CN106056041A - Near-infrared palm vein image identification method - Google Patents
Near-infrared palm vein image identification method Download PDFInfo
- Publication number
- CN106056041A CN106056041A CN201610334223.0A CN201610334223A CN106056041A CN 106056041 A CN106056041 A CN 106056041A CN 201610334223 A CN201610334223 A CN 201610334223A CN 106056041 A CN106056041 A CN 106056041A
- Authority
- CN
- China
- Prior art keywords
- image
- palm vein
- convolutional neural
- vein image
- neural networks
- 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
- G06V40/12—Fingerprints or palmprints
- G06V40/1382—Detecting the live character of the finger, i.e. distinguishing from a fake or cadaver finger
- G06V40/1388—Detecting the live character of the finger, i.e. distinguishing from a fake or cadaver finger using image processing
Abstract
The invention discloses a near-infrared palm vein image identification method. An infrared industrial camera is utilized to collect a vein image of a palm of a person, then normalization, bineryzation and median filtering are carried out on an original image, pre-processing is carried out to obtain a target image with certain characteristics, subsequently, operation such as multilayer convolution and pooling is carried out on the image for training, and a reasonable weight matrix is obtained. Proved by experience, the weight matrix can be applied to palm vein image identification in a relatively small range. The system method has the advantages that the identification speed is high in the small range, the system is simple and the identification rate is high.
Description
Technical field
The present invention is one and utilizes convolutional neural networks to process and identify the technology of digital picture.
Background technology
The world today, identification is more and more important, either arrives bank's transacting business, still takes
Aircraft, the least to opening Yishanmen, it is required for identifying identity.Traditional auth method includes certificate, key, user name
With identity contents such as passwords, due to by external thing, once mark or password are stolen or forget, its identity be easy for by
Other people pretend to be or replace.Then bio-identification receives the attention of more and more people, and the popularity rate of current fingerprint recognition is the highest.But
It is because fingerprint is easily copied, and area is less.The reliability of iris identification is high and easily copys, but its cost is the highest, mesh
Before can't popularize in a large number.Palm vein recognition speed and stability: using infrared induction, user can have the authority of oneself
Quickly scanning, and touch scanner without reality.Additionally, to biological characteristic authentication or the accuracy of scanner in most of environment
The little negative effect of reliability.Palm vein identification equipment the most in the market is the most accurate on the high side, and one
As rising of can not consuming of family and enterprise.
Summary of the invention
The present invention uses convolutional neural networks algorithm, for this field of palm vein image identification provide a kind of efficiently,
Accurate solution.
The present invention is directed to palm vein image identification adopted the technical scheme that: utilize autonomous Design based on near-infrared
The original image of the palm vein that the palm vein image harvester of camera extracts, then to the compression of images collected, dynamic
State binaryzation and enhancement process, finally the palm vein image using pretreatment carries out the training of convolutional neural networks also as input
Coupling.Palm vein identification process has following five steps:
(1) image normalization processes: is first reduced by sample image certain proportion, and carries out gray scale normalization process, sample
The standard convention of image is same variance and same average；
(2) Dynamic Binarization processes: use big law (OSTU) method to carry out palm vein image at Dynamic Binarization
Reason, i.e. divides an image into multiple region, and each region carries out binary conversion treatment respectively, can avoid again image well simultaneously
Not UNICOM and the appearance of pseudo-image；
(3) medium filtering processes: image carries out medium filtering, removes substantial amounts of noise spot and isolated point in image；
(4) convolutional neural networks training: first pass through propagation stage forward, by sample from input layer through convolutional neural networks
In conversion (process of conversion is: ground floor convolution, ground floor pond, second layer convolution, second layer pond) step by step, be sent to
Output layer.Again through the back-propagation stage, carry out error transfer factor, constantly adjust weight matrix；
(5) palm vein image identification: the palm vein image of input has been carried out convolutional Neural net in Sample Storehouse
The image of network training is compared, and draws recognition result.
Accompanying drawing explanation
Fig. 1 is sample artwork；
Fig. 2 is the image after overcompression and normalized；
Fig. 3 is the image after Dynamic Binarization processes；
Fig. 4 is the image after medium filtering processes.
Detailed description of the invention
Referring to the drawings and embodiment the present invention will be described in detail.The scope of the present invention is not by these embodiments
Restriction, the scope of the present invention proposes in detail in the claims.
The palm that the present invention utilizes the palm vein image harvester based near infrared camera of autonomous Design to extract is quiet
The original image of arteries and veins, then to the compression of images collected, Dynamic Binarization and enhancement process, finally the palm with pretreatment is quiet
Arteries and veins image carries out the training of convolutional neural networks as input and mates.Palm vein identification process has following five steps:
(1) image normalization processes: is first reduced by sample image certain proportion, and carries out gray scale normalization process, sample
The standard convention of image is same variance and same average；
(2) Dynamic Binarization processes: use big law (OSTU) method to carry out palm vein image at Dynamic Binarization
Reason, i.e. divides an image into multiple region, and each region carries out binary conversion treatment respectively, can avoid again image well simultaneously
Not UNICOM and the appearance of pseudo-image；
(3) medium filtering processes: image carries out medium filtering, removes substantial amounts of noise spot and isolated point in image；
(4) convolutional neural networks training: first pass through propagation stage forward, by sample from input layer through convolutional neural networks
In conversion (process of conversion is: ground floor convolution, ground floor pond, second layer convolution, second layer pond) step by step, be sent to
Output layer.Again through the back-propagation stage, carry out error transfer factor, constantly adjust weight matrix；
(5) palm vein image identification: the palm vein image of input has been carried out convolutional Neural net in Sample Storehouse
The image of network training is compared, and draws recognition result.
Utilize the palm vein image recognizer that the present invention proposes, Fig. 1 artwork operated, first pass around compression and
Normalized, obtains Fig. 2, to reduce operand, to improve processing speed；Next is carried out at Dynamic Binarization and medium filtering
Reason, the most as shown in Figure 3 and Figure 4, the image after finally being processed；Finally carry out convolutional neural networks training, the knot obtained
Really typing Sample Storehouse.
If user needs to be identified, only need to upload palm vein image, system can be by this image and institute in Sample Storehouse
There is the image through convolutional neural networks training to compare, it can be deduced that comparison information, identify successfully.
For the checking of the palm vein image recognition methods that we use, we use ten foldings in cross validation
Cross validation method.Being divided into by data set very, in turn by wherein 9 parts as training data, 1 part, as test data, is carried out
Test.Test all can draw corresponding accuracy (or error rate) every time.Putting down of the accuracy (or error rate) of the result of 10 times
Average is as the estimation to arithmetic accuracy, and (such as 10 times 10 foldings intersections are tested typically to also need to carry out repeatedly 10 folding cross validations
Card), then seek its average, as the estimation to algorithm accuracy.Why select to be divided into data set 10 parts, be because by profit
The lot of experiments carried out with mass data collection, the different learning art of use, shows that 10 foldings are the appropriate of the best error estimation of acquisition
Select, and have some rationales to may certify that this point.
Our data set is 3 people that we collect with infrared camera, everyone palm picture of 20.Test
During, we use 500 times for the frequency of training of convolution kernel, and concrete the result is as shown in the table:
| Misclassification rate | Sensitivity | Specificity | |
| First man | 2/22=9.09% | 20/ (20+0)=100% | 38/ (38+2)=95% |
| Second people | 1/20=5% | 19/ (19+1)=95% | 39/ (39+1)=97.5% |
| 3rd people | 0/18=0% | 18/ (18+2)=90% | 40/ (40+0)=100% |
| Amount to | 4.70% | 95% | 97.5% |
Misclassification rate, refers to it is not the palm photo of this people, but is not identified as the percentage ratio of the palm photo of this people.
Sensitivity, refers to it is the palm photo of this people, the percentage ratio being correctly validated.
Specificity, refers to it is not the palm photo of this people, and be identified as is not the percentage ratio of this people's palm photo simultaneously yet.
Claims (1)
1. the present invention utilizes the palm vein that the palm vein image harvester based near infrared camera of autonomous Design extracts
Original image, then to the compression of images collected, Dynamic Binarization and enhancement process, finally with the palm vein of pretreatment
Image carries out the training of convolutional neural networks as input and mates.
Palm vein image identification process has following five steps:
(1) image normalization processes: is first reduced by sample image certain proportion, and carries out gray scale normalization process, sample image
Standard convention be same variance and same average；
(2) Dynamic Binarization processes: use big law (OSTU) method palm vein image to be carried out Dynamic Binarization process, i.e.
Divide an image into multiple region, each region is carried out respectively binary conversion treatment, image can be avoided well not join again simultaneously
The appearance of logical and pseudo-image；
(3) medium filtering processes: image carries out medium filtering, removes substantial amounts of noise spot and isolated point in image；
(4) convolutional neural networks training: first pass through propagation stage forward, by sample from input layer through convolutional neural networks by
The conversion (process of conversion is: ground floor convolution, ground floor pond, second layer convolution, second layer pond) of level, is sent to output
Layer.Again through the back-propagation stage, carry out error transfer factor, constantly adjust weight matrix；
(5) palm vein image identification: the palm vein image of input has been carried out convolutional neural networks instruction in Sample Storehouse
The image practiced is compared, and draws recognition result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610334223.0A CN106056041A (en) | 2016-05-18 | 2016-05-18 | Near-infrared palm vein image identification method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610334223.0A CN106056041A (en) | 2016-05-18 | 2016-05-18 | Near-infrared palm vein image identification method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106056041A true CN106056041A (en) | 2016-10-26 |
Family
ID=57177172
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610334223.0A Pending CN106056041A (en) | 2016-05-18 | 2016-05-18 | Near-infrared palm vein image identification method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106056041A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107256395A (en) * | 2017-06-12 | 2017-10-17 | 成都芯软科技股份公司 | Vena metacarpea extracting method and device |
| CN109034034A (en) * | 2018-07-12 | 2018-12-18 | 广州麦仑信息科技有限公司 | A kind of vein identification method based on nitrification enhancement optimization convolutional neural networks |
| WO2022005337A1 (en) * | 2020-06-29 | 2022-01-06 | Autonomous Non-Profit Organization For Higher Education «Skolkovo Institute Of Science And Technology» | Veins mask projection alignment |
| WO2022005336A1 (en) * | 2020-06-29 | 2022-01-06 | Autonomous Non-Profit Organization For Higher Education «Skolkovo Institute Of Science And Technology» | Noise-resilient vasculature localization method with regularized segmentation |
-
2016
- 2016-05-18 CN CN201610334223.0A patent/CN106056041A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107256395A (en) * | 2017-06-12 | 2017-10-17 | 成都芯软科技股份公司 | Vena metacarpea extracting method and device |
| CN109034034A (en) * | 2018-07-12 | 2018-12-18 | 广州麦仑信息科技有限公司 | A kind of vein identification method based on nitrification enhancement optimization convolutional neural networks |
| WO2022005337A1 (en) * | 2020-06-29 | 2022-01-06 | Autonomous Non-Profit Organization For Higher Education «Skolkovo Institute Of Science And Technology» | Veins mask projection alignment |
| WO2022005336A1 (en) * | 2020-06-29 | 2022-01-06 | Autonomous Non-Profit Organization For Higher Education «Skolkovo Institute Of Science And Technology» | Noise-resilient vasculature localization method with regularized segmentation |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| De Souza et al. | Deep texture features for robust face spoofing detection | |
| WO2019200781A1 (en) | Receipt recognition method and device, and storage medium | |
| US8369595B1 (en) | Texture features for biometric authentication | |
| Jain et al. | Handwritten signature verification using shallow convolutional neural network | |
| CN110269625B (en) | Novel multi-feature fusion electrocardio authentication method and system | |
| CN106056041A (en) | Near-infrared palm vein image identification method | |
| TW201211913A (en) | A method for recognizing the identity of user by palm vein biometric | |
| Liu et al. | One-class fingerprint presentation attack detection using auto-encoder network | |
| CN105760841A (en) | Identify recognition method and identify recognition system | |
| Stojanović et al. | A novel neural network based approach to latent overlapped fingerprints separation | |
| Hu et al. | Toward driver face recognition in the intelligent traffic monitoring systems | |
| CN106529395A (en) | Signature image recognition method based on deep brief network and k-means clustering | |
| Xu et al. | An effective recognition approach for contactless palmprint | |
| CN110147769B (en) | Finger vein image matching method | |
| Li | Feature extraction, recognition, and matching of damaged fingerprint: Application of deep learning network | |
| WO2021027155A1 (en) | Verification method and apparatus based on finger vein image, and storage medium and computer device | |
| Bakheet et al. | Robust fingerprint minutiae extraction and matching based on improved SIFT features | |
| Patil et al. | Fingerprint classification using artificial neural network | |
| CN111062345B (en) | Training method and device for vein recognition model and vein image recognition device | |
| Latha et al. | A novel method for person authentication using retinal images | |
| Hsia et al. | A new enhancement edge detection of finger-vein identification for carputer system | |
| Bharath et al. | A novel multimodal hand database for biometric authentication | |
| Zhong et al. | Hand dorsal vein recognition based on deep hash network | |
| Jain et al. | Palm vein technology for biometrics | |
| Yang et al. | Efficient method for high-resolution fingerprint image enhancement using deep residual network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20161026 |