CN106599855A - Softmax-based face recognizing method - Google Patents
Softmax-based face recognizing method Download PDFInfo
- Publication number
- CN106599855A CN106599855A CN201611175952.2A CN201611175952A CN106599855A CN 106599855 A CN106599855 A CN 106599855A CN 201611175952 A CN201611175952 A CN 201611175952A CN 106599855 A CN106599855 A CN 106599855A
- Authority
- CN
- China
- Prior art keywords
- face
- softmax
- test sample
- dimension
- matrix
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/168—Feature extraction; Face representation
Landscapes
- Engineering & Computer Science (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Human Computer Interaction (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
- Collating Specific Patterns (AREA)
Abstract
The invention relates to the technology of face recognition, aiming at addressing large operation and long time consumption in using deep learning to extract the characteristics of face images. The invention provides a softmax-based face recognizing method. The technical solution includes the following steps: firstly selecting a plurality of face images of N persons as test samples, extracting higher dimensional face characteristics of the test samples, creating a higher dimensional matrix, wherein N is an integer, then adopting the softmax algorithm, training the obtained higher dimensional matrix by using a corresponding characteristic projection matrix, conducting dimensional reduction on the higher dimensional matrix, upon face recognition, separately projecting a to-be-tested face image and test samples by using the characteristic projection matrix, finally discriminating the obtained projection of the to-be-tested face image and the projection of the test samples by using the distance discrimination method to obtain the result of face recognition. According to the invention, the method has high recognition precision and is suitable for face recognition systems.
Description
Technical field
The present invention relates to face recognition technology.
Background technology
The range of application of face recognition technology is more and more wider, and has portioned product to have enter into daily life
In, but in some specific application scenarios, its accuracy of identification and speed also much can not meet the application in actual environment.It is existing
Stage increasing scholar, research institution and enterprise are put in the research of deep learning, and deep learning is in speech recognition, figure
As the fields such as classification have obtained good application, also it is unwilling to be outshone in field of face identification, constantly has within nearly 2 years pertinent literature to send out
Table, refreshes the discrimination of conventional face's recognizer, but extracts facial image characteristic operation amount greatly using deep learning, takes
It is long, it is impossible to meet and require in real time.
The content of the invention
The invention aims to solve that current deep learning extracts that facial image characteristic operation amount is big, time-consuming asks
A kind of topic, there is provided face identification method based on softmax.
The present invention solves its technical problem, and the technical scheme of employing is, based on the face identification method of softmax, it is special
Levy and be, comprise the following steps：
Step 1, some facial images of the N number of people of selection extract its higher-dimension face characteristic as test sample, create higher-dimension
Matrix, N is positive integer；
Step 2, using softmax algorithms, the higher dimensional matrix to obtaining trains corresponding Projection Character matrix, and it is dropped
Dimension；
When step 3, recognition of face, test facial image is treated respectively and test sample is thrown using Projection Character matrix
Shadow；
Step 4, the projection of facial image to be tested for obtaining and the projection of test sample are carried out using discriminant by distance
Differentiate, obtain face recognition result.
Specifically, in step 1, the mode for extracting higher-dimension face characteristic is by high-LBP algorithms or deep learning
Algorithm carries out the extraction of higher-dimension face characteristic.
Further, in step 1, some facial images for choosing N number of people extract its higher-dimension people as test sample
Face feature, creates higher dimensional matrix and refers to：The M facial image altogether of N number of people is chosen as test sample, its higher-dimension face is extracted
Feature, if the intrinsic dimensionality for extracting is P, then the higher dimensional matrix for being created is M × P, and M is positive integer, and P is positive integer.
Specifically, in step 2, the employing softmax algorithms, the higher dimensional matrix to obtaining trains corresponding Projection Character
Matrix, be to the method for its dimensionality reduction：Using softmax algorithms to higher dimensional matrix M × P process, the feature for obtaining N × P is thrown
Shadow matrix.
Further, in step 4, the discriminant by distance is Euclidean distance diagnostic method.
Specifically, in step 1, before higher-dimension face characteristic is extracted, also the face images in test sample are entered
Row pretreatment；
Before step 3, also treating test facial image carries out pretreatment.
Further, the pretreatment includes modification image size, image is corrected and illumination pretreatment.
The invention has the beneficial effects as follows, in the present invention program, by the above-mentioned face identification method based on softmax,
Dimensionality reduction can be carried out to the higher-dimension face characteristic obtained using high-LBP and deep learning algorithm etc., so as to judge identification face
Operand and time-consuming duration are reduced during image, so as to realize that real-time face is recognized, and therefore accuracy of identification can't be reduced.
Specific embodiment
With reference to embodiment, technical scheme is described in detail.
Face identification method based on softmax of the present invention is：Some facial image conducts of N number of people are chosen first
Test sample, extracts its higher-dimension face characteristic, creates higher dimensional matrix, and N is positive integer, then using softmax algorithms, to what is obtained
Higher dimensional matrix trains corresponding Projection Character matrix, to its dimensionality reduction, in recognition of face, test facial image and survey is treated respectively
This is projected sample using Projection Character matrix, finally by the projection of the facial image to be tested for obtaining and the throwing of test sample
Shadow differentiated using discriminant by distance, obtains face recognition result.
Embodiment
The face identification method based on softmax of the embodiment of the present invention, it is comprised the following steps：
Step 1, some facial images of the N number of people of selection extract its higher-dimension face characteristic as test sample, create higher-dimension
Matrix, N is positive integer.
In this step, the mode for extracting higher-dimension face characteristic is existing by high-LBP algorithms or deep learning algorithm etc.
Algorithm carries out the extraction of higher-dimension face characteristic.
Some facial images of N number of people are then chosen as test sample, its higher-dimension face characteristic is extracted, higher dimensional matrix is created
Can be specially：The M facial image altogether of N number of people is chosen as test sample, its higher-dimension face characteristic is extracted, if extract
Intrinsic dimensionality is P, then the higher dimensional matrix for being created is M × P, and M is positive integer, and P is positive integer.
Before higher-dimension face characteristic is extracted, can also pretreatment be carried out to the face images in test sample, to carry
High follow-up accuracy of identification.
Step 2, using softmax algorithms, the higher dimensional matrix to obtaining trains corresponding Projection Character matrix, and it is dropped
Dimension.
If in step 1, choosing some facial images of N number of people as test sample, its higher-dimension face characteristic is extracted, created
Higher dimensional matrix is specially：The M facial image altogether of N number of people is chosen as test sample, its higher-dimension face characteristic is extracted, if carrying
The intrinsic dimensionality for taking is P, then the higher dimensional matrix for being created is M × P, and M is positive integer, and P is positive integer.
Then in this step, using softmax algorithms, the higher dimensional matrix to obtaining trains corresponding Projection Character matrix, right
The method of its dimensionality reduction is：Using softmax algorithms to higher dimensional matrix M × P process, the Projection Character matrix of N × P is obtained.
When step 3, recognition of face, test facial image is treated respectively and test sample is thrown using Projection Character matrix
Shadow.
Before this step, can also treat test facial image carries out pretreatment.If before test sample has been carried out pre-
Process, then now need to carry out and test sample identical pretreatment, to improve accuracy of identification, if before test sample is not carried out
Pretreatment, then can also now carry out pretreatment, to improve accuracy of identification.
Step 4, the projection of facial image to be tested for obtaining and the projection of test sample are carried out using discriminant by distance
Differentiate, obtain face recognition result.
In this step, discriminant by distance can be for Euclidean distance diagnostic method etc., to by the projection of facial image to be tested
The facial image of closest people in the classification being identified as belonging to the facial image closest with it, i.e. test sample.
In this example, pretreatment includes modification image size, image is corrected and illumination pretreatment etc..
Claims (7)
1. the face identification method of softmax is based on, it is characterised in that comprised the following steps：
Step 1, some facial images of the N number of people of selection extract its higher-dimension face characteristic as test sample, create higher-dimension square
Battle array, N is positive integer；
Step 2, using softmax algorithms, the higher dimensional matrix to obtaining trains corresponding Projection Character matrix, to its dimensionality reduction；
When step 3, recognition of face, test facial image is treated respectively and test sample is projected using Projection Character matrix；
Step 4, the projection of facial image to be tested for obtaining and the projection of test sample are differentiated using discriminant by distance,
Obtain face recognition result.
2. the face identification method of softmax is based on as claimed in claim 1, it is characterised in that in step 1, the extraction
The mode of higher-dimension face characteristic is that the extraction of higher-dimension face characteristic is carried out by high-LBP algorithms or deep learning algorithm.
3. the face identification method of softmax is based on as claimed in claim 1, it is characterised in that in step 1, the selection N
Personal some facial images extract its higher-dimension face characteristic as test sample, create higher dimensional matrix and refer to：Choose N number of people
M facial image altogether as test sample, extract its higher-dimension face characteristic, if extract intrinsic dimensionality be P, then created
Higher dimensional matrix be M × P, M is positive integer, and P is positive integer.
4. the face identification method of softmax is based on as claimed in claim 3, it is characterised in that in step 2, the employing
Softmax algorithms, the higher dimensional matrix to obtaining trains corresponding Projection Character matrix, is to the method for its dimensionality reduction：Using
Softmax algorithms obtain the Projection Character matrix of N × P to higher dimensional matrix M × P process.
5. the face identification method of softmax is based on as claimed in claim 1, it is characterised in that in step 4, the distance
Diagnostic method is Euclidean distance diagnostic method.
6. the face identification method based on softmax as described in claim 1 or 2 or 3 or 4 or 5, it is characterised in that step 1
In, before higher-dimension face characteristic is extracted, also pretreatment is carried out to the face images in test sample；
Before step 3, also treating test facial image carries out pretreatment.
7. the face identification method of softmax is based on as claimed in claim 6, it is characterised in that the pretreatment includes repairing
Change plan and corrected as size, to image and illumination pretreatment.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611175952.2A CN106599855A (en) | 2016-12-19 | 2016-12-19 | Softmax-based face recognizing method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611175952.2A CN106599855A (en) | 2016-12-19 | 2016-12-19 | Softmax-based face recognizing method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106599855A true CN106599855A (en) | 2017-04-26 |
Family
ID=58601459
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611175952.2A Pending CN106599855A (en) | 2016-12-19 | 2016-12-19 | Softmax-based face recognizing method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106599855A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110472491A (en) * | 2019-07-05 | 2019-11-19 | 深圳壹账通智能科技有限公司 | Abnormal face detecting method, abnormality recognition method, device, equipment and medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103955676A (en) * | 2014-05-12 | 2014-07-30 | 苏州大学 | Human face identification method and system |
| US20150221338A1 (en) * | 2014-02-05 | 2015-08-06 | Elena Shaburova | Method for triggering events in a video |
| CN104899579A (en) * | 2015-06-29 | 2015-09-09 | 小米科技有限责任公司 | Face recognition method and face recognition device |
| CN105138993A (en) * | 2015-08-31 | 2015-12-09 | 小米科技有限责任公司 | Method and device for building face recognition model |
| CN106228201A (en) * | 2016-06-20 | 2016-12-14 | 电子科技大学 | A kind of anti-Deceiving interference method of synthetic aperture radar based on shade characteristic |
-
2016
- 2016-12-19 CN CN201611175952.2A patent/CN106599855A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150221338A1 (en) * | 2014-02-05 | 2015-08-06 | Elena Shaburova | Method for triggering events in a video |
| CN103955676A (en) * | 2014-05-12 | 2014-07-30 | 苏州大学 | Human face identification method and system |
| CN104899579A (en) * | 2015-06-29 | 2015-09-09 | 小米科技有限责任公司 | Face recognition method and face recognition device |
| CN105138993A (en) * | 2015-08-31 | 2015-12-09 | 小米科技有限责任公司 | Method and device for building face recognition model |
| CN106228201A (en) * | 2016-06-20 | 2016-12-14 | 电子科技大学 | A kind of anti-Deceiving interference method of synthetic aperture radar based on shade characteristic |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110472491A (en) * | 2019-07-05 | 2019-11-19 | 深圳壹账通智能科技有限公司 | Abnormal face detecting method, abnormality recognition method, device, equipment and medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Feng et al. | Detecting altered fingerprints | |
| Bartlett et al. | Recognizing facial expression: machine learning and application to spontaneous behavior | |
| Saudagare et al. | Facial expression recognition using neural network–An overview | |
| CN103226835B (en) | Based on method for tracking target and the system of online initialization gradient enhancement regression tree | |
| Sinha et al. | Class-wise difficulty-balanced loss for solving class-imbalance | |
| CN104318221A (en) | Facial expression recognition method based on ELM | |
| EP4075395A3 (en) | Method and apparatus of training anti-spoofing model, method and apparatus of performing anti-spoofing, and device | |
| CN104834916A (en) | Multi-face detecting and tracking method | |
| CN102156887A (en) | Human face recognition method based on local feature learning | |
| EP3905123A3 (en) | Human body identification method and apparatus, electronic device, storage medium and program product | |
| CN107885723A (en) | Conversational character differentiating method and system | |
| Orrite et al. | Hog-based decision tree for facial expression classification | |
| CN109685071A (en) | Brain electricity classification method based on the study of common space pattern feature width | |
| Si et al. | Towards robust visual question answering: Making the most of biased samples via contrastive learning | |
| Alonso-Fernandez et al. | Impact of signature legibility and signature type in off-line signature verification | |
| CN106056088A (en) | Single-sample face recognition method based on self-adaptive virtual sample generation criterion | |
| CN106599855A (en) | Softmax-based face recognizing method | |
| CN114639152A (en) | Multi-modal voice interaction method, device, equipment and medium based on face recognition | |
| CN110472051A (en) | A kind of event detecting method indicating study based on variable quantity | |
| Menon et al. | Data augmentation and transfer learning applied to charcoal image classification | |
| CN109614927A (en) | Micro- Expression Recognition based on front and back frame difference and Feature Dimension Reduction | |
| Ni et al. | Particle filter tracking with online multiple instance learning | |
| CN106203339A (en) | A kind of based on the alignment of multiple coupled differentiation localized mass across angle gait recognition method | |
| Haselhoff et al. | An evolutionary optimized vehicle tracker in collaboration with a detection system | |
| Pereira et al. | Application of genetic algorithms to improve the reliability of an iris recognition system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170426 |
|
| RJ01 | Rejection of invention patent application after publication |