convolutional network;

The candidate face image is calculated by the second depth convolutional network, obtains the candidate face image Reliability values, the second depth convolutional network are the depth convolutional network for verification;

If the reliability values of the candidate face image are greater than preset threshold, it is determined as final facial image.

It is optional:

It is described to include: by the first depth convolutional network acquisition candidate face image

Face thermodynamic chart is generated by the first depth convolutional network;

Part hottest point is determined from the face thermodynamic chart, and using the local hottest point as candidate face position;

According to the candidate face position acquisition candidate face image.

It is optional:

It is described to include: before according to the candidate face position acquisition candidate face image

Judge the candidate face position with the presence or absence of overlapping;

If so, merging the candidate face position of overlapping.

It is optional:

Include: before the acquisition candidate face image by the first depth convolutional network

Generate the first depth convolutional network;

Facial image and inhuman face image are acquired, and using the facial image and inhuman face image as training sample;

Pass through training sample training the first depth convolutional network.

It is optional:

The second depth convolutional network is multiple depth convolutional networks, by the second depth convolutional network to the candidate Facial image is calculated, and the reliability values for obtaining the candidate face image include:

The candidate face image is calculated respectively by the multiple depth convolutional network, obtains the candidate Multiple reliability values of face image;

The reliability values of the candidate face image are obtained according to the multiple reliability values.

It is optional:

The first depth convolutional network include multilayer, successively are as follows: the first input layer, the first convolutional layer, the first output layer, First maximum pond layer, the second output layer, the first activation primitive layer, the second convolutional layer, third output layer, the second activation primitive Layer, third convolutional layer and the 4th output layer;The second depth convolutional network includes multilayer, successively are as follows: the second input layer, Four convolutional layers, the 5th output layer, the second maximum pond layer, the 6th output layer, third activation primitive layer, the 5th convolutional layer, the 7th Output layer, third maximum pond layer, the 8th output layer, the 4th activation primitive layer, full articulamentum and the 9th output layer.

Second aspect of the present invention provides a kind of human face detection device, comprising:

Module is obtained, for obtaining candidate face image, the first depth convolution net by the first depth convolutional network Network is the full convolutional network for initial survey;

First processing module is obtained for being calculated by the second depth convolutional network the candidate face image The reliability values of the candidate face image, the second depth convolutional network are the preset depth convolution net for verification Network;

Determination module is determined as final if the reliability values for the candidate face image are greater than preset threshold Facial image.

It is optional:

The acquisition module includes:

Generation unit, for generating face thermodynamic chart by the first depth convolutional network;

First processing units, for determining part hottest point from the face thermodynamic chart, and by the local hottest point As candidate face position;

Acquiring unit, for according to the candidate face position acquisition candidate face image.

It is optional:

Described device further include:

Judgment module, for judging the candidate face position with the presence or absence of overlapping;

Second processing module merges overlapping if judging that the candidate face position has overlapping for judgment module The candidate face position.

It is optional:

Described device further include:

Generation module, for generating the first depth convolutional network;

Third processing module, for acquiring facial image and inhuman face image, and by the facial image and non-face figure As being used as training sample;

Training module, for passing through training sample training the first depth convolutional network.

It is optional:

The second depth convolutional network is multiple depth convolutional networks, and the first processing module includes:

Computing unit, for being calculated respectively the candidate face image by the multiple depth convolutional network, Obtain multiple reliability values of the candidate face image;

The second processing unit, for obtaining the reliability number of the candidate face image according to the multiple reliability values Value.

It is optional:

The first depth convolutional network include multilayer, successively are as follows: the first input layer, the first convolutional layer, the first output layer, First maximum pond layer, the second output layer, the first activation primitive layer, the second convolutional layer, third output layer, the second activation primitive Layer, third convolutional layer and the 4th output layer;The second depth convolutional network includes multilayer, successively are as follows: the second input layer, Four convolutional layers, the 5th output layer, the second maximum pond layer, the 6th output layer, third activation primitive layer, the 5th convolutional layer, the 7th Output layer, third maximum pond layer, the 8th output layer, the 4th activation primitive layer, full articulamentum and the 9th output layer.

As can be seen from the above technical solutions, the embodiment of the present invention has the advantage that the network for initial survey is full volume Product network, so that the present invention can handle and improve Face datection speed to the image of arbitrary size, it is in addition of the invention Candidate face image do not limit face image or side face image, so the present invention can also detect positive face and side simultaneously Face.

Detailed description of the invention

Fig. 1 is method for detecting human face of embodiment of the present invention one embodiment schematic diagram;

Fig. 2 is one schematic diagram of the first depth of embodiment of the present invention convolutional network;

Fig. 3 is one schematic diagram of the second depth of embodiment of the present invention convolutional network;

Fig. 4 is human face detection device of embodiment of the present invention one embodiment schematic diagram.

Specific embodiment

The embodiment of the invention provides a kind of method for detecting human face and device, can be to the image of arbitrary size at Reason, and positive face and side face can be detected simultaneously, improve detection speed.

Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on Embodiment in the present invention, those skilled in the art's every other implementation obtained without creative efforts Example, shall fall within the protection scope of the present invention.

Description and claims of this specification and term " first ", " second ", " third ", " in above-mentioned attached drawing The (if present)s such as four " are to be used to distinguish similar objects, without being used to describe a particular order or precedence order.It should manage The data that solution uses in this way are interchangeable under appropriate circumstances, so that the embodiments described herein can be in addition to illustrating herein Or the sequence other than the content of description is implemented.In addition, term " includes " and " having " and their any deformation, it is intended that Cover it is non-exclusive include, for example, containing the process, method, system, product or equipment of a series of steps or units need not limit In step or unit those of is clearly listed, but may include be not clearly listed or for these process, methods, produce The other step or units of product or equipment inherently.

Referring to Fig. 1, method for detecting human face one embodiment includes: in the embodiment of the present invention

101, candidate face image is obtained by the first depth convolutional network, which is preset use In the full convolutional network of initial survey;

In the present embodiment, depth convolutional network is generally only used for the image of fixed size, is classified or is identified, this hair The first depth convolutional network in bright is that full convolutional network allows to be applicable in by the network structure using full convolutional network In the image of arbitrary size.

Optionally, in some embodiments of the invention, candidate face image tool is obtained by the first depth convolutional network Body are as follows:

Face thermodynamic chart is generated by the first depth convolutional network;

Part hottest point is determined from face thermodynamic chart, and using local hottest point as candidate face position;

According to candidate face position acquisition candidate face image.

It should be noted that the first depth convolutional network can be a mininet, mininet is for generating face Thermodynamic chart;Then local hottest point is found from face thermodynamic chart, as candidate face position;According to candidate face position from original Candidate face image is intercepted out in figure.Such as original image is a photo, includes a little girl and a desk in photo, leads to The facial image of little girl can be intercepted out from the photo by crossing the first depth convolutional network.

It is further alternative, in some embodiments of the invention, according to candidate face position acquisition candidate face image Include: before

Judge candidate face position with the presence or absence of overlapping;

If so, merging the candidate face position of overlapping.

Optionally, in some embodiments of the invention, by the first depth convolutional network obtain candidate face image it Before include:

Generate the first depth convolutional network;

Facial image and inhuman face image are acquired, and using facial image and inhuman face image as training sample;

Pass through training sample the first depth convolutional network of training.

It should be noted that the present invention acquires a large amount of face and inhuman face image as training sample for training first Depth convolutional network.In general, facial image is located in a biggish image, by a rectangular area come accurate identification face Position and size, the rectangle become face frame;Random disturbance is added by the size and location to face frame, intercepts out face Image, to detect the facial image of different directions, is repeatedly obtained from a facial image so then into row stochastic rotation Multiple training samples;Usually original facial image needs 10000 or more, the training sample extracted up to 100000 or more, with Just the first depth convolutional network learns to better feature.

The selection of inhuman face image can intercept at random in the pictures such as picture with scenes, the object picture of not face, non- The number of facial image is significantly larger than the number of facial image, for example is greater than 1,000,000.

The first depth convolutional network is trained by existing neural network tool and training sample, determines first The parameter of depth convolutional network.When being trained to the first depth convolutional network, if having used cross entropy loss function, finally Convolution is needed using two convolution sums, but after the completion of training in use, due to only using the defeated of first convolution kernel Out as a result, therefore the result of second convolution kernel does not have to calculate, can remove.

It should be noted that the second depth convolutional network and the first depth convolutional network are similar, it can also pass through training sample Training, details are not described herein again.

Optionally, in some embodiments of the invention, the first depth convolutional network includes multilayer, successively are as follows: first is defeated Enter layer, the first convolutional layer, the first output layer, the first maximum pond layer, the second output layer, the first activation primitive layer, the second convolution Layer, third output layer, the second activation primitive layer, third convolutional layer and the 4th output layer, the first depth is rolled up in order to facilitate understanding Product network, is below described in detail the first depth convolutional network:

Referring to Fig. 2, Fig. 2 indicates a kind of preset full convolutional network for initial survey, the first depth convolution net that is to say Network, 3 channels (H × W) image that the first input layer inputs a width arbitrary dimension use 32 5 × 5 convolution in the first convolutional layer Core carries out convolution, and the first output layer obtains the output image in 32 channels, and height and width reduce 4 pixels simultaneously, i.e., (H-4) × (W-4);Then first maximum pond layer carries out one time 4 × 4 maximum pondization processing, pixel number is reduced to original 1/4, i.e., Second output layer obtains ((H-4)/4) × ((W-4)/4) in 32 channels, and then the first activation primitive layer carries out ReLU activation primitive Processing;Following second convolutional layer carries out convolution using 64 7 × 7 convolution kernels, and third output layer obtains the output figure in 64 channels Picture, i.e. ((H-4)/4-6) × ((W-4)/4-6), the second activation primitive layer reuse the processing of ReLU activation primitive;Last third Convolutional layer carries out convolution using 11 × 1 convolution kernel, and the 4th output layer obtains 1 channel ((H-4)/4-6) × ((W-4)/4- 6), that is, last face probability graph, i.e. thermodynamic chart, local maximum point is possible face on the thermodynamic chart.

The full convolutional network, which is equivalent to, is mapped as a probability value the small cube image of 32 × 32 pixels, therefore using should The face (32 × 32) of full convolutional network one scale of intelligent measurement, will detect the face of other scales, original image is needed to scale It detects again afterwards, the number and scaling for needing to scale are determined according to the range for the face size to be detected.

Occasion higher for rate request, can further decrease the size of the full convolutional network, such as input picture Using single pass gray level image, or convolution nuclear volume is reduced, such as first time convolution is 4 convolution kernels, second is 16 A convolution kernel can greatly speed up the speed of processing.

The first depth convolutional network for initial survey is full convolutional network, at the image to arbitrary size Reason;Secondly it requires to be that speed is fast, precision can be slightly lower, and usual reliability can reach 99.3% or more.

102, candidate face image is calculated by the second depth convolutional network, obtains the reliable of candidate face image Property numerical value, which is the preset depth convolutional network for verification;

In the present embodiment, requirement of the second depth convolutional network to speed for verification is not needed too strictly, but is needed Higher reliability, in general, reliability values need to reach 99.7% or more.The second depth convolutional network for verification can be with It is designed as one or more, in the case where multiple, in order to more reliable, multiple depth convolutional networks are in structure or on convolution kernel The diversity ratio of design is larger, in order to form complementation, can excavate the different characteristic in image.In addition, for verification The image of fixed size can be used as input in two depth convolutional networks, therefore is not required for full convolutional network.In order to reach Higher reliability, the number of plies and image channel number of the second depth convolutional network can be bigger, and Fig. 3 is a kind of typical second depth Convolutional network schematic diagram, the second depth convolutional network in Fig. 3 may include multilayer, successively are as follows: the second input layer, Volume Four product Layer, the 5th output layer, the second maximum pond layer, the 6th output layer, third activation primitive layer, the 5th convolutional layer, the 7th output layer, Third maximum pond layer, the 8th output layer, the 4th activation primitive layer, full articulamentum and the 9th output layer.Specifically, second is defeated Enter layer and input 3 channels (32 × 32) image, in Volume Four lamination, carries out convolution using 32 11 × 11 convolution kernels, the 5th Output layer obtains the output image in 32 channels (22 × 22), then the second maximum pond layer carries out one time 2 × 2 maximum pond Hua Chu Reason, the 6th output layer obtain the output image in 32 channels (11 × 11), and then third activation primitive layer carries out ReLU activation primitive Processing;Following 5th convolutional layer carries out convolution using 64 3 × 3 convolution kernels, and the 7th output layer obtains 64 channels (9 × 9) Image is exported, third maximum pond layer carries out one time 3 × 3 maximum pondization processing, and the 8th output layer obtains 64 channels (3 × 3) Output image, the 4th activation primitive layer reuse ReLU activation primitive processing, full articulamentum obtain 576 input values and 2 output valves, wherein 576 × 2 represent one 576 × 2 parameter matrix.9th output layer finally obtains two values, this two A numerical value can finally be used to calculate face and non-face probability, finally take the probability of face.

Optionally, in some embodiments of the invention, if the second depth convolutional network is multiple depth convolutional networks, step Rapid 102 specifically:

Candidate face image is calculated respectively by multiple depth convolutional networks, obtains the multiple of candidate face image Reliability values;

The reliability values of candidate face image are obtained according to multiple reliability values.

Specifically, multiple reliability values are averaged, using average value as the reliability values of candidate face image; Alternatively, multiple reliability values are maximized, using maximum value as the reliability values of candidate face image;Alternatively, by more A reliability values take weighted value, using weighted value as the reliability values of candidate face image, can also use other methods, It is not construed as limiting herein.

If 103, the reliability values of candidate face image are greater than preset threshold, it is determined as final facial image.

In the present embodiment, reliability values are used to indicate the reliability of candidate face image, and the value of preset threshold can be with It is 99.7%, can also is other reasonable values, be not construed as limiting herein.

In the present embodiment, the network for initial survey is full convolutional network, so that the present invention can be to the image of arbitrary size Face datection speed is handled and is improved, in addition candidate face image of the invention does not limit face image or side face Image, so the present invention can also detect positive face and side face simultaneously.

Referring to Fig. 4, human face detection device one embodiment includes: in the embodiment of the present invention

Module 201 is obtained, for obtaining candidate face image, the first depth convolution net by the first depth convolutional network Network is the preset full convolutional network for initial survey;

First processing module 202 is waited for being calculated by the second depth convolutional network candidate face image The reliability values for face image of choosing, the second depth convolutional network are the preset depth convolutional network for verification;

Determination module 203 is determined as final people if the reliability values for candidate face image are greater than preset threshold Face image.

In the present embodiment, the network for initial survey is full convolutional network, so that the present invention can be to the image of arbitrary size Face datection speed is handled and is improved, in addition candidate face image of the invention does not limit face image or side face Image, so the present invention can also detect positive face and side face simultaneously.

Optionally, obtaining module 201 includes:

Generation unit, for generating face thermodynamic chart by the first depth convolutional network;

First processing units, for determining part hottest point from face thermodynamic chart, and using local hottest point as candidate Face location;

Acquiring unit, for according to candidate face position acquisition candidate face image.

Further, the device further include:

Judgment module, for judging candidate face position with the presence or absence of overlapping;

Second processing module merges the time of overlapping if judging that candidate face position has overlapping for judgment module Select face location.

Optionally, the device further include:

Generation module, for generating the first depth convolutional network;

Third processing module is made for acquiring facial image and inhuman face image, and by facial image and inhuman face image For training sample;

Training module, for passing through training sample the first depth convolutional network of training.

Further, if the second depth convolutional network is multiple depth convolutional networks.First processing module 202 includes:

Computing unit obtains candidate for calculating respectively candidate face image by multiple depth convolutional networks Multiple reliability values of facial image;

The second processing unit, for obtaining the reliability values of candidate face image according to multiple reliability values.

Specifically, multiple reliability values are averaged, using average value as the reliability values of candidate face image; Alternatively, multiple reliability values are maximized, using maximum value as the reliability values of candidate face image;Alternatively, by more A reliability values take weighted value, using weighted value as the reliability values of candidate face image, can also use other methods, It is not construed as limiting herein.

Optionally, the first depth convolutional network includes multilayer, successively are as follows: the first input layer, the first convolutional layer, the first output Layer, the first maximum pond layer, the second output layer, the first activation primitive layer, the second convolutional layer, third output layer, the second activation letter Several layers, third convolutional layer and the 4th output layer;The second depth convolutional network includes multilayer, successively are as follows: the second input layer, Volume Four lamination, the 5th output layer, the second maximum pond layer, the 6th output layer, third activation primitive layer, the 5th convolutional layer, the Seven output layers, third maximum pond layer, the 8th output layer, the 4th activation primitive layer, full articulamentum and the 9th output layer.

It is apparent to those skilled in the art that for convenience and simplicity of description, the system of foregoing description, The specific work process of device and unit, can refer to corresponding processes in the foregoing method embodiment, and details are not described herein.

In several embodiments provided herein, it should be understood that disclosed system, device and method can be with It realizes by another way.For example, the apparatus embodiments described above are merely exemplary, for example, the unit It divides, only a kind of logical function partition, there may be another division manner in actual implementation, such as multiple units or components It can be combined or can be integrated into another system, or some features can be ignored or not executed.Another point, it is shown or The mutual coupling, direct-coupling or communication connection discussed can be through some interfaces, the indirect coupling of device or unit It closes or communicates to connect, can be electrical property, mechanical or other forms.

The unit as illustrated by the separation member may or may not be physically separated, aobvious as unit The component shown may or may not be physical unit, it can and it is in one place, or may be distributed over multiple In network unit.It can select some or all of unit therein according to the actual needs to realize the mesh of this embodiment scheme 's.

It, can also be in addition, the functional units in various embodiments of the present invention may be integrated into one processing unit It is that each unit physically exists alone, can also be integrated in one unit with two or more units.Above-mentioned integrated list Member both can take the form of hardware realization, can also realize in the form of software functional units.

If the integrated unit is realized in the form of SFU software functional unit and sells or use as independent product When, it can store in a computer readable storage medium.Based on this understanding, technical solution of the present invention is substantially The all or part of the part that contributes to existing technology or the technical solution can be in the form of software products in other words It embodies, which is stored in a storage medium, including some instructions are used so that a computer Equipment (can be personal computer, server or the network equipment etc.) executes the complete of each embodiment the method for the present invention Portion or part steps.And storage medium above-mentioned includes: USB flash disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic or disk etc. are various can store journey The medium of sequence code.

The above, the above embodiments are merely illustrative of the technical solutions of the present invention, rather than its limitations;Although referring to before Stating embodiment, invention is explained in detail, those skilled in the art should understand that: it still can be to preceding Technical solution documented by each embodiment is stated to modify or equivalent replacement of some of the technical features;And these It modifies or replaces, the spirit and scope for technical solution of various embodiments of the present invention that it does not separate the essence of the corresponding technical solution.

Claims (10)

1. a kind of method for detecting human face characterized by comprising
Candidate face image is obtained by the first depth convolutional network, the first depth convolutional network is preset for initial survey Full convolutional network, the first depth convolutional network includes multilayer, successively are as follows: the first input layer, the first convolutional layer, first defeated Layer, the first maximum pond layer, the second output layer, the first activation primitive layer, the second convolutional layer, third output layer, the second activation out Function layer, third convolutional layer and the 4th output layer;
The candidate face image is calculated by the second depth convolutional network, obtains the reliable of the candidate face image Property numerical value, the second depth convolutional network is the preset depth convolutional network for verification, the second depth convolution net Network includes multilayer, successively are as follows: the second input layer, Volume Four lamination, the 5th output layer, the second maximum pond layer, the 6th output layer, Third activation primitive layer, the 5th convolutional layer, the 7th output layer, third maximum pond layer, the 8th output layer, the 4th activation primitive Layer, full articulamentum and the 9th output layer;
If the reliability values of the candidate face image are greater than preset threshold, it is determined as final facial image.
2. the method according to claim 1, wherein the first depth convolutional network that passes through obtains candidate face Image includes:
Face thermodynamic chart is generated by the first depth convolutional network;
Part hottest point is determined from the face thermodynamic chart, and using the local hottest point as candidate face position;
According to the candidate face position acquisition candidate face image.
3. according to the method described in claim 2, it is characterized in that, described according to the candidate face position acquisition candidate face Include: before image
Judge the candidate face position with the presence or absence of overlapping;
If so, merging the candidate face position of overlapping.
4. the method according to claim 1, wherein the first depth convolutional network that passes through obtains candidate face Include: before image
Generate the first depth convolutional network;
Facial image and inhuman face image are acquired, and using the facial image and inhuman face image as training sample;
Pass through training sample training the first depth convolutional network.
5. the method according to claim 1, which is characterized in that the second depth convolutional network is more A depth convolutional network calculates the candidate face image by the second depth convolutional network, obtains the candidate The reliability values of face image include:
The candidate face image is calculated respectively by the multiple depth convolutional network, obtains the candidate face figure Multiple reliability values of picture;
The reliability values of the candidate face image are obtained according to the multiple reliability values.
6. a kind of human face detection device characterized by comprising
Module is obtained, for obtaining candidate face image by the first depth convolutional network, the first depth convolutional network is The preset full convolutional network for initial survey, the first depth convolutional network include multilayer, successively are as follows: the first input layer, the One convolutional layer, the first output layer, the first maximum pond layer, the second output layer, the first activation primitive layer, the second convolutional layer, third Output layer, the second activation primitive layer, third convolutional layer and the 4th output layer;
First processing module obtains described for being calculated by the second depth convolutional network the candidate face image The reliability values of candidate face image, the second depth convolutional network are the preset depth convolutional network for verification, The second depth convolutional network includes multilayer, successively are as follows: the second input layer, Volume Four lamination, the 5th output layer, the second maximum It is pond layer, the 6th output layer, third activation primitive layer, the 5th convolutional layer, the 7th output layer, third maximum pond layer, the 8th defeated Layer, the 4th activation primitive layer, full articulamentum and the 9th output layer out;
Determination module is determined as final face if the reliability values for the candidate face image are greater than preset threshold Image.
7. device according to claim 6, which is characterized in that the acquisition module includes:
Generation unit, for generating face thermodynamic chart by the first depth convolutional network;
First processing units, for determining part hottest point from the face thermodynamic chart, and using the local hottest point as Candidate face position;
Acquiring unit, for according to the candidate face position acquisition candidate face image.
8. device according to claim 7, which is characterized in that described device further include:
Judgment module, for judging the candidate face position with the presence or absence of overlapping;
Second processing module merges the described of overlapping if judging that the candidate face position has overlapping for judgment module Candidate face position.
9. device according to claim 6, which is characterized in that described device further include:
Generation module, for generating the first depth convolutional network;
Third processing module is made for acquiring facial image and inhuman face image, and by the facial image and inhuman face image For training sample;
Training module, for passing through training sample training the first depth convolutional network.
10. according to device described in claim 6 to 9 any one, which is characterized in that the second depth convolutional network is more A depth convolutional network, the first processing module include:
Computing unit is obtained for being calculated respectively the candidate face image by the multiple depth convolutional network Multiple reliability values of the candidate face image;
The second processing unit, for obtaining the reliability values of the candidate face image according to the multiple reliability values.
CN201610590134.2A 2016-07-25 2016-07-25 A kind of method for detecting human face and device CN106295502B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201610590134.2A CN106295502B (en) 2016-07-25 2016-07-25 A kind of method for detecting human face and device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201610590134.2A CN106295502B (en) 2016-07-25 2016-07-25 A kind of method for detecting human face and device

Publications (2)

Publication Number Publication Date
CN106295502A CN106295502A (en) 2017-01-04
CN106295502B true CN106295502B (en) 2019-07-12

Family

ID=57652237

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201610590134.2A CN106295502B (en) 2016-07-25 2016-07-25 A kind of method for detecting human face and device

Country Status (1)

Country Link
CN (1) CN106295502B (en)

Families Citing this family (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106845427B (en) * 2017-01-25 2019-12-06 北京深图智服技术有限公司 face detection method and device based on deep learning
CN108446694B (en) * 2017-02-16 2020-11-27 杭州海康威视数字技术股份有限公司 Target detection method and device
CN107145833A (en) * 2017-04-11 2017-09-08 腾讯科技(上海)有限公司 The determination method and apparatus of human face region
CN107403141B (en) * 2017-07-05 2020-01-10 中国科学院自动化研究所 Face detection method and device, computer readable storage medium and equipment
CN107665333A (en) * 2017-08-28 2018-02-06 平安科技(深圳)有限公司 A kind of indecency image identification method, terminal, equipment and computer-readable recording medium based on convolutional neural networks
CN110163033A (en) * 2018-02-13 2019-08-23 京东方科技集团股份有限公司 Positive sample acquisition methods, pedestrian detection model generating method and pedestrian detection method
CN109598212B (en) * 2018-11-20 2020-11-24 北京知道创宇信息技术股份有限公司 Face detection method and device
CN110210457A (en) * 2019-06-18 2019-09-06 广州杰赛科技股份有限公司 Method for detecting human face, device, equipment and computer readable storage medium

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103824054A (en) * 2014-02-17 2014-05-28 北京旷视科技有限公司 Cascaded depth neural network-based face attribute recognition method
CN104992167A (en) * 2015-07-28 2015-10-21 中国科学院自动化研究所 Convolution neural network based face detection method and apparatus
CN105447458A (en) * 2015-11-17 2016-03-30 深圳市商汤科技有限公司 Large scale crowd video analysis system and method thereof
CN105608456A (en) * 2015-12-22 2016-05-25 华中科技大学 Multi-directional text detection method based on full convolution network
CN105631427A (en) * 2015-12-29 2016-06-01 北京旷视科技有限公司 Suspicious personnel detection method and system

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8582807B2 (en) * 2010-03-15 2013-11-12 Nec Laboratories America, Inc. Systems and methods for determining personal characteristics
US9418319B2 (en) * 2014-11-21 2016-08-16 Adobe Systems Incorporated Object detection using cascaded convolutional neural networks

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103824054A (en) * 2014-02-17 2014-05-28 北京旷视科技有限公司 Cascaded depth neural network-based face attribute recognition method
CN104992167A (en) * 2015-07-28 2015-10-21 中国科学院自动化研究所 Convolution neural network based face detection method and apparatus
CN105447458A (en) * 2015-11-17 2016-03-30 深圳市商汤科技有限公司 Large scale crowd video analysis system and method thereof
CN105608456A (en) * 2015-12-22 2016-05-25 华中科技大学 Multi-directional text detection method based on full convolution network
CN105631427A (en) * 2015-12-29 2016-06-01 北京旷视科技有限公司 Suspicious personnel detection method and system

Also Published As

Publication number Publication date
CN106295502A (en) 2017-01-04

Similar Documents

Publication Publication Date Title
JP6547069B2 (en) Convolutional Neural Network with Subcategory Recognition Function for Object Detection
US9633282B2 (en) Cross-trained convolutional neural networks using multimodal images
CN105631439B (en) Face image processing process and device
US9542621B2 (en) Spatial pyramid pooling networks for image processing
CN106934397B (en) Image processing method and device and electronic equipment
US9349076B1 (en) Template-based target object detection in an image
CN105184312B (en) A kind of character detecting method and device based on deep learning
US20170124409A1 (en) Cascaded neural network with scale dependent pooling for object detection
US20170345146A1 (en) Liveness detection method and liveness detection system
CN105138993B (en) Establish the method and device of human face recognition model
Hazirbas et al. Fusenet: Incorporating depth into semantic segmentation via fusion-based cnn architecture
CN104992167B (en) A kind of method for detecting human face and device based on convolutional neural networks
CN103544506B (en) A kind of image classification method and device based on convolutional neural networks
US10242289B2 (en) Method for analysing media content
JP2014232533A (en) System and method for ocr output verification
CN107423701B (en) Face unsupervised feature learning method and device based on generative confrontation network
EP3327583A1 (en) Method and device for searching a target in an image
US20160104058A1 (en) Generic object detection in images
US20180114071A1 (en) Method for analysing media content
CN101558431B (en) Face authentication device
CN101198987B (en) Object detecting device and its learning device
US10262190B2 (en) Method, system, and computer program product for recognizing face
Nguyen et al. Combining deep and handcrafted image features for presentation attack detection in face recognition systems using visible-light camera sensors
JP6159489B2 (en) Face authentication method and system
CN108475331A (en) Use the candidate region for the image-region for including interested object of multiple layers of the characteristic spectrum from convolutional neural networks model

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
TA01 Transfer of patent application right
TA01 Transfer of patent application right

Effective date of registration: 20170824

Address after: 361008, Xiamen three software park, Fujian Province, 8 North Street, room 2001

Applicant after: Xiamen Central Intelligent Information Technology Co., Ltd.

Address before: 361000 Fujian province Xiamen software park two sunrise Road No. 32 403 unit 02 District

Applicant before: XIAMEN ZHONGKONG BIOLOGICAL RECOGNITION INFORMATION TECHNOLOGY CO., LTD.

GR01 Patent grant
GR01 Patent grant
CP03 Change of name, title or address
CP03 Change of name, title or address

Address after: Room 1301, No.132, Fengqi Road, phase III, software park, Xiamen City, Fujian Province

Patentee after: Xiamen Entropy Technology Co., Ltd

Address before: 361008 room 8, 2001 North Avenue, Xiamen Software Park, Fujian, three

Patentee before: XIAMEN ZKTECO BIOMETRIC IDENTIFICATION TECHNOLOGY Co.,Ltd.