CN108038464A - A kind of new HOG features Uygur nationality facial image recognizer - Google Patents
A kind of new HOG features Uygur nationality facial image recognizer Download PDFInfo
- Publication number
- CN108038464A CN108038464A CN201711408669.4A CN201711408669A CN108038464A CN 108038464 A CN108038464 A CN 108038464A CN 201711408669 A CN201711408669 A CN 201711408669A CN 108038464 A CN108038464 A CN 108038464A
- Authority
- CN
- China
- Prior art keywords
- facial image
- face
- image
- distribution diagram
- gradient
- 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
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Human Computer Interaction (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a kind of new HOG features Uygur nationality facial image recognizer, the intrinsic textural characteristics of facial image are protruded by Laplce's filtering transformation first, weaken the interference of inhomogeneous illumination, then low-frequency information caused by long-lost cosine code is retained while the facial image after Laplce's filtering transformation being effectively filtered out high fdrequency component by discrete cosine transform, the image for the recovery being close with protoplast's face image is constructed after inverse discrete cosine transform IDCT processing, filter out in facial image and insensitive medium-high frequency part and reduce its dimension well, excellent basis is provided for subsequent extracted facial image feature；Finally using gradient quality distribution diagram HOG operator extractions its facial images feature and Classification and Identification is carried out to facial image using arest neighbors method.The algorithm its discrimination in the self-built Uygur nationality's face database of Yale B face databases and seminar is greatly improved, and better than other traditional algorithms, and has very strong robustness.
Description
Technical field
The invention belongs to biometrics identification technology field, is related to a kind of Arithmetic of Face Image Recognition, specifically, is related to
A kind of new HOG features Uygur nationality facial image recognizer.
Background technology
Recognition of face is one of most important biometrics identification technology, can extensively social safety, entrance guard safety system,
Applied in the various fields such as teaching.Develop with the maturation of face recognition technology, many Related products are applied to well
In human lives.Meanwhile the research and application of face recognition technology have many problems, predominantly its recognition performance is by many
The influence of aspect factor, as facial image resolution ratio obscures, illumination diversity, multi-pose and abundant the problems such as blocking.
In recent years, many algorithms are proposed for interference of the inhomogeneous illumination to recognition of face effect, traditional face is special
Sign extraction recognizer mainly has the extraction of Gabor face characteristics, and the extraction of local binary patterns LBP face characteristics, local three are worth mould
Formula LTP face characteristics extract, gradient direction quality distribution diagram HOG face characteristics extraction etc..
Gabor face characteristics extraction algorithm has good ability to express, while possesses similar to biological vision system
Characteristic, therefore carried out extensive research application, but since the Dimension Characteristics in its multiple dimensioned direction are excessive, cause its calculating
Process is more complicated, therefore real-time difficult to realize and validity demand are in actual application.LBP and LTP face characteristics carry
Take with simple classification, and the features such as computational complexity excessively simplifies, but to noise interferences such as inhomogeneous illuminations very
Sensitivity, is easily disturbed the influence of environmental factor.Gradient direction quality distribution diagram HOG features are to inhomogeneous illumination, partial occlusion
Deng good and beneficial to complicated calculations with robustness, the pass of researcher is more and more obtained based on the factor of these superiority
Note.
Gradient direction quality distribution diagram (HOG) proposed .HOG features due to direction, light first in 2005 by Dalal
According to etc. factor there is insensitivity, in computer vision research, such as pedestrian identify field be widely used.Arrive
2011 by Deniz et al. first Applications in recognition of face, but achievement in research of the HOG features in recognition of face is opposite
Less, potentiality of the HOG features in face recognition application are also to be excavated.
The content of the invention
The present invention proposes a kind of new HOG features Uygur nationality facial image recognizer, and this method is general based on drawing
Lars filters and gradient quality distribution diagram (HOG) Uygur nationality face recognizer of discrete cosine transform (DCT) fusion.Should
Algorithm protrudes the intrinsic textural characteristics of facial image by Laplce's filtering transformation first, weakens the interference of inhomogeneous illumination,
Then the facial image after Laplce's filtering transformation can so be effectively filtered out into high frequency by discrete cosine transform (DCT)
Retain low-frequency information caused by long-lost cosine code while component, then pass through inverse discrete cosine transform (IDCT) place
The image for the recovery being close with protoplast's face image is constructed after reason, filters out medium-high frequency part in facial image and insensitive
And its dimension is reduced well, provide excellent basis for subsequent extracted facial image feature；Finally utilize gradient matter
The feature of its facial image of amount distribution map (HOG) operator extraction simultaneously carries out Classification and Identification using arest neighbors method to facial image.
Test result indicates that the algorithm in the self-built Uygur nationality's face database of Yale B face databases and seminar its
Discrimination is greatly improved, and better than other traditional algorithms, and has very strong robustness.
Its technical solution is as follows：
A kind of new HOG features Uygur nationality facial image recognizer, comprises the following steps：
(1) Laplce's filtering transformation pretreatment is carried out to training sample facial image；
(2) it is divided into 8 × 8 facial image blocks to completing the pretreated facial image of Laplce's filtering transformation, then will
Image block carries out dct transform, extracts 10, the upper left corner low frequency coefficient characteristic component of its facial image；
(3) 10, the facial image upper left corner low frequency coefficient characteristic component of extraction is passed through into inverse discrete cosine transform IDCT
Obtain its facial image reconstructed similar in original image；
(4) formula is passed throughWith α (x, y)=arctan (Gy/Gx) obtain reconstruct facial image
Gradient；
(5) formula is used1≤k≤K using unit Cell as unit pattern,
Statistical gradient direction quality distribution diagram；
(6) in units of block block, the gradient direction quality distribution diagram in block is normalized；
(7) the gradient quality distribution diagram in each piece is stitched together, obtains the HOG features of final facial image；
(8) Classification and Identification finally is carried out to facial image with arest neighbors method.
Beneficial effects of the present invention are：
The present invention proposes gradient quality distribution diagram (HOG) people of Laplce's filtering and discrete cosine transform (DCT) fusion
Face recognizer.The algorithm filters the edge and textural characteristics of its prominent facial image by Laplce, weakens non-homogeneous light
According to interference, facial image high-frequency information data are eliminated using discrete cosine transform (DCT), retain its low-frequency information data, together
When reduce facial image dimension, and facial image is rebuild with inverse discrete cosine transform (IDCT), although the people of reconstruct
Face image has certain difference with original facial image, but substantially remains the information data of original facial image, and then
Go out face inherent feature using gradient quality distribution diagram (HOG) operator extraction, classification is identified.Calculation proposed by the invention
Method can extract facial image feature well under illumination interference to a certain extent, and achieve superior Classification and Identification effect
Fruit, the experiment simulation in the self-built Uygur nationality's face database of Yale B face databases and seminar show, the calculation
Discrimination is up to 95% and the self-built Uygur nationality of seminar in Yale B face databases of the method under different characteristic dimension
Its discrimination is up to 98.5% in face database, its discrimination has very big compared with other Arithmetic of Face Image Recognition
Improve, identification discrimination has good robustness.
Brief description of the drawings
Fig. 1 parts Uygur nationality face database；
Fig. 2 facial image identification process；
Fig. 3 Yale B face databases；
Uygur nationality's face database under Fig. 4 inhomogeneous illuminations；
Fig. 5 Uygur nationality face database discrimination；
Fig. 6 Yale B face database discriminations.
Embodiment
Technical scheme is described in more detail with reference to the accompanying drawings and detailed description.
1 Laplce filters
Facial image often shows different textures, angle point and other identification informations on different scale, wherein drawing
This filtering of pula is exactly one of which multiscale analysis instrument, and it is filtered that its energy structure will show more textural characteristics
After ripple.Its expression is：
In order to face image processing, be denoted as two-dimensional discrete form：
G (x, y)=f (x, y)+c [▽2f(x,y)] (3)
Above formula (3) is inner, and g (x, y) is target image, and f (x, y) is original image.When mask center coefficient is timing, c is
1；When mask center coefficient is bears, c is -1.Facial image can protrude facial image after Laplce's filtering
The intrinsic textural characteristics in edge.
2 discrete cosine transforms (DCT)
Discrete cosine transform (DCT) is widely used in the compression processing of view data, and data message is converted to frequency domain
Information, and important people's image data information is gathered on low frequency coefficient.In discrete cosine transform (DCT) algorithm, one
Width facial image matrix, can be first by two-dimensional feature vector when handling single bivector matrix as bivector matrix
After extraction, then the two-dimensional feature vector extracted before it is put into its preprepared feature database to carry out behind
Identification；When needing new facial image being added in storehouse, first to its pretreatment work, then to the facial image
Dct transform is carried out, does not have to thus scruple other facial images, even if the face image data in storehouse is very big, then calculation amount
Will not be very big.
For the facial image f (x, y) of a width m × n, his DCT expression formulas can be described as：
In above formula, C (u, v) be known as in DCT fixed coefficient, dct transform is orthogonally transformed, and facial image passes through DCT
After conversion, wherein the big coefficient of numerical value is mainly gathered in the non-high frequency section in the upper left corner, that is to say, that passes through the master after dct transform
Information data amount is wanted to be gathered on a small number of non-high frequent coefficient values.
Two-dimension discrete cosine transform (DCT) inverible transform, its inverse transformation, which is referred to as inverse discrete cosine transform (IDCT), is,
It is defined as：
Using IDCT, but obtain the facial image of reconstruction, although although facial image and the original facial image rebuild not
It is completely similar, but its main information data is all kept down, so, the interference such as illumination of its facial image
Factor reduces the dimension of facial image all by certain weakening, and subsequently preferably extraction facial image feature is played very
Big effect.
3 HOG feature extractions describe
HOG features describe operator and come from SIFT algorithms, are description of useful image local difference information, more to expression
The linear transformations such as posture changing, strong and weak, the complex scene change of illumination have superior robustness.Gradient quality distribution diagram (HOG)
Feature extraction is achieved in the range of pattern-recognition and successfully applied, such as facial image identification, human body pedestrian detection, target identification
Tracking etc..In computer vision and image procossing, HOG Feature Descriptors be by Local gradient direction quality distribution diagram come
Calculate and statistics produces its Image salient features, specific calculating process is as follows：
(1) gradient of each pixel is calculated, is shown below：
Gx(x, y)=I (x+1, y)-I (x-1, y) (8)
Gy(x, y)=I (x, y+1)-I (x, y-1) (9)
Then, the amplitude G (x, y) of gradient and the direction α (x, y) of gradient are estimated
α (x, y)=arctan (Gy/Gx)(11)
(2) facial image is isolated into the similar unit mode Cell of several sizes, if unit mode size is s × s
Two-dimentional pixel, and the quality distribution diagram Cell-HOG of its gradient direction is estimated using the unit mode statistical unit main as its.
Gradient direction is divided into K uniform intervals (Bin), uses Vk(x, y) represents contribution of the pixel (x, y) to k-th of Bin
Weighted value, the value algorithm of weighted value：
For each unit mode Celli, calculate its gradient direction quality distribution diagram vectorIts
In
(3) t × t adjacent Cell is made into a block Block.The block is moved to one to the right or downwards in the picture
The size of a Cell, to obtain next piece of therefore, there is overlapping between block and block.For each piece, institute in splicing block
There is the gradient quality distribution diagram vector of unit, obtain the gradient direction quality distribution diagram vector B lock-HOG of block, length definition
For Kt2, then, by each piece of gradient quality distribution diagram vector normalized.The method used is by L2Mould normalizes, i.e.,
For vector v, if
In above formula, V is the HOG two-dimensional feature vectors of a block before standardization, | | v | |kRepresent k-norm computings, ε
It is minimum, for preventing the phenomenon that denominator is zero in above formula, and then causes it that infinite value is calculated.
(4) finally, the gradient direction quality distribution diagram fusion of all Block is got up, obtains it and finally enter image
HOG features
The face feature general introduction of 4 Uygur nationality and the structure of database
The Uygur nationality inhabits the Xinjiang Uygur Autonomous Regions of northwest China, its macroscopic features is obvious, with
The Hans have different appearance.Although the Uygur nationality is also yellow, due to historical development and the shape of race
Into intermarriage etc. makes them save substantial amounts of white blood lineage.Uygur nationality's man's chaeta is more flourishing, and love is worn the bread, two
Closer to the distance, sunk socket, stature is biger and tall, and shape of face is more more in rectangular face (state's font), high-bridged nose, most of almond eye,
Double-edged eyelid are more, and color development, wink are in yellowish-brown.
Inhabit the appearance bodily form feature of the Uighurs of Southern Xinjiang (South Sinkiang) with inhabiting In The North of Xinjiang
The Uighurs in area (North SinKiang) still has certain difference.North SinKiang the Uighurs is close to former Soviet Union area, no matter from ground
Reason position is still intermarriaged etc. from the point of view of reason, or their colour of skin is white, color development it is most for brown, wink it is shallow, these essential characteristics
All nearly close to white feature.And South Sinkiang the Uighurs is from reasons such as geographical location, humane weather, intermarriages, they
Color development, wink it is more black, and with wear the bread, veil, the custom such as be branded as.
Based on the feature of above Uygur nationality face, we must build the high scale of complete set and high quality dimension I
That race face database.In order to protrude the main feature of Uygur nationality's face, we mainly acquire southern Xinjiang and Yi Li
Uygur nationality's face in area, these regional Uygur nationality's faces remain most national characteristics, have good allusion quotation
Type.We acquire Uygur nationality's face of different sexes, age etc., its prominent face as far as possible in gatherer process
Main feature, haves laid a good foundation for identification machine sort.Adopted according to the requirement and subsidy of problem, used in us
Integrate instrument as high-definition Canon EOS70D slr cameras, gather Uygur nationality's face information data in different environments,
Uygur nationality's face database is constructed, is included under inhomogeneous illumination and partial occlusion, part Uygur nationality face database
As shown in Figure 1：
5 inventive algorithms realize description：
Since discrimination of the face under inhomogeneous illumination is difficult and the problem of poor robustness, the present invention, which proposes, to be based on
Laplce filters and gradient quality distribution diagram (HOG) face recognition algorithms of discrete cosine transform (DCT) fusion.The algorithm
Filtered first by Laplce and the influence of inhomogeneous illumination is weakened, the intrinsic texture edge feature of prominent facial image, so
The non-high frequency inherent feature component of facial image is extracted using discrete cosine transform (DCT) afterwards, while reduces facial image
Intrinsic dimensionality, and the image information being close with protoplast's face image is restored using inverse discrete cosine transform (IDCT) structure, this
Sample can retain the important characteristic information of its face, finally extract its facial image using gradient quality distribution diagram (HOG) operator
Feature.Shown in the flow chart 2 of this algorithm：Specific algorithm step is as follows：
(1) Laplce's filtering transformation pretreatment is carried out to training sample facial image；
(2) it is divided into 8 × 8 facial image blocks to completing the pretreated facial image of Laplce's filtering transformation, then will
Image block carries out dct transform, extracts 10, the upper left corner low frequency coefficient characteristic component of its facial image；
(3) 10, the facial image upper left corner low frequency coefficient characteristic component of extraction is passed through into inverse discrete cosine transform (IDCT)
Obtain its facial image reconstructed similar in original image；
(4) gradient of reconstruct facial image is obtained by formula (10) and (11)；
(5) with formula (12) using unit Cell as unit pattern, statistical gradient direction quality distribution diagram；
(6) in units of block block, the gradient direction quality distribution diagram in block is normalized；
(7) the gradient quality distribution diagram in each piece is stitched together, obtains the HOG features of final facial image；
(8) Classification and Identification finally is carried out to facial image with arest neighbors method.
6 analysis of simulation experiment：
The present invention selects Yale B face databases to carry out inventive algorithm replication experiment.Yale B face databases are
The standard faces storehouse of one variable illumination, the face database are the faces for having multi-direction illumination variation and multi-pose change
Database, covers 10 people altogether, constructs being total under 576 kinds of multi-poses and more illumination conditions (9 kinds of posture × 64 kind illumination)
5760 width facial images.The facial image of all selections has all done similar pretreatment work, covers cutting, correction, scaling
Deng as shown in Figure 3.
Meanwhile the present invention also uses the self-built uygur nationality of Xijiang's face information database of problem group membership and is tested
Confirmation is tested.Uygur nationality's facial image is all from each area in Xinjiang, and different age group, different sexes, have typical represent
Property.As shown in Figure 4.
6.1 based on the face identification rate under algorithms of different
The present invention is tieed up by the different algorithms Xinjiang self-built to Yale B face databases and seminar of the present invention first
I carries out Experimental comparison by your race's face database.Two kinds of different face databases are normalized at work respectively first
Facial image, i.e., is normalized to the facial image of 64 × 64,32 × 32,16 × 16 sizes by reason, is then calculated with tradition DCT
Method, tradition LBP algorithms, tradition HOG feature extraction algorithms, DCT+LBP algorithms, HOG+LBP blending algorithms and inventive algorithm
Experiment comparison is carried out, the classification of face is finally realized using arest neighbors method.The results are shown in Table 1：
Face identification rate of the table 1 based on algorithms of different
By 1 result of table it may be seen that (exemplified by 64 × 64), by the self-built Uygur nationality's face of seminar of the present invention
In discrimination, the gradient quality distribution diagram proposed by the present invention based on Laplce's filtering and discrete cosine transform (DCT) fusion
(HOG) Uygur nationality's face recognizer highest, has reached 98.99%, and secondly the discrimination of HOG+LBP blending algorithms is also very
Height, is 95.56%.The discrimination of other algorithms is also higher.And in the discrimination of Yale B face databases, people
Face discrimination power has reached 93.23%, illustrates that algorithm performance proposed by the invention is still higher, is secondly merged for HOG+LBP
Algorithm is 85.53%.
We contrast the self-built Uygur nationality's face database and Yale B face databases of seminar from table 1, we
As can be seen that the common ground of the two face databases is under inhomogeneous illumination, and carried in the present invention
Human face identification rate highest under the algorithm and HOG+LBP blending algorithms that go out, this is primarily due to direction gradient quality distribution diagram
(HOG) algorithm has critically important geometric invariance and luminosity Inalterability of displacement characteristic, algorithm especially proposed by the invention
Laplce's filtering and discrete cosine transform (DCT) work disposal, therefore the human face identification rate of inventive algorithm have been carried out in advance
Performance, which is better than other five kinds of algorithms, illumination etc., has stronger robustness.
6.2 based on the face identification rate under different characteristic dimension
In order to more embody the superiority of institute's algorithm of the present invention, we are tieed up by taking 64 × 64 facial images as an example
The experiment of I your face identification rate of the two kinds of databases of clansman's face and Yale B faces under different intrinsic dimensionalities, such as Fig. 5 and
Shown in Fig. 6：
Uygur nationality's face discrimination under 2 different characteristic dimension of table
Yale B face identification rates under 3 different characteristic dimension of table
From conclusions as can be seen that algorithm proposed by the invention is in Yale B face databases and self-built Uygur
Discrimination highest in race's face database, especially in Uygur nationality's face identification, this is primarily due to the Uighurs
What face database embodied in gatherer process is positive effect light image, andFace database is that have different postures
Under light image, secondly be exactly to protrude the edge of facial image and texture using Laplce's filtering in inventive algorithm
Feature, reduces illumination effect, and then merging gradient quality distribution diagram (HOG) by discrete cosine transform (DCT) can be fine
Extraction its face it is special, so Uygur nationality's face database resolution is higher than Yale B face databases；Certain HOG+
The identification process of LBP blending algorithms is also higher, this is primarily due to both algorithms and embodies direction gradient quality distribution diagram
(HOG) algorithm has the excellent characteristic of geometric invariance and luminosity Inalterability of displacement.
Algorithm proposed by the invention can not only strengthen its facial image identification capability in summary, and to more changes
Different illumination etc. there is good robustness.It can be seen that it is superior to carry out the recognition effect that this algorithm extraction face feature obtains
In other face recognition algorithms, it can extract more abundant more effective than traditional LBP, tradition HOG and tradition DCT scheduling algorithms
Face texture information, has very strong antijamming capability to inhomogeneous illumination, embodies inventive algorithm in extraction face figure
As characteristic aspect is even better.
6.3 time tests are analyzed
Hardware environment used in time test experiment is Intel Corei7,4G memories；Simulated environment is matlab 2014b.
In Yale B face databases and self-built Uygur nationality's face storehouse, everyone 5 width images are chosen as training sample, difference
Using other facial images as test sample.
The time test of 4 algorithms of different of table
By table 4 as it can be seen that inventive algorithm all embodies most in training average time or on identification average time
Good state, overstriking data are optimum state in table.Although inventive algorithm is number two on training average time, mainly
It is because training process is offline, but to the run time of the computer system during whole identification face, identification system
System does not have a great impact, so inventive algorithm is still to be superior to other algorithms, has good real-time.
Finally, in order to verify the superiority of inventive algorithm, we calculate subject's work of individual features description
AUC (Area Under Curve) value of feature (ROC) curve.The part area that AUC value refers to below ROC curve is big
Small, area under the ROC curve by calculating Yale B face databases, assesses the identification capability of its algorithm, can also filter out
Optimal algorithm and its grader, the more big performance that just represent its grader of its value is better, otherwise area under curve is smaller
The performance of the more low then corresponding algorithm of AUC value and grader is poorer, we are by the AUC value of several features come to several spies
Sign carries out performance test, and it is as shown in the table, by AUC value in table, the ROC curve of feature recognition algorithms proposed by the present invention
It is more superior, recognition of face effect is better.
The AUC value of the ROC curve of 5 algorithms of different of table
7 conclusions：
The present invention proposes gradient quality distribution diagram (HOG) people of Laplce's filtering and discrete cosine transform (DCT) fusion
Face recognizer.The algorithm filters the edge and textural characteristics of its prominent facial image by Laplce, weakens non-homogeneous light
According to interference, facial image high-frequency information data are eliminated using discrete cosine transform (DCT), retain its low-frequency information data, together
When reduce facial image dimension, and facial image is rebuild with inverse discrete cosine transform (IDCT), although the people of reconstruct
Face image has certain difference with original facial image, but substantially remains the information data of original facial image, and then
Go out face inherent feature using gradient quality distribution diagram (HOG) operator extraction, classification is identified.Calculation proposed by the invention
Method can extract facial image feature well under illumination interference to a certain extent, and achieve superior Classification and Identification effect
Fruit, the experiment simulation in the self-built Uygur nationality's face database of Yale B face databases and seminar show, the calculation
Discrimination is up to 95% and the self-built Uygur nationality of seminar in Yale B face databases of the method under different characteristic dimension
Its discrimination is up to 98.5% in face database, its discrimination has very big compared with other Arithmetic of Face Image Recognition
Improve, identification discrimination has good robustness.
The foregoing is only a preferred embodiment of the present invention, protection scope of the present invention not limited to this, any ripe
Those skilled in the art are known in the technical scope of present disclosure, the technical solution that can be become apparent to
Simple change or equivalence replacement are each fallen within protection scope of the present invention.
Claims (1)
1. a kind of new HOG features Uygur nationality facial image recognizer, it is characterised in that comprise the following steps：
(1) Laplce's filtering transformation pretreatment is carried out to training sample facial image；
(2) it is divided into 8 × 8 facial image blocks to completing the pretreated facial image of Laplce's filtering transformation, then by image
Block carries out dct transform, extracts 10, the upper left corner low frequency coefficient characteristic component of its facial image；
(3) 10, the facial image upper left corner low frequency coefficient characteristic component of extraction is obtained it by inverse discrete cosine transform IDCT
With original image similar in the facial image that reconstructs；
(4) formula is passed throughWith α (x, y)=arctan (Gy/Gx) obtain the gradient for reconstructing facial image；
(5) formula is used1≤k≤K is using unit Cell as unit pattern, statistics
Gradient direction quality distribution diagram；
(6) in units of block block, the gradient direction quality distribution diagram in block is normalized；
(7) the gradient quality distribution diagram in each piece is stitched together, obtains the HOG features of final facial image；
(8) Classification and Identification finally is carried out to facial image with arest neighbors method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711408669.4A CN108038464A (en) | 2017-12-22 | 2017-12-22 | A kind of new HOG features Uygur nationality facial image recognizer |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711408669.4A CN108038464A (en) | 2017-12-22 | 2017-12-22 | A kind of new HOG features Uygur nationality facial image recognizer |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108038464A true CN108038464A (en) | 2018-05-15 |
Family
ID=62100845
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711408669.4A Pending CN108038464A (en) | 2017-12-22 | 2017-12-22 | A kind of new HOG features Uygur nationality facial image recognizer |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108038464A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108961218A (en) * | 2018-06-11 | 2018-12-07 | 无锡维胜威信息科技有限公司 | Solar power silicon platelet spends extracting method |
| CN109934772A (en) * | 2019-03-11 | 2019-06-25 | 深圳岚锋创视网络科技有限公司 | A kind of image interfusion method, device and portable terminal |
| CN110796828A (en) * | 2019-10-31 | 2020-02-14 | 的卢技术有限公司 | Method and system for searching lost population based on Internet of vehicles |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103440478A (en) * | 2013-08-27 | 2013-12-11 | 电子科技大学 | Face detection method based on HOG characteristics |
| CN104899576A (en) * | 2015-06-23 | 2015-09-09 | 南京理工大学 | Face identification feature extraction method based on Gabor conversion and HOG |
| CN105046224A (en) * | 2015-07-16 | 2015-11-11 | 东华大学 | Block self-adaptive weighted histogram of orientation gradient feature based face recognition method |
| CN105893916A (en) * | 2014-12-11 | 2016-08-24 | 深圳市阿图姆科技有限公司 | New method for detection of face pretreatment, feature extraction and dimensionality reduction description |
| CN106250811A (en) * | 2016-06-15 | 2016-12-21 | 南京工程学院 | Unconfinement face identification method based on HOG feature rarefaction representation |
-
2017
- 2017-12-22 CN CN201711408669.4A patent/CN108038464A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103440478A (en) * | 2013-08-27 | 2013-12-11 | 电子科技大学 | Face detection method based on HOG characteristics |
| CN105893916A (en) * | 2014-12-11 | 2016-08-24 | 深圳市阿图姆科技有限公司 | New method for detection of face pretreatment, feature extraction and dimensionality reduction description |
| CN104899576A (en) * | 2015-06-23 | 2015-09-09 | 南京理工大学 | Face identification feature extraction method based on Gabor conversion and HOG |
| CN105046224A (en) * | 2015-07-16 | 2015-11-11 | 东华大学 | Block self-adaptive weighted histogram of orientation gradient feature based face recognition method |
| CN106250811A (en) * | 2016-06-15 | 2016-12-21 | 南京工程学院 | Unconfinement face identification method based on HOG feature rarefaction representation |
Non-Patent Citations (6)
| Title |
|---|
| 张新平: ""Laplace平滑变换在人脸识别应用中的问题及相关算法"", 《中央民族大学学报(自然科学版)》 * |
| 张梦霞等: ""基于HOG低秩恢复与协同表征的人脸识别方法"", 《电视技术》 * |
| 张秀琴: ""基于DCT和分块2D2PCA的人脸识别"", 《太原科技大学学报》 * |
| 杨恢先等: ""拉普拉斯滤波结合CS-LBP的单样本人脸识别"", 《计算机工程与应用》 * |
| 王小鹏: ""基于图像融合和DCT域增强的人脸识别"", 《兰州交通大学学报》 * |
| 顾岁成等: ""Laplace平滑变换及其在人脸识别中的应用"", 《中国科学:信息科学》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108961218A (en) * | 2018-06-11 | 2018-12-07 | 无锡维胜威信息科技有限公司 | Solar power silicon platelet spends extracting method |
| CN108961218B (en) * | 2018-06-11 | 2021-07-02 | 无锡维胜威信息科技有限公司 | Solar silicon wafer crystal flower extraction method |
| CN109934772A (en) * | 2019-03-11 | 2019-06-25 | 深圳岚锋创视网络科技有限公司 | A kind of image interfusion method, device and portable terminal |
| CN109934772B (en) * | 2019-03-11 | 2023-10-27 | 影石创新科技股份有限公司 | Image fusion method and device and portable terminal |
| CN110796828A (en) * | 2019-10-31 | 2020-02-14 | 的卢技术有限公司 | Method and system for searching lost population based on Internet of vehicles |
| CN110796828B (en) * | 2019-10-31 | 2022-06-14 | 的卢技术有限公司 | Method and system for searching lost population based on Internet of vehicles |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107316031B (en) | Image feature extraction method for pedestrian re-identification | |
| Quan et al. | Dynamic texture recognition via orthogonal tensor dictionary learning | |
| CN102360421B (en) | Face identification method and system based on video streaming | |
| Tapia et al. | Gender classification from iris images using fusion of uniform local binary patterns | |
| CN105825183B (en) | Facial expression recognizing method based on partial occlusion image | |
| Schwartz et al. | Visual material traits: Recognizing per-pixel material context | |
| CN103279745B (en) | A kind of face identification method based on half face multiple features fusion | |
| CN105574534A (en) | Significant object detection method based on sparse subspace clustering and low-order expression | |
| CN106127196A (en) | The classification of human face expression based on dynamic texture feature and recognition methods | |
| CN105139039A (en) | Method for recognizing human face micro-expressions in video sequence | |
| CN103632132A (en) | Face detection and recognition method based on skin color segmentation and template matching | |
| Zhang et al. | Graph fusion for finger multimodal biometrics | |
| CN102855468A (en) | Single sample face recognition method in photo recognition | |
| CN106021330A (en) | A three-dimensional model retrieval method used for mixed contour line views | |
| Ukita | Articulated pose estimation with parts connectivity using discriminative local oriented contours | |
| Khan et al. | Gender classification using image processing techniques: A survey | |
| Yu et al. | Improvement of face recognition algorithm based on neural network | |
| CN108038464A (en) | A kind of new HOG features Uygur nationality facial image recognizer | |
| CN106611158A (en) | Method and equipment for obtaining human body 3D characteristic information | |
| CN104376312B (en) | Face identification method based on bag of words compressed sensing feature extraction | |
| Vu et al. | Improving texture categorization with biologically-inspired filtering | |
| He et al. | Structure-preserving texture smoothing via scale-aware bilateral total variation | |
| Pukhrambam et al. | A smart study on medicinal plants identification and classification using image processing techniques | |
| Wang et al. | Face recognition based on nonsubsampled contourlet transform and block-based kernel fisher linear discriminant | |
| Qiao | Application of Gabor image recognition technology in intelligent clothing design |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180515 |