CN106203242A - A kind of similar image recognition methods and equipment - Google Patents
A kind of similar image recognition methods and equipment Download PDFInfo
- Publication number
- CN106203242A CN106203242A CN201510229654.6A CN201510229654A CN106203242A CN 106203242 A CN106203242 A CN 106203242A CN 201510229654 A CN201510229654 A CN 201510229654A CN 106203242 A CN106203242 A CN 106203242A
- Authority
- CN
- China
- Prior art keywords
- image
- identified
- specific characteristic
- normalized
- authentication
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 49
- 230000004044 response Effects 0.000 claims description 76
- 238000013527 convolutional neural network Methods 0.000 claims description 42
- 238000013507 mapping Methods 0.000 claims description 17
- 238000012795 verification Methods 0.000 claims description 16
- 238000001514 detection method Methods 0.000 claims description 14
- 238000012549 training Methods 0.000 claims description 11
- 230000005540 biological transmission Effects 0.000 claims description 3
- 230000006870 function Effects 0.000 description 14
- 230000008569 process Effects 0.000 description 9
- 210000000887 face Anatomy 0.000 description 8
- 238000010606 normalization Methods 0.000 description 6
- 239000000284 extract Substances 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 3
- 238000000605 extraction Methods 0.000 description 3
- 230000001815 facial effect Effects 0.000 description 3
- 238000006243 chemical reaction Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000003203 everyday effect Effects 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012905 input function Methods 0.000 description 1
- 230000007935 neutral effect Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
- G06V10/761—Proximity, similarity or dissimilarity measures
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Theoretical Computer Science (AREA)
- Medical Informatics (AREA)
- Computing Systems (AREA)
- Databases & Information Systems (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Collating Specific Patterns (AREA)
- Image Analysis (AREA)
Abstract
This application discloses a kind of similar image recognition methods.In determining the first image to be identified corresponding with specific characteristic until contrast district and after the image in region to be identified is alignd with the standard picture preset, will in region to be identified the resolution adjustment of image after alignment to the resolution preset, and the image after adjusting is as normalized image, finally obtain the metric range between normalized image and the normalized image of the second image to be identified of the first image to be identified, determine that according to the size between metric range and default threshold value the specific characteristic of the first image to be identified and the second image to be identified is the most similar.Thus on the premise of ensureing accuracy, carry out identification rapidly and efficiently for the similarity between image to be detected and another image to be detected, provide reference frame for improving the safety of existing system.
Description
Technical field
The application relates to communication technical field, particularly to a kind of similar image recognition methods.The application is same
Time further relate to a kind of similar image identification equipment.
Background technology
Along with the Internet and the development of computer information technology, shopping at network day by day becomes the new of people's shopping
Fashion.For the marketing system that visit capacity and purchase volume are the biggest, every day suffers from up to ten million
Businessman user sells commodity by this marketing system, but also has a lot of lawless person's false impersonation's identity simultaneously
Attempt carries out trading processing in marketing system, thus may produce various violation operation, thus to it
He causes damage at the rights and interests of validated user.The most how based on the photo to be detected uploaded from businessman user and
Whether existing photo of putting on record judges in two kinds of photos is same person, becomes marketing system to be solved
One of problem.
Traditional face authentication method is based primarily upon SIFT (Scale-invariant feature transform, chi
Degree invariant features conversion), the feature such as LBP (Local Binary Patterns, local binary patterns) treats
Face in detection photo and existing photo is described, and then judges two faces by grader
Whether is same person, wherein SIFT is a kind of local feature description for image processing field, this
Plant description there is scale invariability and can detect that key point, SIFT feature are based on object in the picture
Some local appearance point of interest and with the size of image and rotate unrelated, for light, noise, a little
The tolerance that micro-visual angle changes is the most at a relatively high;LBP is a kind of effective texture description operator, it is possible to tolerance and
Extract the texture information of image local, illumination is had invariance.
But, during realizing the application, inventor finds that prior art also exists following shortcoming:
The face authentication algorithm that traditional feature based describes extracts high-dimensional spy often through to human face region
Levy, by the way of grader, carry out face authentication.This kind of algorithm is only the most special for face characteristic
Not significantly picture or photo is the most effective.More complicated in background, in the case of face changes greatly,
Whether the image that identification technology of the prior art often cannot be passed through in two photos accurately is same
People, therefore, how on the premise of ensureing recognition accuracy, for image to be detected and existing image
Carry out identification rapidly and efficiently, become the technical problem that those skilled in the art are urgently to be resolved hurrily.
Summary of the invention
This application provides a kind of similar image recognition methods, be used on the premise of ensureing accuracy, pin
Image to be detected and existing image are carried out identification rapidly and efficiently, and the method includes:
Obtain and corresponding with specific characteristic in the first image to be identified treat contrast district;
Image in described region to be identified is alignd with the standard picture preset, and by after alignment
Image is as the normalized image of described first image to be identified, described standard picture and described specific characteristic
Corresponding;
Determine the metric range between the normalized image of described normalized image and the second image to be identified,
Described metric range exists according to the normalized image of described normalized image and described second image to be identified
Distance in feature space generates, and wherein, similar normalized image is little in the distance of described feature space
In non-similar normalized image in the distance of described feature space;
If described metric range is more than the threshold value preset, confirm described first image to be identified and described second
The specific characteristic of image to be identified is dissimilar;
If described metric range is less than or equal to described threshold value, confirm that described first image to be identified is with described
The specific characteristic of the second image to be identified is similar.
Preferably, obtain and corresponding with specific characteristic in the first image to be identified treat contrast district, particularly as follows:
Described in determining in described first image to be identified according to the detection algorithm corresponding with described specific characteristic
Treat contrast district;
Many with described specific characteristic by treating in contrast district described in default key point regression model acquisition
The key point coordinate that individual key point feature is corresponding.
Preferably, the image in described region to be identified is alignd, specifically with the standard picture preset
For:
According to parameter M, each key point coordinate in described region to be identified is mapped as the pass of the image after aliging
Key point coordinates;
Wherein, each key point coordinate of described standard picture and having marked in image according to described parameter M
The key point Coordinate generation of the image corresponding with described specific characteristic.
Preferably, by the image after alignment as after the normalized image of described first image to be identified,
Also include:
By the resolution adjustment of described normalized image to the resolution preset.
Preferably, the degree between the normalized image of described normalized image and the second image to be identified is determined
Span from, particularly as follows:
The specific characteristic in described normalized image is extracted by convolutional neural networks;
Determine that described specific characteristic is mapping to spy according to convolutional neural networks and distance metric loss function
Levy the eigenvalue behind space, and using described eigenvalue as the eigenvalue of described normalized image;
Determine the spy of the eigenvalue of described normalized image and the normalized image of described second image to be identified
Euclidean distance between value indicative, using described Euclidean distance as described metric range.
Preferably, described specific characteristic is specially face area, and described key point feature at least includes left eye
Region, right eye region, nasal area, left corners of the mouth region and right corners of the mouth region.
Preferably, described convolutional neural networks parameter is to have obtained according to having marked image training, described marks
Note image includes the most similar normalized image of specific characteristic and the mutual dissimilar normalization of specific characteristic
Image.
Accordingly, the application also proposed a kind of similar image identification equipment, including:
Acquisition module, corresponding with specific characteristic in the first image to be identified treats contrast district for obtaining;
Alignment module, for the image in described region to be identified is alignd with the standard picture preset,
And by the image after alignment as the normalized image of described first image to be identified, described standard picture with
Described specific characteristic is corresponding;
Determine module, for determining that the described normalized image of described first image to be identified is waited to know with second
Metric range between the normalized image of other image, described metric range according to described normalized image with
And the distance that the normalized image of described second image to be identified is in feature space generates, wherein, similar
Normalized image empty in described feature less than non-similar normalized image in the distance of described feature space
Between distance;
Identification module, described first to be identified for confirming when described metric range is more than the threshold value preset
The specific characteristic of image and described second image to be identified is dissimilar, and described metric range less than or
The specific characteristic of described first image to be identified and described second image to be identified is confirmed during equal to described threshold value
Similar.
Preferably, described determine module specifically for:
Described in determining in described first image to be identified according to the detection algorithm corresponding with described specific characteristic
Treat contrast district, by default key point regression model obtain described in treat in contrast district with described appointment
The key point coordinate that multiple key point features of feature are corresponding.
Preferably, described alignment module specifically for:
According to parameter M, each key point coordinate in described region to be identified is mapped as the pass of the image after aliging
Key point coordinates, wherein, each key point coordinate of described standard picture and having marked according to described parameter M
The key point Coordinate generation of image corresponding with described specific characteristic in note image.
Preferably, also include:
Adjusting module, for the resolution extremely preset by the resolution adjustment of described normalized image.
Preferably, described acquisition module specifically for:
The specific characteristic in described normalized image is extracted by convolutional neural networks;
Determine that described specific characteristic is mapping to spy according to convolutional neural networks and distance metric loss function
Levy the eigenvalue behind space, and using described eigenvalue as the eigenvalue of described normalized image;
Determine the spy of the eigenvalue of described normalized image and the normalized image of described second image to be identified
Euclidean distance between value indicative, using described Euclidean distance as described metric range.
Preferably, described specific characteristic is specially face area, and described key point feature at least includes left eye
Region, right eye region, nasal area, left corners of the mouth region and right corners of the mouth region.
Preferably, described convolutional neural networks parameter is to have obtained according to having marked image training, described marks
Note image includes the most similar normalized image of specific characteristic and the mutual dissimilar normalization of specific characteristic
Image.
Accordingly, the application also proposed a kind of similar image recognition methods, and the method is applied to client,
Comprise the following steps:
Receiving the ID authentication request of user, described ID authentication request carries first that described user uploads
Image to be identified and the authentication information of described user;
Described ID authentication request is sent to server, so that described server is according to described authentication information
Obtain the second to be identified image corresponding with described user;
Receive the authentication response that described server sends;
Described client shows authentication result according to described authentication response to described user.
Preferably, receive the ID authentication request of user, particularly as follows:
Obtain image and the information of described user input that described user uploads;
Using described image as described first image to be identified, and described information is believed as described certification
Breath;
Described ID authentication request is generated according to described first image to be identified and described authentication information.
Preferably, described authentication response is authentication success response or authentication failure response,
Also include:
Described authentication success response is that described server is confirming that described first image to be identified is with described
The similar generation afterwards of specific characteristic of the second image to be identified;
Described authentication failure response is that described server is confirming that described first image to be identified is with described
Generate after the specific characteristic dissmilarity of the second image to be identified.
Preferably, show authentication result according to described authentication response to described user, particularly as follows:
When receiving described authentication success response, show default with described identity to described user
The interface that certification success response is corresponding;
When receiving described authentication failure response, show default with described identity to described user
The interface that authentication failure response is corresponding, and show the need of carrying out carrying of manual verification to described user
Show information.
Preferably, the default interface corresponding with described authentication failure response is being shown to described user
And after described user shows the need of the prompting carrying out manual verification, also include:
If receiving manual verification's request of described user, the transmission of described ID authentication request is extremely preset
Service end.
Accordingly, the application also proposed a kind of client, including:
Receiver module, for receiving the ID authentication request of user, described ID authentication request is carried described
The first image to be identified that user uploads and the authentication information of described user;
Sending module, for described ID authentication request is sent to server, so that described server root
The second to be identified image corresponding with described user is obtained according to described authentication information;
Receiver module, for receiving the authentication response that described server sends;
Display module, for showing authentication result according to described authentication response to described user.
Preferably, described receiver module specifically for:
Obtain image and the information of described user input that described user uploads, using described image as institute
State the first image to be identified, and using described information as described authentication information, wait to know according to described first
Other image and described authentication information generate described ID authentication request.
Preferably, described authentication response is authentication success response or authentication failure response,
Also include:
Described authentication success response is that described server is confirming that described first image to be identified is with described
The similar generation afterwards of specific characteristic of the second image to be identified;
Described authentication failure response is that described server is confirming that described first image to be identified is with described
Generate after the specific characteristic dissmilarity of the second image to be identified.
Preferably, described display module, specifically for recognizing when described receiver module receives described identity
During card success response, show the default interface corresponding with described authentication success response to described user;
Or, described display module, specifically for losing when described receiver module receives described authentication
When losing response, show the default interface corresponding with described authentication failure response to described user, with
And show the need of the information carrying out manual verification to described user.
Preferably, show default with described when described receiver module to described user at described display module
Interface that authentication failure response is corresponding and showing the need of carrying out manual verification's to described user
After prompting, also receiving manual verification's request of described user, described receiver module indicates described transmission
Described ID authentication request is sent to the service end preset by module.
Accordingly, the application also proposed a kind of similar image recognition methods, and the method is applied to server,
Comprise the following steps:
Receiving the ID authentication request sent by described client, described ID authentication request carries described use
The first image to be identified that family is uploaded and the authentication information of described user;
According to the second image to be identified that the inquiry of described authentication information is corresponding with described user;
Obtain and corresponding with specific characteristic in the first image to be identified treat contrast district;
Image in described region to be identified is alignd with the standard picture preset, and by after alignment
Image is as the normalized image of described first image to be identified, described standard picture and described specific characteristic
Corresponding;
Determine the metric range between the normalized image of described normalized image and the second image to be identified,
Described metric range exists according to the normalized image of described normalized image and described second image to be identified
Distance in feature space generates, and wherein, similar normalized image is little in the distance of described feature space
In non-similar normalized image in the distance of described feature space;
If described metric range is more than the threshold value preset, confirm described first image to be identified and described second
The specific characteristic of image to be identified is dissimilar, and returns authentication failure response to described client;
If described metric range is less than or equal to described threshold value, confirm that described first image to be identified is with described
The specific characteristic of the second image to be identified is similar, and returns authentication success response to described client.
Preferably, obtain and corresponding with specific characteristic in the first image to be identified treat contrast district, particularly as follows:
Described in determining in described first image to be identified according to the detection algorithm corresponding with described specific characteristic
Treat contrast district;
Many with described specific characteristic by treating in contrast district described in default key point regression model acquisition
The key point coordinate that individual key point feature is corresponding.
Preferably, the image in described region to be identified is alignd, specifically with the standard picture preset
For:
According to parameter M, each key point coordinate in described region to be identified is mapped as the pass of the image after aliging
Key point coordinates;
Wherein, each key point coordinate of described standard picture and marked image according to described parameter M
In the key point Coordinate generation of the image corresponding with described specific characteristic.
Preferably, by the image after alignment as after the normalized image of described first image to be identified,
Also include:
By the resolution adjustment of described normalized image to the resolution preset.
Preferably, the degree between the normalized image of described normalized image and the second image to be identified is determined
Span from, particularly as follows:
The specific characteristic in described normalized image is extracted by convolutional neural networks;
Determine that described specific characteristic is mapping to spy according to convolutional neural networks and distance metric loss function
Levy the eigenvalue behind space, and using described eigenvalue as the eigenvalue of described normalized image;
Determine the spy of the eigenvalue of described normalized image and the normalized image of described second image to be identified
Euclidean distance between value indicative, using described Euclidean distance as described metric range.
Preferably, described specific characteristic is specially face area, and described key point feature at least includes left eye
Region, right eye region, nasal area, left corners of the mouth region and right corners of the mouth region.
Preferably, described convolutional neural networks parameter is to have obtained according to having marked image training, described marks
Note image includes the most similar normalized image of specific characteristic and the mutual dissimilar normalization of specific characteristic
Image.
Accordingly, the application also proposed a kind of server, including:
Receiver module, the ID authentication request sent by described client with reception, described authentication please
Ask and carry the first image to be identified and the authentication information of described user that described user uploads;
Enquiry module, for the second to be identified figure corresponding with described user according to the inquiry of described authentication information
Picture;
Acquisition module, corresponding with specific characteristic in the first image to be identified treats contrast district for obtaining;
Alignment module, for the image in described region to be identified is alignd with the standard picture preset,
And by the image after alignment as the normalized image of described first image to be identified, described standard picture with
Described specific characteristic is corresponding;
Determine module, for determining that the described normalized image of described first image to be identified is waited to know with second
Metric range between the normalized image of other image, described metric range according to described normalized image with
And the distance that the normalized image of described second image to be identified is in feature space generates, wherein, similar
Normalized image empty in described feature less than non-similar normalized image in the distance of described feature space
Between distance;
Identification module, described first to be identified for confirming when described metric range is more than the threshold value preset
The specific characteristic of image and described second image to be identified is dissimilar, and described metric range less than or
The specific characteristic of described first image to be identified and described second image to be identified is confirmed during equal to described threshold value
Similar;
At described identification module, sending module, for confirming that described first image to be identified is treated with described second
Authentication failure response, Yi Ji is returned to described client when identifying the specific characteristic dissmilarity of image
Described identification module confirms the specific characteristic phase of described first image to be identified and described second image to be identified
Like time to described client return authentication success response.
Preferably, described determine module specifically for:
Described in determining in described first image to be identified according to the detection algorithm corresponding with described specific characteristic
Treat contrast district, by default key point regression model obtain described in treat in contrast district with described appointment
The key point coordinate that multiple key point features of feature are corresponding.
Preferably, described alignment module specifically for according to parameter M by each key in described region to be identified
Point coordinates is mapped as the key point coordinate of the image after alignment, wherein, described mark according to described parameter M
Each key point coordinate of quasi-image and marked the key of image corresponding with described specific characteristic in image
Point coordinates generates.
Preferably, also include:
Adjusting module, for the resolution extremely preset by the resolution adjustment of described normalized image.
Preferably, described acquisition module specifically for:
The specific characteristic in described normalized image is extracted by convolutional neural networks;
Determine that described specific characteristic is mapping to spy according to convolutional neural networks and distance metric loss function
Levy the eigenvalue behind space, and using described eigenvalue as the eigenvalue of described normalized image;
Determine the spy of the eigenvalue of described normalized image and the normalized image of described second image to be identified
Euclidean distance between value indicative, using described Euclidean distance as described metric range.
Preferably, described specific characteristic is specially face area, and described key point feature at least includes left eye
Region, right eye region, nasal area, left corners of the mouth region and right corners of the mouth region.
Preferably, described convolutional neural networks parameter is to have obtained according to having marked image training, described marks
Note image includes the most similar normalized image of specific characteristic and the mutual dissimilar normalization of specific characteristic
Image.
As can be seen here, by applying the technical scheme of the application, with finger in determining the first image to be identified
Determine feature corresponding treat contrast district and by the image in region to be identified and preset standard picture carry out right
Qi Hou, will in region to be identified the resolution adjustment of image after alignment to the resolution preset, and
Image after adjusting as normalized image, finally obtain the normalized image of the first image to be identified with
Metric range between the normalized image of the second image to be identified, according to metric range and the threshold value preset
Between size determine that the specific characteristic of the first image to be identified and the second image to be identified is the most similar.From
And on the premise of ensureing accuracy, for the similarity between image to be detected and another image to be detected
Carry out identification rapidly and efficiently, provide reference frame for improving the safety of existing system.
Accompanying drawing explanation
Fig. 1 is the schematic flow sheet of a kind of similar image recognition methods that the application proposes;
Fig. 2 is the convolutional neural networks structure chart training facial modeling in the application specific embodiment;
Fig. 3 is the schematic flow sheet carrying out depth measure study in the application specific embodiment;
Fig. 4 is the convolutional neural networks structure chart training face authentication in the application specific embodiment;
Fig. 5 is the schematic flow sheet that in the application specific embodiment, client carries out similar image identification;
Fig. 6 is the schematic flow sheet that in the application specific embodiment, server carries out similar image identification;
Fig. 7 is the structural representation of a kind of similar image identification equipment that the application proposes;
Fig. 8 is the structural representation of a kind of client that the application proposes;
Fig. 9 is the structural representation of a kind of server that the application proposes.
Detailed description of the invention
Along with popularizing of mobile device, face authentication plays an important role in increasing place.
But disturbed by a lot of other objective factors during face authentication, added the reasons such as angle, typically
Comprise the face in the image of face and often can not directly carry out Characteristic Contrast and extraction.Because should
Problem, present applicant proposes the recognition methods for similar image, and the method can be by means of network environment
Under computer equipment realize.Wherein it is mainly used in the service as system background that similarity is judged
Device, and user oriented client both the most compatible key mapping input and the mobile device of touch-screen input, also
Can be PC equipment, client passes through wired with server or wirelessly realizing network is connected.
As it is shown in figure 1, the schematic flow sheet of a kind of similar image recognition methods proposed by the application, bag
Include following steps:
S101, obtains and corresponding with specific characteristic in the first image to be identified treats contrast district.
Carry out similarity at the picture for face class to judge to differentiate whether two photos are same accurately
During people, orient some key points (such as eyes, nose, the corners of the mouth etc.) of face exactly
The face requisite step of alignment.Therefore, in the application preferred embodiment, treat that contrast district can lead to
Cross and determine that the mode of key point coordinate (relevant to face) obtains.Specifically, determining that first waits to know
In other image corresponding with specific characteristic treat contrast district during, can first according to and described specify spy
The detection algorithm levying correspondence determines and treats contrast district described in described first image to be identified, then passes through
The multiple key points with described specific characteristic in contrast district are treated described in the key point regression model acquisition preset
The key point coordinate that feature is corresponding, determines accurately with this and treats contrast district.Correspondingly, it is intended that feature can
For face area, and key point feature at least includes left eye region, right eye region, nasal area, Zuo Zui
Angular zone and right corners of the mouth region.
According to one embodiment of the application, have employed degree of depth convolutional neural networks and realize returning of face key point
Return.The structure of the neutral net in this specific embodiment is as in figure 2 it is shown, include 4 convolutional layers and 2
Full articulamentum.The most front 3 convolutional layers comprise maximum pond (max pooling) operation, last
Convolutional layer only contains only convolution operation.First full articulamentum contains 100 nodes, second full connection
Become and have 10 nodes, represent the coordinate of 5 key points of face.Return and use Euclidean distance conduct loss
Function, expression formula is as follows:
X represents the coordinate of the key point of mark, represents and passes through convolution
The coordinate of the key point of neural network prediction.
By minimizing above-mentioned loss function, this specific embodiment uses stochastic gradient descent algorithm to optimize mould
Parameter in type, thus training obtains predicting the model of face key point.
Secondly, after each key point coordinate in described region to be identified is mapped as alignment according to parameter M by this step
The key point coordinate of image, wherein each key point coordinate and of described standard picture according to parameter M
The key point Coordinate generation of image corresponding with described specific characteristic in mark image.
In above-mentioned specific embodiment as a example by parameter, this specific embodiment defines 5 in standard faces
Key point position, is left eye respectively, right eye, nose, the left corners of the mouth, right corners of the mouth position, and by examining
The face measured carries out rotating, translates, scales, thus snaps to standard faces.In tentative standard face
Characteristic point position be (x, y), it was predicted that the position of the characteristic point obtained is (x', y'), then both relations
For:
Wherein the parameter of position is a=s cos θ, b=s sin θ, c=tx, d=ty4 unknown parameters, in order to
Solve this 4 parameters, need 4 equatioies.So that the result of alignment is more healthy and stronger, this is concrete real
Execute example to be mapped by 5 points, set up system of linear equations, by the least square method meter of system of linear equations
Calculate the system changing system of linear equations.Specific as follows:
Above-mentioned equation is stated as the form of system of linear equations, can become:
By minimizingWherein solve x=(MTM)-1y
Above procedure is the detailed generation process of parameter M, and those skilled in the art can be according to this parameter M pair
Facial image carries out registration process, and on this basis, other can obtain the improvement implementation of this parameter M
Belong to the protection domain of the application.
S102, aligns the image in described region to be identified with the standard picture preset, and by right
Image after Qi is as the normalized image of described first image to be identified.
In order to reduce the interference of other objective factors so that comparing result is more accurate, and the application is really
Fixed the most corresponding with specific characteristic behind contrast district, need the image in region to be identified and default mark
Quasi-image aligns.In the preferred embodiment of the application, face is snapped to a standard by this step
Face, those skilled in the art can arrange the face of this standard based on existing contrast standard, these
All within the protection domain of the application.
Additionally, in order to further by image standardization so that process, after above procedure terminates, this
Apply for the resolution adjustment of normalized image to the resolution preset.According to one embodiment of the application,
If need to zoom to human face region this specification of 39x39 according to default parameter, this step is crucial by face
Point coordinates information normalizes in 39x39 metric space.
S103, determines described normalized image and second image to be identified of described first image to be identified
Metric range between normalized image, described metric range is according to described normalized image and described
The normalized image of two images to be identified distance in feature space generates, wherein, and similar normalization
Image described feature space distance less than non-similar normalized image described feature space away from
From.
Based on described above, in the application preferred embodiment, first pass through convolutional neural networks and extract
Specific characteristic in described normalized image, loses letter according to convolutional neural networks and distance metric subsequently
Number determines described specific characteristic eigenvalue after mapping to feature space, and using described eigenvalue as institute
State the eigenvalue of normalized image, finally determine that the eigenvalue of described normalized image is waited to know with described second
Euclidean distance between the eigenvalue of the normalized image of other image, using described Euclidean distance as described degree
Span from.
For the special screne of face authentication (two faces compare), the technical scheme knot of the application
Degree of depth convolutional neural networks and metric learning are closed to train face authentication model.Degree of depth convolutional neural networks
At present in the field of image understanding, classify including image, image retrieval, target detection, recognition of face etc.,
It is widely used in a variety of applications.The method adding grader with traditional feature is compared, and convolutional neural networks has
Feature self study, model generalization ability are good etc. a little.Metric learning is by being carried out linearly by feature space
Or nonlinear mapping, so that identical face characteristic distance is less than different face characteristic distances.
It should be noted that described convolutional neural networks parameter is to have obtained according to having marked image training, institute
State and mark image and include that the most similar normalized image of specific characteristic and specific characteristic are the most dissimilar
Normalized image.
Concrete, in order to obtain face authentication model based on depth measure study, the reality that the application is concrete
Executing and use the mode of (pair-wise) in pairs to carry out sample mark in example, each sample packages contains 2 portrait figures
Sheet, if in two images not being same person, represents negative sample, if same person is expressed as
Positive sample.Multiple pictures that positive sample belongs to same person by collection carry out combination of two generation, negative sample
This is not then by being that the picture of same person carries out emulation and generates.Sample is being carried out Face datection, is passing through
The key point of face is predicted, face snaps to mark by the face key point forecast model that training obtains
Quasi-face and after image resolution ratio is zoomed to a series of processes such as 39x39, can train based on degree of depth degree
The face authentication model of amount study.
As it is shown on figure 3, two images in each group of sample through Face datection, key point location and
After face alignment, it is input to convolutional neural networks, extracts, by convolution, the face characteristic that study is arrived.Wherein
Parameter W of the network on the left side and the right is shared.The distance of feature is finally carried out in high-rise semantic space
Tolerance.Depth measure study mainly comprises 2 parts.One of them is parameter W, represents and needs training
The parameter of the convolutional neural networks obtained, another one is distance metric loss function.With traditional people
Face identification is different, and the input of the application is 2 faces, and last loss is also to weigh 2 faces spy
Levy distance spatially.In this specific embodiment, the structure of the convolutional neural networks of use is as shown in Figure 4,
Comprise 4 convolutional layers and 2 full articulamentums, wherein connect maximum sample level after 3 convolutional layers.?
Big sample level makes the feature extracted have translation invariance, and reduces computation complexity.Finally will
The feature nonlinear mapping of face to 100 dimension feature spaces in.
As can be seen here, metric learning be find a transformation space, in this space similar sample away from
From reducing, inhomogeneity sample distance increases.Therefore this step first passes through metric learning and finds a non-thread
Property conversion, face is transformed to a feature space from original pixels so that similar in this space
Face is apart from little, and dissimilar face distance is big.Face is extracted subsequently special by degree of depth convolutional neural networks
Levy, finally combine Feature Mapping to the feature space that convolutional neural networks is acquired by metric learning.By
Being that facial image continues nonlinear mapping in convolutional neural networks, thus obtained feature representation is with artificial
The feature of design is compared more healthy and stronger, and face authentication accuracy rate is higher.
S104, if described metric range is more than the threshold value preset, confirms described first image to be identified and institute
The specific characteristic stating the second image to be identified is dissimilar.
S105, if described metric range is less than or equal to described threshold value, confirms described first image to be identified
Similar to the specific characteristic of described second image to be identified.
In the specific embodiment of S103,100 dimensional features finally obtained will carry out metric learning, this mistake
The loss function of Cheng Caiyong is as follows:
WhereinRepresent the logic loss function of broad sense.(Xi,Xj) ∈ P represents sample
This set, lijRepresent the classification of sample, lij=1 represents XiAnd XjIt is same person, lij=-1 represents XiAnd Xj
Not being same person, W is the parameter of model, FW(Xi) represent when "current" model parameter is W, it is mapped to
The value of the feature of 100 dimensions.Represent two faces distance in feature space, away from
Two from the least expression faces are the most similar.As (Xi,Xj) when being same person, lij=1, loss function along withIncrease and increase, accordingly as (Xi,Xj) when not being same person loss function with
?Increase and reduce, τ indicates whether it is the threshold value of same person.
Parameter W of model can be obtained by minimizing loss function.Preferably, technical staff can adopt
Try to achieve the gradient of corresponding parameter with chain type Rule for derivation and use stochastic gradient descent method (SGD) to optimize
The parameter of computation model, other are capable of the computation model of effect of optimization equally at the protection model of the application
Within enclosing.
Based on described above, when a pair image input, the face in detection figure respectively, according to face district
Territory carries out feature point extraction and aligns with face.Then the model obtained is trained to carry out spy according in step before
Levy extraction and by Feature Mapping to the space of 100 dimensions, finally calculate two face characteristics European away from
From, when distance is to represent it is not same person more than or equal to τ, it is exactly otherwise same person.
Above scheme describes how to judge its most similar process for one group of image to be identified in detail,
In concrete implementation scene, this process can be completed jointly by client and server.Scene is realized at this
In, user can use the such as mobile terminal such as smart mobile phone, panel computer to carry out the upper of picture and information
Pass, it is possible to by PC terminal uploading pictures and relevant information.As the main body that image is processed, clothes
Business device can be the data server that sets up in advance of system operator or the webserver.
As the tie between user and server, client is mainly used in forwarding the input content of user
To server, server the identity of user is verified by the content inputted according to user, End-Customer
The result returned according to server is held to be shown to user.Below first similar to client-side
Image-recognizing method is introduced, as it is shown in figure 5, comprise the following steps:
S501, receives the ID authentication request of user, and described ID authentication request is carried described user and uploaded
The first image to be identified and the authentication information of described user.
It should be noted that the form to client does not limit in the present embodiment, client can be PC
Equipment, it is also possible to for mobile terminal equipment.But all can provide a user with picture to upload and information input
Function, concrete, first client obtains image that described user uploads and described user input
Information, subsequently using described image as described first image to be identified, and using described information as described
Authentication information, generates described identity finally according to described first image to be identified and described authentication information and recognizes
Card request.
S502, by described ID authentication request send to server so that described server according to described in recognize
The second image to be identified that card acquisition of information is corresponding with described user.
After getting the authentication information of user, server can obtain corresponding second the treating of this user accordingly
Identify image.In the particular embodiment, the identity information that server provides according to user, data base
Image in this user identity of middle inquiry card, and using this image as the second image to be identified, so that it is determined that
Whether the image of self that user uploads mates with its ID Card Image.
S503, receives the authentication response that described server sends.
In the present embodiment, authentication response is authentication success response or authentication failure response,
Wherein authentication success response is that described server is confirming described first image to be identified and described second
The similar generation afterwards of specific characteristic of image to be identified, and authentication failure response is described server
Life after confirming the described first image to be identified specific characteristic dissmilarity with described second image to be identified
Become.
S504, shows authentication result according to described authentication response to described user.
Identity-based certification success response or authentication failure response, the specific implementation process of this step
As follows:
(1) when described client receives described authentication success response, described client is to institute
State user and show the default interface corresponding with described authentication success response;
(2) when described client receives described authentication failure response, described client is to institute
State user and show the default interface corresponding with described authentication failure response, and to described user's exhibition
Show the need of the information carrying out manual verification.
Owing to the application is automatically to be judged one group of image to be identified, therefore to enter one by equipment
Step avoids the impact that error is brought, while returning failure response to user, and can be simultaneously to user's exhibition
Show the need of the information carrying out manual verification.If the user thinks that need to resubmit manual examination and verification
If, then notifying clients input manual verification request the most again, and client is receiving described use
After manual verification's request at family, will ID authentication request send to the service end preset.
Being more than the flow process of client, be mainly used in realizing between user and server is mutual, real below
Execute the similar image recognition methods that example is server side, as shown in Figure 6, comprise the steps:
S601, receives the ID authentication request sent by described client, and described ID authentication request is carried
The first image to be identified that described user uploads and the authentication information of described user;
S602, according to the second image to be identified that the inquiry of described authentication information is corresponding with described user;
S603, obtains and corresponding with specific characteristic in the first image to be identified treats contrast district;
S604, aligns the image in described region to be identified with the standard picture preset, and by right
Image after Qi is as the normalized image of described first image to be identified, described standard picture and described finger
Determine feature corresponding;
S605, determines the tolerance between the normalized image of described normalized image and the second image to be identified
Distance, described metric range is according to described normalized image and the normalization of described second image to be identified
Image distance in feature space generates, and wherein, similar normalized image is at described feature space
Distance less than non-similar normalized image in the distance of described feature space;
S606, if described metric range is more than the threshold value preset, confirms described first image to be identified and institute
The specific characteristic stating the second image to be identified is dissimilar, and unsuccessfully rings to the return authentication of described client
Should;
S607, if described metric range is less than or equal to described threshold value, confirms described first image to be identified
Similar to the specific characteristic of described second image to be identified, and return authentication success to described client
Response.
For reaching above technical purpose, the application also proposed a kind of similar image identification equipment, such as Fig. 7
Shown in, including:
Acquisition module 710, corresponding with specific characteristic in the first image to be identified treats contrast district for obtaining
Territory;
Alignment module 720 is right for the image in described region to be identified and the standard picture preset being carried out
Together, and by the image after alignment as the normalized image of described first image to be identified, described standard drawing
As corresponding with described specific characteristic;
Determine module 730, for determining that the described normalized image of described first image to be identified is treated with second
Identifying the metric range between the normalized image of image, described metric range is according to described normalized image
And the distance that the normalized image of described second image to be identified is in feature space generates, wherein, phase
As normalized image described feature space distance less than non-similar normalized image in described feature
The distance in space;
When described metric range is more than the threshold value preset, identification module 740, for confirming that described first waits to know
Other image is dissimilar with the specific characteristic of described second image to be identified, and is less than at described metric range
Or during equal to described threshold value, confirm that described first image to be identified is special with the appointment of described second image to be identified
Levy similar.
In concrete application scenarios, described determine module specifically for:
Described in determining in described first image to be identified according to the detection algorithm corresponding with described specific characteristic
Treat contrast district, by default key point regression model obtain described in treat in contrast district with described appointment
The key point coordinate that multiple key point features of feature are corresponding.
In concrete application scenarios, described alignment module specifically for:
According to parameter M, each key point coordinate in described region to be identified is mapped as the pass of the image after aliging
Key point coordinates, wherein, each key point coordinate of described standard picture and having marked according to described parameter M
The key point Coordinate generation of image corresponding with described specific characteristic in note image.
In concrete application scenarios, also include:
Adjusting module, for the resolution extremely preset by the resolution adjustment of described normalized image.
In concrete application scenarios, described acquisition module specifically for:
The specific characteristic in described normalized image is extracted by convolutional neural networks;
Determine that described specific characteristic is mapping to spy according to convolutional neural networks and distance metric loss function
Levy the eigenvalue behind space, and using described eigenvalue as the eigenvalue of described normalized image;
Determine the spy of the eigenvalue of described normalized image and the normalized image of described second image to be identified
Euclidean distance between value indicative, using described Euclidean distance as described metric range.
In concrete application scenarios, described specific characteristic is specially face area, described key point feature
At least include left eye region, right eye region, nasal area, left corners of the mouth region and right corners of the mouth region.
In concrete application scenarios, described convolutional neural networks parameter is to train according to marking image
Arriving, the described image that marked includes the most similar normalized image of specific characteristic and specific characteristic the most not
Similar normalized image.
The application also proposed a kind of client, as shown in Figure 8, and including:
Receiver module 810, for receiving the ID authentication request of user, described ID authentication request carries institute
State the first image to be identified and the authentication information of described user that user uploads;
Sending module 820, for described ID authentication request is sent to server, so that described server
The second to be identified image corresponding with described user is obtained according to described authentication information;
Described receiver module 810, is additionally operable to receive the authentication response that described server sends;
Display module 830, for showing authentication result according to described authentication response to described user.
In concrete application scenarios, described receiver module specifically for:
Obtain image and the information of described user input that described user uploads, using described image as institute
State the first image to be identified, and using described information as described authentication information, wait to know according to described first
Other image and described authentication information generate described ID authentication request.
In concrete application scenarios, described authentication response is recognized for authentication success response or identity
Card failure response, also includes:
Described authentication success response is that described server is confirming that described first image to be identified is with described
The similar generation afterwards of specific characteristic of the second image to be identified;
Described authentication failure response is that described server is confirming that described first image to be identified is with described
Generate after the specific characteristic dissmilarity of the second image to be identified.
In concrete application scenarios, described display module, specifically for receiving when described receiver module
During to described authentication success response, show default successfully ringing with described authentication to described user
Answer corresponding interface;Or, described display module, specifically for receiving described when described receiver module
During authentication failure response, show default corresponding with described authentication failure response to described user
Interface, and to described user show the need of the information carrying out manual verification.
In concrete application scenarios, when described receiver module is shown to described user at described display module
Preset the interface corresponding with described authentication failure response and to described user show the need of entering
After the prompting of row manual verification, also receive manual verification's request of described user, described receiver module
Described sending module is indicated to send described ID authentication request to the service end preset.
The embodiment of the present application also proposed a kind of server, as it is shown in figure 9, include:
Receiver module 910, the ID authentication request sent by described client with reception, described authentication
The first image to be identified and the authentication information of described user that described user uploads is carried in request;
Enquiry module 920, to be identified for inquiring about second corresponding with described user according to described authentication information
Image;
Acquisition module 930, corresponding with specific characteristic in the first image to be identified treats contrast district for obtaining
Territory;
Alignment module 940 is right for the image in described region to be identified and the standard picture preset being carried out
Together, and by the image after alignment as the normalized image of described first image to be identified, described standard drawing
As corresponding with described specific characteristic;
Determine module 950, for determining that the described normalized image of described first image to be identified is treated with second
Identifying the metric range between the normalized image of image, described metric range is according to described normalized image
And the distance that the normalized image of described second image to be identified is in feature space generates, wherein, phase
As normalized image described feature space distance less than non-similar normalized image in described feature
The distance in space;
When described metric range is more than the threshold value preset, identification module 960, for confirming that described first waits to know
Other image is dissimilar with the specific characteristic of described second image to be identified, and is less than at described metric range
Or during equal to described threshold value, confirm that described first image to be identified is special with the appointment of described second image to be identified
Levy similar;
Sending module 970, for confirming described first image to be identified and described second at described identification module
Authentication failure response is returned to described client during the specific characteristic dissmilarity of image to be identified, and
The specific characteristic of described first image to be identified and described second image to be identified is confirmed at described identification module
Authentication success response is returned to described client time similar.
In concrete application scenarios, described determine module specifically for:
Described in determining in described first image to be identified according to the detection algorithm corresponding with described specific characteristic
Treat contrast district, by default key point regression model obtain described in treat in contrast district with described appointment
The key point coordinate that multiple key point features of feature are corresponding.
In concrete application scenarios, described alignment module specifically for:
According to parameter M, each key point coordinate in described region to be identified is mapped as the pass of the image after aliging
Key point coordinates, wherein, each key point coordinate of described standard picture and having marked according to described parameter M
The key point Coordinate generation of image corresponding with described specific characteristic in note image.
In concrete application scenarios, also include:
Adjusting module, for the resolution extremely preset by the resolution adjustment of described normalized image.
In concrete application scenarios, described acquisition module specifically for:
The specific characteristic in described normalized image is extracted by convolutional neural networks;
Determine that described specific characteristic is mapping to spy according to convolutional neural networks and distance metric loss function
Levy the eigenvalue behind space, and using described eigenvalue as the eigenvalue of described normalized image;
Determine the spy of the eigenvalue of described normalized image and the normalized image of described second image to be identified
Euclidean distance between value indicative, using described Euclidean distance as described metric range.
In concrete application scenarios, described specific characteristic is specially face area, described key point feature
At least include left eye region, right eye region, nasal area, left corners of the mouth region and right corners of the mouth region.
In concrete application scenarios, described convolutional neural networks parameter is to train according to marking image
Arriving, the described image that marked includes the most similar normalized image of specific characteristic and specific characteristic the most not
Similar normalized image.
Through the above description of the embodiments, those skilled in the art is it can be understood that arrive this Shen
Please be realized by hardware, it is also possible to the mode adding necessary general hardware platform by software realizes.
Based on such understanding, the technical scheme of the application can embody with the form of software product, and this is soft
Part product can be stored in a non-volatile memory medium, and (can be CD-ROM, USB flash disk, movement be hard
Dish etc.) in, including some instructions with so that a computer equipment (can be personal computer, take
Business device, or the network equipment etc.) each implements the method described in scene to perform the application.
It will be appreciated by those skilled in the art that accompanying drawing is a schematic diagram being preferable to carry out scene, in accompanying drawing
Module or flow process not necessarily implement necessary to the application.
It will be appreciated by those skilled in the art that the module in the device implemented in scene can be according to implementing scene
Describe and carry out being distributed in the device implementing scene, it is also possible to carry out respective change and be disposed other than this enforcement
In one or more devices of scene.The module of above-mentioned enforcement scene can merge into a module, it is possible to
To be further split into multiple submodule.
Above-mentioned the application sequence number, just to describing, does not represent the quality implementing scene.
The several scenes that are embodied as being only the application disclosed above, but, the application is not limited to
This, the changes that any person skilled in the art can think of all should fall into the protection domain of the application.
Claims (25)
1. a similar image recognition methods, it is characterised in that including:
Obtain and corresponding with specific characteristic in the first image to be identified treat contrast district;
Image in described region to be identified is alignd with the standard picture preset, and by after alignment
Image is as the normalized image of described first image to be identified, described standard picture and described specific characteristic
Corresponding;
Determine the metric range between the normalized image of described normalized image and the second image to be identified,
Described metric range exists according to the normalized image of described normalized image and described second image to be identified
Distance in feature space generates, and wherein, similar normalized image is little in the distance of described feature space
In non-similar normalized image in the distance of described feature space;
If described metric range is more than the threshold value preset, confirm described first image to be identified and described second
The specific characteristic of image to be identified is dissimilar;
If described metric range is less than or equal to described threshold value, confirm that described first image to be identified is with described
The specific characteristic of the second image to be identified is similar.
2. the method for claim 1, it is characterised in that with finger in acquisition the first image to be identified
Determine feature corresponding treat contrast district, particularly as follows:
Described in determining in described first image to be identified according to the detection algorithm corresponding with described specific characteristic
Treat contrast district;
Many with described specific characteristic by treating in contrast district described in default key point regression model acquisition
The key point coordinate that individual key point feature is corresponding.
3. method as claimed in claim 2, it is characterised in that by the image in described region to be identified
Align with default standard picture, particularly as follows:
According to parameter M, each key point coordinate in described region to be identified is mapped as the pass of the image after aliging
Key point coordinates;
Wherein, each key point coordinate of described standard picture and marked image according to described parameter M
In the key point Coordinate generation of the image corresponding with described specific characteristic.
4. method as claimed in claim 3, it is characterised in that the image after aliging is as described
After the normalized image of the first image to be identified, also include:
By the resolution adjustment of described normalized image to the resolution preset.
5. the method for claim 1, it is characterised in that determine described normalized image and second
Metric range between the normalized image of image to be identified, particularly as follows:
The specific characteristic in described normalized image is extracted by convolutional neural networks;
Determine that described specific characteristic is mapping to spy according to convolutional neural networks and distance metric loss function
Levy the eigenvalue behind space, and using described eigenvalue as the eigenvalue of described normalized image;
Determine the spy of the eigenvalue of described normalized image and the normalized image of described second image to be identified
Euclidean distance between value indicative, using described Euclidean distance as described metric range.
6. the method as described in any one of claim 1-5, it is characterised in that described specific characteristic is concrete
For face area, described key point feature at least includes left eye region, right eye region, nasal area, a left side
Corners of the mouth region and right corners of the mouth region.
7. method as claimed in claim 6, it is characterised in that
Described convolutional neural networks parameter is to have obtained according to having marked image training, described has marked image bag
Include the most similar normalized image of specific characteristic and the mutual dissimilar normalized image of specific characteristic.
8. a similar image identification equipment, it is characterised in that including:
Acquisition module, corresponding with specific characteristic in the first image to be identified treats contrast district for obtaining;
Alignment module, for the image in described region to be identified is alignd with the standard picture preset,
And by the image after alignment as the normalized image of described first image to be identified, described standard picture with
Described specific characteristic is corresponding;
Determine module, for determining that the described normalized image of described first image to be identified is waited to know with second
Metric range between the normalized image of other image, described metric range according to described normalized image with
And the distance that the normalized image of described second image to be identified is in feature space generates, wherein, similar
Normalized image empty in described feature less than non-similar normalized image in the distance of described feature space
Between distance;
Identification module, described first to be identified for confirming when described metric range is more than the threshold value preset
The specific characteristic of image and described second image to be identified is dissimilar, and described metric range less than or
The specific characteristic of described first image to be identified and described second image to be identified is confirmed during equal to described threshold value
Similar.
9. equipment as claimed in claim 8, it is characterised in that described determine module specifically for:
Described in determining in described first image to be identified according to the detection algorithm corresponding with described specific characteristic
Treat contrast district, by default key point regression model obtain described in treat in contrast district with described appointment
The key point coordinate that multiple key point features of feature are corresponding.
10. equipment as claimed in claim 9, it is characterised in that described alignment module specifically for:
According to parameter M, each key point coordinate in described region to be identified is mapped as the pass of the image after aliging
Key point coordinates, wherein, each key point coordinate of described standard picture and having marked according to described parameter M
The key point Coordinate generation of image corresponding with described specific characteristic in note image.
11. equipment as claimed in claim 8, it is characterised in that described acquisition module specifically for:
The specific characteristic in described normalized image is extracted by convolutional neural networks;
Determine that described specific characteristic is mapping to spy according to convolutional neural networks and distance metric loss function
Levy the eigenvalue behind space, and using described eigenvalue as the eigenvalue of described normalized image;
Determine the spy of the eigenvalue of described normalized image and the normalized image of described second image to be identified
Euclidean distance between value indicative, using described Euclidean distance as described metric range.
12. 1 kinds of similar image recognition methodss, are applied to client, it is characterised in that the method includes:
Receiving the ID authentication request of user, described ID authentication request carries first that described user uploads
Image to be identified and the authentication information of described user;
Described ID authentication request is sent to server, so that described server is according to described authentication information
Obtain the second to be identified image corresponding with described user;
Receive the authentication response that described server sends;
Authentication result is shown to described user according to described authentication response.
13. methods as claimed in claim 12, it is characterised in that receive the ID authentication request of user,
Particularly as follows:
Obtain image and the information of described user input that described user uploads;
Using described image as described first image to be identified, and described information is believed as described certification
Breath;
Described ID authentication request is generated according to described first image to be identified and described authentication information.
14. methods as claimed in claim 13, it is characterised in that described authentication response is identity
Certification success response or authentication failure response, also include:
Described authentication success response is that described server is confirming that described first image to be identified is with described
The similar generation afterwards of specific characteristic of the second image to be identified;
Described authentication failure response is that described server is confirming that described first image to be identified is with described
Generate after the specific characteristic dissmilarity of the second image to be identified.
15. methods as claimed in claim 14, it is characterised in that according to described authentication response to
Described user shows authentication result, particularly as follows:
When receiving described authentication success response, show default with described identity to described user
The interface that certification success response is corresponding;
When receiving described authentication failure response, show default with described identity to described user
The interface that authentication failure response is corresponding, and show the need of carrying out carrying of manual verification to described user
Show information.
16. methods as claimed in claim 15, it is characterised in that showing default to described user
The interface corresponding with described authentication failure response and show the need of carrying out artificial to described user
After the prompting of checking, also include:
If receiving manual verification's request of described user, the transmission of described ID authentication request is extremely preset
Service end.
17. 1 kinds of clients, it is characterised in that including:
Receiver module, for receiving the ID authentication request of user, described ID authentication request is carried described
The first image to be identified that user uploads and the authentication information of described user;
Sending module, for described ID authentication request is sent to server, so that described server root
The second to be identified image corresponding with described user is obtained according to described authentication information;
Described receiver module, is additionally operable to receive the authentication response that described server sends;
Display module, for showing authentication result according to described authentication response to described user.
18. 1 kinds of similar image recognition methodss, are applied to server, it is characterised in that the method includes:
Receiving the ID authentication request sent by client, described ID authentication request carries what user uploaded
First image to be identified and the authentication information of described user;
According to the second image to be identified that the inquiry of described authentication information is corresponding with described user;
Obtain and corresponding with specific characteristic in the first image to be identified treat contrast district;
Image in described region to be identified is alignd with the standard picture preset, and by after alignment
Image is as the normalized image of described first image to be identified, described standard picture and described specific characteristic
Corresponding;
Determine the metric range between the normalized image of described normalized image and the second image to be identified,
Described metric range exists according to the normalized image of described normalized image and described second image to be identified
Distance in feature space generates, and wherein, similar normalized image is little in the distance of described feature space
In non-similar normalized image in the distance of described feature space;
If described metric range is more than the threshold value preset, confirm described first image to be identified and described second
The specific characteristic of image to be identified is dissimilar, and returns authentication failure response to described client;
If described metric range is less than or equal to described threshold value, confirm that described first image to be identified is with described
The specific characteristic of the second image to be identified is similar, and returns authentication success response to described client.
19. methods as claimed in claim 18, it is characterised in that obtain in the first image to be identified with
What specific characteristic was corresponding treats contrast district, particularly as follows:
Described in determining in described first image to be identified according to the detection algorithm corresponding with described specific characteristic
Treat contrast district;
Many with described specific characteristic by treating in contrast district described in default key point regression model acquisition
The key point coordinate that individual key point feature is corresponding.
20. methods as claimed in claim 19, it is characterised in that by the figure in described region to be identified
As aliging with default standard picture, particularly as follows:
According to parameter M, each key point coordinate in described region to be identified is mapped as the pass of the image after aliging
Key point coordinates;
Wherein, each key point coordinate of described standard picture and marked image according to described parameter M
In the key point Coordinate generation of the image corresponding with described specific characteristic.
21. methods as claimed in claim 18, it is characterised in that determine described normalized image and the
Metric range between the normalized image of two images to be identified, particularly as follows:
The specific characteristic in described normalized image is extracted by convolutional neural networks;
Determine that described specific characteristic is mapping to spy according to convolutional neural networks and distance metric loss function
Levy the eigenvalue behind space, and using described eigenvalue as the eigenvalue of described normalized image;
Determine the spy of the eigenvalue of described normalized image and the normalized image of described second image to be identified
Euclidean distance between value indicative, using described Euclidean distance as described metric range.
22. 1 kinds of servers, it is characterised in that including:
Receiver module, the ID authentication request sent by described client with reception, described authentication please
Ask and carry the first image to be identified and the authentication information of described user that described user uploads;
Enquiry module, for the second to be identified figure corresponding with described user according to the inquiry of described authentication information
Picture;
Acquisition module, corresponding with specific characteristic in the first image to be identified treats contrast district for obtaining;
Alignment module, for the image in described region to be identified is alignd with the standard picture preset,
And by the image after alignment as the normalized image of described first image to be identified, described standard picture with
Described specific characteristic is corresponding;
Determine module, for determining that the described normalized image of described first image to be identified is waited to know with second
Metric range between the normalized image of other image, described metric range according to described normalized image with
And the distance that the normalized image of described second image to be identified is in feature space generates, wherein, similar
Normalized image empty in described feature less than non-similar normalized image in the distance of described feature space
Between distance;
Identification module, described first to be identified for confirming when described metric range is more than the threshold value preset
The specific characteristic of image and described second image to be identified is dissimilar, and described metric range less than or
The specific characteristic of described first image to be identified and described second image to be identified is confirmed during equal to described threshold value
Similar;
At described identification module, sending module, for confirming that described first image to be identified is treated with described second
Authentication failure response, Yi Ji is returned to described client when identifying the specific characteristic dissmilarity of image
Described identification module confirms the specific characteristic phase of described first image to be identified and described second image to be identified
Like time to described client return authentication success response.
23. servers as claimed in claim 22, it is characterised in that described determine module specifically for:
Described in determining in described first image to be identified according to the detection algorithm corresponding with described specific characteristic
Treat contrast district, by default key point regression model obtain described in treat in contrast district with described appointment
The key point coordinate that multiple key point features of feature are corresponding.
24. servers as claimed in claim 23, it is characterised in that described alignment module specifically for:
According to parameter M, each key point coordinate in described region to be identified is mapped as the pass of the image after aliging
Key point coordinates, wherein, each key point coordinate of described standard picture and having marked according to described parameter M
The key point Coordinate generation of image corresponding with described specific characteristic in note image.
25. servers as claimed in claim 22, it is characterised in that described acquisition module specifically for:
The specific characteristic in described normalized image is extracted by convolutional neural networks;
Determine that described specific characteristic is mapping to spy according to convolutional neural networks and distance metric loss function
Levy the eigenvalue behind space, and using described eigenvalue as the eigenvalue of described normalized image;
Determine the spy of the eigenvalue of described normalized image and the normalized image of described second image to be identified
Euclidean distance between value indicative, using described Euclidean distance as described metric range.
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510229654.6A CN106203242B (en) | 2015-05-07 | 2015-05-07 | Similar image identification method and equipment |
PCT/CN2016/079158 WO2016177259A1 (en) | 2015-05-07 | 2016-04-13 | Similar image recognition method and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510229654.6A CN106203242B (en) | 2015-05-07 | 2015-05-07 | Similar image identification method and equipment |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106203242A true CN106203242A (en) | 2016-12-07 |
CN106203242B CN106203242B (en) | 2019-12-24 |
Family
ID=57217488
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510229654.6A Active CN106203242B (en) | 2015-05-07 | 2015-05-07 | Similar image identification method and equipment |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN106203242B (en) |
WO (1) | WO2016177259A1 (en) |
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107451965A (en) * | 2017-07-24 | 2017-12-08 | 深圳市智美达科技股份有限公司 | Distort face image correcting method, device, computer equipment and storage medium |
CN108012080A (en) * | 2017-12-04 | 2018-05-08 | 广东欧珀移动通信有限公司 | Image processing method, device, electronic equipment and computer-readable recording medium |
WO2018137358A1 (en) * | 2017-01-24 | 2018-08-02 | 北京大学 | Deep metric learning-based accurate target retrieval method |
CN108573201A (en) * | 2017-03-13 | 2018-09-25 | 金德奎 | A kind of user identity identification matching process based on face recognition technology |
CN108921209A (en) * | 2018-06-21 | 2018-11-30 | 杭州骑轻尘信息技术有限公司 | Image identification method, device and electronic equipment |
CN108932727A (en) * | 2017-12-29 | 2018-12-04 | 浙江宇视科技有限公司 | Face tracking method and device |
CN109429519A (en) * | 2017-06-30 | 2019-03-05 | 北京嘀嘀无限科技发展有限公司 | System and method for verifying the authenticity of certificate photograph |
CN109459873A (en) * | 2018-11-12 | 2019-03-12 | 广州小鹏汽车科技有限公司 | A kind of test method, device, Auto-Test System and storage medium |
WO2019149186A1 (en) * | 2018-02-01 | 2019-08-08 | 北京京东尚科信息技术有限公司 | Method and apparatus for generating information |
CN110781917A (en) * | 2019-09-18 | 2020-02-11 | 北京三快在线科技有限公司 | Method and device for detecting repeated image, electronic equipment and readable storage medium |
US10636139B2 (en) | 2017-02-15 | 2020-04-28 | Htc Corporation | Image processing apparatus utilizing keypoints in an image analysis and method thereof |
CN112508109A (en) * | 2020-12-10 | 2021-03-16 | 锐捷网络股份有限公司 | Training method and device for image recognition model |
CN112560971A (en) * | 2020-12-21 | 2021-03-26 | 上海明略人工智能(集团)有限公司 | Image classification method and system for active learning self-iteration |
CN114359933A (en) * | 2021-11-18 | 2022-04-15 | 珠海读书郎软件科技有限公司 | Cover image identification method |
US11610391B2 (en) | 2019-12-30 | 2023-03-21 | Industrial Technology Research Institute | Cross-domain image comparison method and system using semantic segmentation |
Families Citing this family (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113688737A (en) * | 2017-12-15 | 2021-11-23 | 北京市商汤科技开发有限公司 | Face image processing method, face image processing device, electronic apparatus, storage medium, and program |
CN108804996B (en) * | 2018-03-27 | 2022-03-04 | 腾讯科技(深圳)有限公司 | Face verification method and device, computer equipment and storage medium |
CN109345770A (en) * | 2018-11-14 | 2019-02-15 | 深圳市尼欧科技有限公司 | A kind of child leaves in-vehicle alarm system and child leaves interior alarm method |
CN110084161B (en) * | 2019-04-17 | 2023-04-18 | 中山大学 | Method and system for rapidly detecting key points of human skeleton |
CN112464689A (en) * | 2019-09-06 | 2021-03-09 | 佳能株式会社 | Method, device and system for generating neural network and storage medium for storing instructions |
CN111079644B (en) * | 2019-12-13 | 2023-06-06 | 四川新网银行股份有限公司 | Method for assisting photographing based on distance and joint point identification external force and storage medium |
CN115393405A (en) * | 2021-05-21 | 2022-11-25 | 北京字跳网络技术有限公司 | Image alignment method and device |
CN113744769A (en) * | 2021-09-06 | 2021-12-03 | 盐城市聚云网络科技有限公司 | Storage device for computer information security product |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101510257A (en) * | 2009-03-31 | 2009-08-19 | 华为技术有限公司 | Human face similarity degree matching method and device |
CN102375970A (en) * | 2010-08-13 | 2012-03-14 | 北京中星微电子有限公司 | Identity authentication method based on face and authentication apparatus thereof |
CN103678984A (en) * | 2013-12-20 | 2014-03-26 | 湖北微模式科技发展有限公司 | Method for achieving user authentication by utilizing camera |
CN103824051A (en) * | 2014-02-17 | 2014-05-28 | 北京旷视科技有限公司 | Local region matching-based face search method |
Family Cites Families (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6738063B2 (en) * | 2002-02-07 | 2004-05-18 | Siemens Corporate Research, Inc. | Object-correspondence identification without full volume registration |
CN101669824B (en) * | 2009-09-22 | 2012-01-25 | 浙江工业大学 | Biometrics-based device for detecting indentity of people and identification |
CN102629320B (en) * | 2012-03-27 | 2014-08-27 | 中国科学院自动化研究所 | Ordinal measurement statistical description face recognition method based on feature level |
CN103207898B (en) * | 2013-03-19 | 2016-08-03 | 天格科技(杭州)有限公司 | A kind of similar face method for quickly retrieving based on local sensitivity Hash |
-
2015
- 2015-05-07 CN CN201510229654.6A patent/CN106203242B/en active Active
-
2016
- 2016-04-13 WO PCT/CN2016/079158 patent/WO2016177259A1/en active Application Filing
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101510257A (en) * | 2009-03-31 | 2009-08-19 | 华为技术有限公司 | Human face similarity degree matching method and device |
CN102375970A (en) * | 2010-08-13 | 2012-03-14 | 北京中星微电子有限公司 | Identity authentication method based on face and authentication apparatus thereof |
CN103678984A (en) * | 2013-12-20 | 2014-03-26 | 湖北微模式科技发展有限公司 | Method for achieving user authentication by utilizing camera |
CN103824051A (en) * | 2014-02-17 | 2014-05-28 | 北京旷视科技有限公司 | Local region matching-based face search method |
Non-Patent Citations (1)
Title |
---|
MOHAMED KHALIL-HANI ETC.: "A Convolutional Neural Network Approach for Face Verification", 《2014 INTERNATIONAL CONFERENCE ON HIGH PERFORMANCE COMPUTING & SIMULATION (HPCS)》 * |
Cited By (23)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2018137358A1 (en) * | 2017-01-24 | 2018-08-02 | 北京大学 | Deep metric learning-based accurate target retrieval method |
US11080845B2 (en) | 2017-02-15 | 2021-08-03 | Htc Corporation | Image processing apparatus utilizing keypoints in an image analysis and method thereof |
TWI695324B (en) * | 2017-02-15 | 2020-06-01 | 宏達國際電子股份有限公司 | Image processing apparatus and method thereof |
US10636139B2 (en) | 2017-02-15 | 2020-04-28 | Htc Corporation | Image processing apparatus utilizing keypoints in an image analysis and method thereof |
CN108573201A (en) * | 2017-03-13 | 2018-09-25 | 金德奎 | A kind of user identity identification matching process based on face recognition technology |
CN109429519A (en) * | 2017-06-30 | 2019-03-05 | 北京嘀嘀无限科技发展有限公司 | System and method for verifying the authenticity of certificate photograph |
CN107451965A (en) * | 2017-07-24 | 2017-12-08 | 深圳市智美达科技股份有限公司 | Distort face image correcting method, device, computer equipment and storage medium |
CN107451965B (en) * | 2017-07-24 | 2019-10-18 | 深圳市智美达科技股份有限公司 | Distort face image correcting method, device, computer equipment and storage medium |
CN108012080B (en) * | 2017-12-04 | 2020-02-04 | Oppo广东移动通信有限公司 | Image processing method, image processing device, electronic equipment and computer readable storage medium |
CN108012080A (en) * | 2017-12-04 | 2018-05-08 | 广东欧珀移动通信有限公司 | Image processing method, device, electronic equipment and computer-readable recording medium |
CN108932727A (en) * | 2017-12-29 | 2018-12-04 | 浙江宇视科技有限公司 | Face tracking method and device |
WO2019149186A1 (en) * | 2018-02-01 | 2019-08-08 | 北京京东尚科信息技术有限公司 | Method and apparatus for generating information |
US11250292B2 (en) | 2018-02-01 | 2022-02-15 | Beijing Jingdong Shangke Information Technology Co., Ltd. | Method and apparatus for generating information |
CN108921209A (en) * | 2018-06-21 | 2018-11-30 | 杭州骑轻尘信息技术有限公司 | Image identification method, device and electronic equipment |
CN109459873A (en) * | 2018-11-12 | 2019-03-12 | 广州小鹏汽车科技有限公司 | A kind of test method, device, Auto-Test System and storage medium |
CN110781917A (en) * | 2019-09-18 | 2020-02-11 | 北京三快在线科技有限公司 | Method and device for detecting repeated image, electronic equipment and readable storage medium |
CN110781917B (en) * | 2019-09-18 | 2021-03-02 | 北京三快在线科技有限公司 | Method and device for detecting repeated image, electronic equipment and readable storage medium |
US11610391B2 (en) | 2019-12-30 | 2023-03-21 | Industrial Technology Research Institute | Cross-domain image comparison method and system using semantic segmentation |
TWI797420B (en) * | 2019-12-30 | 2023-04-01 | 財團法人工業技術研究院 | Cross-domain image comparison method and system |
CN112508109A (en) * | 2020-12-10 | 2021-03-16 | 锐捷网络股份有限公司 | Training method and device for image recognition model |
CN112560971A (en) * | 2020-12-21 | 2021-03-26 | 上海明略人工智能(集团)有限公司 | Image classification method and system for active learning self-iteration |
CN114359933A (en) * | 2021-11-18 | 2022-04-15 | 珠海读书郎软件科技有限公司 | Cover image identification method |
CN114359933B (en) * | 2021-11-18 | 2022-09-20 | 珠海读书郎软件科技有限公司 | Cover image identification method |
Also Published As
Publication number | Publication date |
---|---|
CN106203242B (en) | 2019-12-24 |
WO2016177259A1 (en) | 2016-11-10 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106203242A (en) | A kind of similar image recognition methods and equipment | |
Khan et al. | Deep unified model for face recognition based on convolution neural network and edge computing | |
CN106897675B (en) | Face living body detection method combining binocular vision depth characteristic and apparent characteristic | |
CN107688823B (en) | A kind of characteristics of image acquisition methods and device, electronic equipment | |
WO2022161286A1 (en) | Image detection method, model training method, device, medium, and program product | |
CN103530599B (en) | The detection method and system of a kind of real human face and picture face | |
WO2020107847A1 (en) | Bone point-based fall detection method and fall detection device therefor | |
CN105205453B (en) | Human eye detection and localization method based on depth self-encoding encoder | |
CN111091075B (en) | Face recognition method and device, electronic equipment and storage medium | |
CN102629320B (en) | Ordinal measurement statistical description face recognition method based on feature level | |
KR20160083900A (en) | Systems and methods for facial representation | |
CN105447532A (en) | Identity authentication method and device | |
CN110765882B (en) | Video tag determination method, device, server and storage medium | |
CN112132099A (en) | Identity recognition method, palm print key point detection model training method and device | |
CN113515988B (en) | Palm print recognition method, feature extraction model training method, device and medium | |
CN106303599A (en) | A kind of information processing method, system and server | |
KR20180038169A (en) | Safety classification method of the city image using deep learning-based data feature | |
US20220327189A1 (en) | Personalized biometric anti-spoofing protection using machine learning and enrollment data | |
CN108549883A (en) | A kind of face recognition methods again | |
CN111507320A (en) | Detection method, device, equipment and storage medium for kitchen violation behaviors | |
CN116152870A (en) | Face recognition method, device, electronic equipment and computer readable storage medium | |
CN117058723B (en) | Palmprint recognition method, palmprint recognition device and storage medium | |
CN103984918A (en) | Human face image recognition method based on intra-class and inter-class variation | |
TWI617999B (en) | Double identification door access system and method thereof | |
Lin et al. | A lightweight face verification based on adaptive cascade network and triplet loss function |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |