CN109145799A - A kind of object discrimination method based on multi-layer information - Google Patents
A kind of object discrimination method based on multi-layer information Download PDFInfo
- Publication number
- CN109145799A CN109145799A CN201810918127.XA CN201810918127A CN109145799A CN 109145799 A CN109145799 A CN 109145799A CN 201810918127 A CN201810918127 A CN 201810918127A CN 109145799 A CN109145799 A CN 109145799A
- Authority
- CN
- China
- Prior art keywords
- sample
- image
- examined
- feature
- master
- 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
-
- 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
- G06V40/172—Classification, e.g. identification
Abstract
The invention discloses a kind of object discrimination method based on multi-layer information, it include: that master sample feature is established, acquire the multispectral image of collected master sample, and extract the feature of multispectral image, spectrum includes at least visible light and ultraviolet spectra or/infrared spectroscopy, to obtain the feature on master sample surface layer and internal layer;The Image Acquisition of sample to be examined, collects the multispectral image of sample to be examined, and wavelength when spectral wavelength is established with master sample feature is identical;Above-mentioned collected sample to be examined is handled, the characteristics of image of sample to be examined is extracted, the classification of the characteristics of image is identical as master sample;Object identifies, and through master sample feature compared with sample to be examined feature, judges whether sample to be examined is identical as master sample or has certain attribute.The present invention carries out Image Acquisition and feature extraction to the multi-layer information for being authenticated object using multiple spectrum, can effectively identify camouflage, the accuracy that lifting object identifies.
Description
Technical field
The present invention relates to object authentication technique field, especially a kind of object discrimination method based on multi-layer information.
Background technique
Object identifies field at present, for example gate inhibition's identification, identification technology continue to introduce new.For example, some is using increase
Dimension (such as two-dimension human face identification is promoted to three-dimensional face identification), increases identification object (as combined face, fingerprint, sound), with
And improve accuracy of identification (passing through various algorithms) etc..But various technologies have its deficiency, and increase dimension, raising accuracy of identification can
To promote recognition effect to a certain degree, but with the promotion of camouflage, still cannot meet the needs in sensitive occasion.And
It is cumbersome to increase identification content operation, is also easy to imitate, therefore recognition effect is not good enough yet.
Summary of the invention
The present invention provides a kind of object discrimination method, can improve identification effect by the extraction to object multi-layer information.
To achieve the above object, technical scheme is as follows:
A kind of object discrimination method based on multi-layer information, comprising:
Master sample feature is established, and the multispectral image of collected master sample is acquired, and extracts multispectral image
Feature, spectrum includes at least visible light and ultraviolet spectra or/infrared spectroscopy, to obtain master sample surface layer and internal layer
Feature;
The Image Acquisition of sample to be examined, collects the multispectral image of sample to be examined, and spectral wavelength and master sample are special
Wavelength when sign is established is identical;
Above-mentioned collected sample to be examined is handled, the characteristics of image of sample to be examined is extracted, the characteristics of image
Classification is identical as master sample;
Object identify, through master sample feature compared with sample to be examined feature, judge sample to be examined whether with standard
Sample is identical or has certain attribute.
In one embodiment, master sample feature is established adopts including master sample data collection, master sample image
The image characteristics extraction of collection, master sample, wherein when master sample data collection, each sample establishes more parts of spectrum picture numbers
According to respectively corresponding the information of different layers.
In one embodiment, when master sample feature is established, after the image characteristics extraction of master sample, further include
The classification of standard sample database and library is built, establishes the database of master sample.
In one embodiment, when the Image Acquisition of sample to be examined, the collection ring of sample to be examined and master sample is controlled
Border is similar, or/and is made up by video camera light filling unit.
In one embodiment, above-mentioned collected sample to be examined is handled, when extracting characteristics of image, is also wrapped
Image standardization is included, so that sample to be examined is consistent with the size of master sample, brightness.
In one embodiment, object identifies by the way of based on sample distance, comprising: each spectrum obtains one
A distance, is denoted as: R=[R1, R2..., Rn], wherein n is spectrum quantity;Each spectrum takes k samples in the top, and assigns
Give the different weight of this k ranking results, weighting value coefficient are as follows: { e1, e2..., ek};Weight coefficient { E is arranged to each spectrum1,
E2..., En, EnCoefficient magnitude is set according to different levels confidence level height;Final score is expressed as the weighting of all rankings
As a result:Wherein,Indicate occur in the k result in the top of every class spectrum picture
YiRanking, eiIndicate the weight of corresponding ranking, EmIndicate the weight of different spectrum picture features, finally, acquirement point is highest
Ranking assert that sample to be examined is identical as master sample as comparison result.
In one embodiment, object, which identifies, uses the identification method based on machine learning, including multispectral image to melt
Conjunction, model training, objective identification.
In one embodiment, Multispectral Image Fusion is used based on the preceding method to fusion, comprising: is set and is extracted
Characteristics of image indicates that wherein F=[f1, f2 ..., fn] indicates the spectrum picture feature extracted, by all spectrum with F
Feature line up a matrix, FF (i, j)=[F1, F2 ..., Fj]T, using space projection or dimension reduction method to FF matrix into
Row dimensionality reduction and characteristics extraction indicate the characteristics of image of all spectrum with the data after dimensionality reduction.
In one embodiment, the pattern representation of model training input is the dimensionality reduction that Multispectral Image Fusion step obtains
Data afterwards, output are sample class.
In one embodiment, in objective identification, final recognition result is the comparison knot of comprehensive different spectrum pictures
Fruit is weighted, and obtains point highest sample as final qualification result and assert to be checked if top score is more than preset threshold
Sample is identical as master sample.
It is adopted the beneficial effects of the present invention are: the present invention carries out image to the multi-layer information for being authenticated object using multiple spectrum
Collection and feature extraction can effectively identify camouflage, the accuracy that lifting object identifies.
Detailed description of the invention
Fig. 1 is the schematic diagram that the embodiment of the present invention is applied to recognition of face.
Specific embodiment
Method based on multispectral scanner, this programme can detect the information of different layers, such as facial image, it is seen that
Light can only detect skin surface, and ultraviolet light is right due to that can detect the information of skin inner layer with certain penetration power
In other objects, such as water fruits and vegetables etc., and it is the same.With reference to the accompanying drawing and example, by taking recognition of face as an example, to this hair
It is bright to be described further.
In the present embodiment, as shown in Figure 1, the object discrimination method based on multi-layer information includes the following steps.
Step S1, master sample feature are established.The step for be collected when being embodied according to the difference of the application, below
It is described in detail.
Step S11, the collection of sample data
According to the difference of task, using different collection modes.In the present embodiment, it is based on multispectral scanner image, we
Case is available infrared, and the image of the difference spectrum such as Uv and visible light, when collecting sample data, each sample is established more
Part spectral image data indicates the mark to every a material with Y, wherein Y=[y1, y2 ..., yn], and yn indicates each
The feature of spectrum picture since different spectrum represent the data of different skin layer, such as ultraviolet can see that skin is deeper on the face
The spot information etc. of layer;As shown in Figure 1, left-side images are the image of the skin surface taken with visible light, right side
Image for the skin inner layer taken with ultraviolet spectra.
Step S12, the acquisition of sample image
Carrying out shooting sampling to sample based on multispectral camera, (spectral wavelength used is 300 nanometers optional, 500 receive
Rice, 800 nanometers etc. it is a variety of), each such object can acquire multiple spectrum image, and this programme is set to M, wherein M=
[M1, M2, M3 ..., Mn], Mi are the image taken under some spectrum;
Step S13, the image characteristics extraction of java standard library
Spectrum picture is layered by this programme according to the difference of its spectral wavelength, and the wavelength of light is smaller, and penetration power is stronger,
Therefore the image that smaller wavelength can see is closer to deep layer.For each tomographic image (each spectrum picture), this programme
Carrying out image preprocessing and feature extraction respectively, these features may is that edge feature according to the difference of the application, Hog histogram,
Gabor wavelet, Haar small echo, SIFT feature etc. are also possible to convolution feature.
Step S14, the classification of java standard library and builds library
With RGB, the differences such as gray scale or depth image have N kind difference spectrum picture for each target this programme
It is described, by taking recognition of face as an example, firstly for a certain face, this programme has collected multiple spectrum pictures, secondly we
The purpose of case is from outer skin layer, and the various aspects information such as skin endosexine and deep skin is analyzed;Based on this, this programme
Classification, which is carried out, as the attribute of sample from each layer of information (corresponding each section of spectrum picture) builds library.
Step S2, the Image Acquisition of sample to be examined.It is to meet sample to be examined and master sample as far as possible same first
It is acquired in light source or proximate sources environment, this condition can be adjusted by external light filling unit;It can also be from the later period
Improved in software algorithm, such as brightness normalization etc..
Step S3 handles the image of sample to be examined.May include following steps:
Step S31, image standardization: sample-size is consistent with java standard library first, (wherein secondly by brightness normalization
Brightness is pixel R, the average value of tri- color of G, B) and setting contrast algorithm (average brightness * contrast rating) to image carry out
Normalization, wherein the normalized operation of brightness is in this programme: by the average brightness of each pixel subtracted image, contrast tune
Whole is by being adjusted multiplied by contrast rating;
Step S32, the extraction of characteristics of image to be checked, treatment process of this process with master sample.
Step S4, object identify.This example provides two kinds of identification methods: it is based on sample distance and is based on machine learning, under
Face is described respectively.
The mode based on sample distance is introduced first.Discrimination process is actually to find in master sample and to sample
The most similar sample of this distance.
Distance mathematically has a many representation methods, such as Euclidean distance, COS distance etc., wherein Euclidean distance are as follows:Each available distance of spectrum (each layer data), is denoted as: R=[R1, R2..., Rn], wherein n
For spectrum quantity.Meanwhile each spectrum takes K (closely located) samples in the top, and such as: spectrum picture Mi takes { d1,
d2..., dkTotal k near preceding as a result, weight bigger importance in the top in order to highlight, this programme assign this k sequence
As a result different weights, weighting value coefficient are as follows: { e1, e2..., ek};In addition, for recognition of face, the really degree of the deep information
It is higher, and surface layer information is easier to be interfered or pretend, therefore the weight coefficient of Mi different spectrum is set to: { E1,
E2..., En, it is higher with the confidence level for highlighting the deep information.Final score can use the weighted results for being expressed as all rankings:
Wherein,Indicate y occur in the k result in the top of every class spectrum pictureiRanking, eiIt indicates to correspond to
The weight of ranking, EmIndicate the weight of different spectrum (different levels) characteristics of image, finally, this programme take highest scoring (away from
It is close from most) ranking as comparison result.
Identification method based on machine learning, this programme can use " manual Feature extraction~+ conventional machines learning method "
Method further identified, can also be distinguished that these methods are all relative maturities based on deep learning method.
However, since this programme there are many different spectrum pictures describes target, the processing of identification process can be different.
Final recognition result should be the synthesis result after multiple spectrum image co-registration.This programme specifically comprises the following steps.
Step S41, the fusion of multispectral image
This programme is used based on the preceding method to fusion, when specific implementation, this programme set the characteristics of image that extracts with F come
It indicates, wherein F=[f1, f2 ..., fn] indicates a spectrum picture feature extracting, and this programme is by the feature of all spectrum
Line up a matrix, FF (i, j)=[F1, F2 ..., Fj]T, this programme is projected using space or dimension reduction method (PCA, LDA)
Deng carrying out dimensionality reduction and characteristics extraction to FF matrix, the characteristics of image of all spectrum, final base are indicated with the data after dimensionality reduction
Be trained traditional machine learning model in this feature;Here, the characteristics of image of manual extraction may is that binaryzation
The edge feature of image, HOG histogram feature, Haar wavelet character, SIFT feature etc.;Machine learning algorithm may is that
SVM, neural network, decision tree etc.;
Step S42, model training: this programme is the same target to be described based on multi-layer information, therefore the sample inputted is retouched
Stating is multidimensional data, and output is sample class, and certain multidimensional data can be the feature vector after Data Dimensionality Reduction;
Step S43, objective identification: by taking recognition of face as an example, final recognition result is that comprehensive face different layers (are not shared the same light
Spectrogram picture) comparison result be weighted, obtain point highest sample as final qualification result, if top score is more than default
Threshold value (such as similarity 99) then assert that face to be checked is consistent with target face.
The present invention is based on the object discrimination methods of multilayer feature can be applied to other a variety of occasions.Such as it reflects for historical relic
Fixed (such as ceramics) is identified by the feature of cultural artifact surface and internal layer (ceramic layer below glaze layer), can be more accurately
The true and false made an appraisal of a cultural relic.Such as in fruit inspection, clearly imaging surface property is captured with visible spectrum, to check whether into
It is ripe;The sign of decaying of early stage, such as mechanical abrasion and internal insect pest can be checked near infrared light.If fruit is correct face
Color, most of near infrared light can reflect.When the absorption light in a region is relatively more, it is likely to the eucaryotic cell structure in this region
Different from other regions, indicating internal has rotten or damage.So as to identify the maturity of fruit by this attribute and decline
Lose situation.
As can be seen from the above description, the present invention carries out Image Acquisition to the multi-layer information for being authenticated object using multiple spectrum
With feature extraction, camouflage, the accuracy that lifting object identifies can be effectively identified.
Claims (10)
1. a kind of object discrimination method based on multi-layer information characterized by comprising
Master sample feature is established, and the multispectral image of collected master sample is acquired, and extracts the spy of multispectral image
Sign, spectrum includes at least visible light and ultraviolet spectra or/infrared spectroscopy, to obtain the spy on master sample surface layer and internal layer
Sign;
The Image Acquisition of sample to be examined, collects the multispectral image of sample to be examined, and spectral wavelength is built with master sample feature
Wavelength immediately is identical;
Above-mentioned collected sample to be examined is handled, the characteristics of image of sample to be examined, the classification of the characteristics of image are extracted
It is identical as master sample;
Object identify, through master sample feature compared with sample to be examined feature, judge sample to be examined whether with master sample
It is identical or have certain attribute.
2. object discrimination method according to claim 1, which is characterized in that it includes master sample that master sample feature, which is established,
The image characteristics extraction of data collection, master sample Image Acquisition, master sample, it is each wherein when master sample data collection
A sample establishes more parts of spectral image datas, respectively corresponds the information of different layers.
3. object discrimination method according to claim 1, which is characterized in that when master sample feature is established, in standard sample
After this image characteristics extraction, further include the classification of standard sample database and build library, establishes the database of master sample.
4. object discrimination method according to claim 1, which is characterized in that when the Image Acquisition of sample to be examined, control to
Sample sheet is similar with the collection environment of master sample, or/and is made up by video camera light filling unit.
5. object discrimination method according to claim 1, which is characterized in that at above-mentioned collected sample to be examined
Reason, further includes image standardization when extracting characteristics of image, so that sample to be examined is consistent with the size of master sample, brightness.
6. object discrimination method according to claim 1, which is characterized in that object, which identifies, uses the side based on sample distance
Formula, comprising: each spectrum obtains a distance, is denoted as: R=[R1, R2..., Rn], wherein n is spectrum quantity;Each spectrum
K samples in the top are taken, and assign this k ranking results different weights, weighting value coefficient are as follows: { e1, e2..., ek};
Weight coefficient { E is arranged to each spectrum1, E2..., En, EnCoefficient magnitude is set according to different levels confidence level height;Finally
Score is expressed as the weighted results of all rankings:Wherein,Indicate every class spectrum picture
There is y in k result in the topiRanking, eiIndicate the weight of corresponding ranking, EmIndicate different spectrum picture features
Weight finally obtain point highest ranking and be used as comparison result, identification sample to be examined is identical as master sample.
7. object discrimination method according to claim 1, which is characterized in that object, which identifies, uses the mirror based on machine learning
Other mode, including Multispectral Image Fusion, model training, objective identification.
8. object discrimination method according to claim 7, which is characterized in that Multispectral Image Fusion is used based on preceding to melting
The method of conjunction, comprising: set the characteristics of image extracted is indicated with F, and wherein F=[f1, f2 ..., fn] indicates extract one
The feature of all spectrum is lined up a matrix, FF (i, j)=[F1, F2 ..., Fj] by a spectrum picture featureT, utilize space
Projection or dimension reduction method carry out dimensionality reduction and characteristics extraction to FF matrix, and the figure of all spectrum is indicated with the data after dimensionality reduction
As feature.
9. object discrimination method according to claim 8, which is characterized in that the pattern representation of model training input is mostly light
Data after the dimensionality reduction that spectrogram is obtained as fusion steps, output is sample class.
10. object discrimination method according to claim 9, which is characterized in that in objective identification, final recognition result is
The comparison result of comprehensive different spectrum pictures is weighted, and obtains point highest sample as final qualification result, if highest obtains
Dividing is more than preset threshold, then assert that sample to be examined is identical as master sample.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810918127.XA CN109145799A (en) | 2018-08-13 | 2018-08-13 | A kind of object discrimination method based on multi-layer information |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810918127.XA CN109145799A (en) | 2018-08-13 | 2018-08-13 | A kind of object discrimination method based on multi-layer information |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109145799A true CN109145799A (en) | 2019-01-04 |
Family
ID=64792859
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810918127.XA Pending CN109145799A (en) | 2018-08-13 | 2018-08-13 | A kind of object discrimination method based on multi-layer information |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109145799A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111160145A (en) * | 2019-12-13 | 2020-05-15 | 广州懿锝文化创意有限公司 | Method for identifying uniqueness of porcelain |
CN111739003A (en) * | 2020-06-18 | 2020-10-02 | 上海电器科学研究所(集团)有限公司 | Machine vision algorithm for appearance detection |
CN112907571A (en) * | 2021-03-24 | 2021-06-04 | 南京鼓楼医院 | Target judgment method based on multispectral image fusion recognition |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5621858A (en) * | 1992-05-26 | 1997-04-15 | Ricoh Corporation | Neural network acoustic and visual speech recognition system training method and apparatus |
CN101211410A (en) * | 2007-12-25 | 2008-07-02 | 哈尔滨工业大学 | Multi-light spectrum palm print identity authentication method and its special-purpose collection instrument |
CN101894263A (en) * | 2010-05-24 | 2010-11-24 | 中国科学院合肥物质科学研究院 | Computer-aided classification system and classification method for discriminating mapped plant species based on level set and local sensitivity |
CN103884661A (en) * | 2014-02-21 | 2014-06-25 | 浙江大学 | Soil total nitrogen real-time detection method based on soil visible-near infrared spectrum library |
-
2018
- 2018-08-13 CN CN201810918127.XA patent/CN109145799A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5621858A (en) * | 1992-05-26 | 1997-04-15 | Ricoh Corporation | Neural network acoustic and visual speech recognition system training method and apparatus |
CN101211410A (en) * | 2007-12-25 | 2008-07-02 | 哈尔滨工业大学 | Multi-light spectrum palm print identity authentication method and its special-purpose collection instrument |
CN101894263A (en) * | 2010-05-24 | 2010-11-24 | 中国科学院合肥物质科学研究院 | Computer-aided classification system and classification method for discriminating mapped plant species based on level set and local sensitivity |
CN103884661A (en) * | 2014-02-21 | 2014-06-25 | 浙江大学 | Soil total nitrogen real-time detection method based on soil visible-near infrared spectrum library |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111160145A (en) * | 2019-12-13 | 2020-05-15 | 广州懿锝文化创意有限公司 | Method for identifying uniqueness of porcelain |
CN111160145B (en) * | 2019-12-13 | 2023-09-19 | 广州懿锝文化创意有限公司 | Method for identifying porcelain uniqueness |
CN111739003A (en) * | 2020-06-18 | 2020-10-02 | 上海电器科学研究所(集团)有限公司 | Machine vision algorithm for appearance detection |
CN111739003B (en) * | 2020-06-18 | 2022-11-18 | 上海电器科学研究所(集团)有限公司 | Machine vision method for appearance detection |
CN112907571A (en) * | 2021-03-24 | 2021-06-04 | 南京鼓楼医院 | Target judgment method based on multispectral image fusion recognition |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110516576B (en) | Near-infrared living body face recognition method based on deep neural network | |
US9317761B2 (en) | Method and an apparatus for determining vein patterns from a colour image | |
Lau et al. | Automatically early detection of skin cancer: Study based on nueral netwok classification | |
US20060222212A1 (en) | One-dimensional iris signature generation system and method | |
CN109145799A (en) | A kind of object discrimination method based on multi-layer information | |
Tolosana et al. | Towards fingerprint presentation attack detection based on convolutional neural networks and short wave infrared imaging | |
JP4202271B2 (en) | Pattern comparison | |
Narang et al. | Face recognition in the SWIR band when using single sensor multi-wavelength imaging systems | |
CN107918773B (en) | Face living body detection method and device and electronic equipment | |
US11854289B2 (en) | Biometric identification using composite hand images | |
CN104809450B (en) | Wrist vena identification system based on online extreme learning machine | |
CN116849612B (en) | Multispectral tongue picture image acquisition and analysis system | |
CN106845449A (en) | A kind of image processing apparatus, method and face identification system | |
Bhagwat et al. | A framework for crop disease detection using feature fusion method | |
Ding et al. | Feature extraction of hyperspectral images for detecting immature green citrus fruit | |
Borkar et al. | IRIS Recognition System | |
Janes et al. | A low cost system for dorsal hand vein patterns recognition using curvelets | |
Richard et al. | Colour local pattern: a texture feature for colour images | |
Nithya et al. | Nail based disease analysis at earlier stage using median filter in image processing | |
Gong et al. | An optimized wavelength band selection for heavily pigmented iris recognition | |
Chugh | An accurate, efficient, and robust fingerprint presentation attack detector | |
Cornett et al. | Effects of postmortem decomposition on face recognition | |
VigneshJanarthanan | Hybrid Multi-Core Algorithm Based Image Segmentation for Plant Disease Identification using Mobile Application | |
Chen | A Highly Efficient Biometrics Approach for Unconstrained Iris Segmentation and Recognition | |
Kumar et al. | A Neoteric Procedure for Spotting and Segregation of Ailments in Mediciative Plants using Image Processing Techniques. |
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 | ||
WD01 | Invention patent application deemed withdrawn after publication | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190104 |