AU2013269822B2 - Method for detecting drop shadows on an initial image - Google Patents

Method for detecting drop shadows on an initial image Download PDF

Info

Publication number
AU2013269822B2
AU2013269822B2 AU2013269822A AU2013269822A AU2013269822B2 AU 2013269822 B2 AU2013269822 B2 AU 2013269822B2 AU 2013269822 A AU2013269822 A AU 2013269822A AU 2013269822 A AU2013269822 A AU 2013269822A AU 2013269822 B2 AU2013269822 B2 AU 2013269822B2
Authority
AU
Australia
Prior art keywords
image
size
step during
value
pixels
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.)
Active
Application number
AU2013269822A
Other versions
AU2013269822A1 (en
Inventor
Pierre ESCAMILLA
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Idemia Identity and Security France SAS
Original Assignee
Idemia Identity and Security France SAS
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Idemia Identity and Security France SAS filed Critical Idemia Identity and Security France SAS
Publication of AU2013269822A1 publication Critical patent/AU2013269822A1/en
Application granted granted Critical
Publication of AU2013269822B2 publication Critical patent/AU2013269822B2/en
Assigned to IDEMIA IDENTITY & SECURITY FRANCE reassignment IDEMIA IDENTITY & SECURITY FRANCE Request to Amend Deed and Register Assignors: MORPHO
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/60—Extraction of image or video features relating to illumination properties, e.g. using a reflectance or lighting model

Abstract

The invention concerns a method (300) for detecting drop shadows on an image which consists of successively reducing the resolution of the captured image, calculating the variance of same and deducing, from said variances, whether or not drop shadows are present.

Description

The invention concerns a method (300) for detecting drop shadows on an image which consists of successively reducing the resolution of the captured image, calculating the variance of same and deducing, from said variances, whether or not drop shadows are present.
(57) Abrege : L'invention conceme un precede de detection (300) d'omhres portees sur une image qui consiste a reduire successivement la resolution de l'image capturee, a en calculer la variance et a partir de ces variances d'en deduire la presence ou non d'omhres portees.
wo 2013/178510 Al llllllllllllllllllllllllllllllllllll^
TR), OAPI (BF, BJ, CF, CG, CI, CM, GA, GN, GQ, GW, ML, MR, NE, SN, TD, TG).
Publiee :
— avec rapport de recherche Internationale (Art. 21(3))
Method for detecting cast shadows on an initial image
The present invention concerns a method for detecting shadows on an image, in particular on an image of a face of an individual, as well as a detection device suitable for implementing such a method. It finds an application in the field of biometric recognition and in particular in the field of identification of an individual by facial analysis.
Facial recognition identification is used for protecting installations such as, for example, buildings or machines, or for obtaining the granting of rights, such as for example the issue of an identity card, the payment of a pension, etc. This technology makes it possible to dispense with access codes or cards, which may be stolen or falsified. The use of this technology reinforces security since the probability that two persons have two identical faces is low.
A method for identifying an individual by facial analysis is known, which comprises a step of capturing an image of the face of said individual, a step of processing shadows cast on the image thus captured, and a step of comparing the image thus processed with reference images stored in a database.
2013269822 27 Jun2018
The step of processing the cast shadows is performed by cast shadow processing software that performs various processing operations tending to make disappear the shadows that are cast on the captured image by the body parts of the individual himself or by objects external to said individual.
The processing operations that are currently used for reducing shadows are not entirely satisfactory since they process all the images, even those that do not have a cast shadow.
Any discussion of documents, devices, acts or knowledge in this specification is included to explain the context of the invention. It should not be taken as an admission that any of the material formed part of the prior art base or the common general knowledge in the relevant art on or before the priority date of the claims herein.
“Comprises/comprising” when used in this specification is taken to specify the presence of stated features, integers, steps or components but does not preclude the presence or addition of one or more other features, integers, steps, components or groups thereof.
It would be desirable to provide a method for detecting shadows on an image that, when it is applied to an image, makes it possible to detect whether this image comprises cast shadows and whether it must be processed in order to reduce the shadows.
To this end, in accordance with a first aspect of the invention, there is provided a 20 method for detecting shadows cast on an image of size 0, said method comprising:
- a learning phase during which a linear discriminant analysis is carried out on each image of a series of test images to determine the characteristics of a “cast shadows” class;
- a capture step during which an acquisition device captures and optionally resizes the image of size 0 of the face of an individual so that the size 0 image has a definition of n*n where n = 2N and where N is a natural integer greater than or equal to 2, for p varying from 0 to N-l
- a division step during which the image of size p is divided into np *np pixels where np = 2(N^p)
- a first step of normalization of the image of size p, during which the value X of each pixel of the image of size p is normalized in the form of a value
Xp
V
ΕΣ<%>!
-=1 7=1
2013269822 27 Jun2018
- a computing step during which the variance Y, of the image of size p is computed,
- a reduction step during which the resolution of the image of size p is reduced by a factor of 2 into an image of size p+1, where the values Xf+1 of the pixels of the image of size p+1 are calculated with respect to the normalized values x, of the pixels of the image of size p,
- an incrementation step during which the index p is incremented by 1,
- a test step during which the value of the index p is compared with N,
- if the result of the test step is negative, a looping step during which the detection method loops onto the division step,
- a second normalization step during which the variance σρ. of each reduced image of size p where p varies from 1 to N-l is divided by the variance σ0οί the image of size 0,
- a vectorization step during which a characteristics vector is constructed from all the variances thus normalized σ0 σ0 σ0
- a multiplication step during which the scalar product of the characteristics vector and the predetermined eigenvector issuing from the LDA is calculated,
- a comparison step during which the scalar product thus calculated is compared with a reference threshold value, and
- a classification step during which the image of size 0 is classified in the cast shadows class if the score is greater than the reference threshold value, and is not classified in the cast shadows class if the score is less than the reference threshold value.
Advantageously, the value Xf+1 of each pixel is a sum of the normalized values xf, of the corresponding four pixels of the image of size p and given by the formula
3a
2013269822 27 Jun2018 yP — γΐ γΐ i i Λ ij 2st=0 £jj=O Xi+J{,j+l yp+1 _ V V ^9 LiZ-i X!+KJ+1 k=0 7=0
In accordance with another aspect of the invention, there is provided a detection device suitable for implementing the above method of the first aspect of the invention, said detection device comprising:
- learning means carrying out a linear discriminant analysis on each image of a series of test images to determine the characteristics of a “cast shadows” class;
- an acquisition device intended to capture and optionally resize an image of size 0 of the face of an individual so that the image of size 0 has a definition of n*n where n = 2N and where N is a natural integer greater than or equal to 2,
- dividing means intended to divide an image of size p into np * np pixels where np = 2(N~P'>,
- first normalization means intended to normalize the value of xf each pixel of the
Xp.
image of size 'p in the form of a value xf = , 11 I n n ,ΣΣ<χΑ \ i=l j=l
- computing means intended to compute the varianceσp, of the image of size p, [continued on page 4]
- reduction means intended to create an image of size “p+1” by reducing the resolution of the image of size “p” by a factor of 2, where the values X/’1 of the pixels of the image of size “p+1” are computed with respect to the normalized values Xf of the pixels of the image of size “p”,
- incrementation means intended to increment the index “p” by 1,
- test means intended to test the value of the index “p” with respect to N,
- second normalization means intended to divide the variance opof each reduced image of size “p” where p varies from 1 to N-l by the variance σθ of the image of size “0”,
- vectorization means intended to construct a characteristics vector from all , , . σ, , the normalized variances —L,,... , σο σο σο
- multiplication means intended to calculate the scalar product of the characteristics vector and the predetermined eigenvector issuing from the LDA,
- comparison means intended to compare the scalar product with a reference threshold value, and
- classification means intended to classify the image of size “0” in the “cast shadows” class if the score is above the reference threshold value, and not to classify the image of size “0” in the “cast shadows” class if the score is below the reference threshold value.
The features of the invention mentioned above, as well as others, will emerge more clearly from a reading of the following description of an example embodiment, said description being given in relation to the accompanying drawings, among which:
Fig. 1 depicts an image of a face of an individual at a first stage of a shadowdetection method according to the invention,
Fig. 2 depicts the image of Fig. 1 at a subsequent stage of the shadow-detection method,
Fig. 3 is an algorithm of a detection method according to the invention, and
Fig. 4 shows a detection device according to the invention.
The invention that will be described below is more specifically oriented to the detection of shadows on an image of a face of an individual, but it applies in the same way if the image depicts only part of said face, such as for example a peripheral region of an eye.
Cast shadows are details of scale and not of frequency, they are localized spatially and have a large size in the case of the most problematic ones, and are characterized by the appearance of an abrupt jump in the image and therefore high variation in the image.
It is sought to characterize the spatial variation according to a scale on a captured image.
Fig. 1 shows an initial image 100 of a face of an individual captured by a suitable acquisition device (402, Fig. 4), such as a CCD sensor, and optionally resized in order to be able to be divided into n*n pixels where n is written 2N and N is a natural integer strictly greater than 1.
The initial image 100 is said to be an image of size “0”.
In the case of an initial grey-level image 100, the value Χθ of each pixel varies between 0 and 255. In order to dispense with variations in luminosity from one initial image 100 to another, the value Χθ of each pixel of the initial image 100 is canonically normalized from 0 to 1 in the form of a normalized value χθ where i is the number of y
the row and j is the number of the column of the pixel in question.
In the embodiment of the invention presented here, the normalization is effected on the basis of the so-called L2 norm and is represented by the formula:
X° *S = | ‘J Μ ,ΕΣσΤ
V 1=1 7=1
According to other embodiments, it is possible to use a different normalization such as for example the so-called LI norm or the so-called infinite L norm.
The initial variance o0of the initial image 100 is calculated from normalized values of the pixels of the initial image 100 by the formula:
σ0 = ZZ(A>-m)2 ’ where m=—— ΣΣ
X.
-=1 7=1 n*n (1).
i=l 7=1
Fig. 2 shows an image 200 the resolution of which has been reduced by 2 with respect to the initial image 100. The reduced image 200 is thus divided into n .n . , — * —pixels.
2
The reduced image 200 is said to be of size “1” since it results from a reduction by two of the resolution of the image of size “0”. In general terms, an image of size “p” results from a reduction by two of the resolution of the image of size “p-1”. The number “p” varies from 0 to N-l.
The image of size “p” is divided into np *n pixels where np = 2(Ν~Ρ>> .
For the image of size “p”, the value of each pixel is denoted Xf, where i is the number of the row and j is the number of the column of the pixel in question.
Each group of four pixels of the image of size “p-1” (100) is grouped together to give a pixel of the image of size “p” (200) and in which the value Xf of each pixel is a sum of the normalized values xU1 of the corresponding four pixels of the image of size “p-1” (100) and given by the formula:
:? Vs 1 ρ—1
Τ’=ΣΣ
X, p-i i+k ,j +Z k=0 1=0 (2).
Each value Xf of the image of size “p” (200) is then normalized according to a formula similar to the formula (0) and the normalized value of each pixel of the image of size “p” (200) is denoted xp and is given, in the case of the L2 norm, by the formula:
xp = xij (3).
<ΣΣ(*Ρ2
V >=1 7=1
The variance σ p of the image of size “p” (200) is calculated by the formula: 1 ΰΡ = np(np -1) i=1 M i
ΣΣ(ΛΜ nA
Figure AU2013269822B2_D0001
(4).
Figure AU2013269822B2_D0002
where
ΣΣThe variance of the image of size “p” (200) is then normalized with respect to the initial variance σθ of the initial image 100 and thusopis divided byo0, and the σ
normalized variance of the image of size “p” is written — (5).
The resolution of the image of size “p” (200) is then reduced in its turn by 2 in order to obtain a new image of size “p+1” and equations 2 to 5 are recalculated.
As long as the number of pixels of the reduced image is greater than or equal to 2*2, a reduction of the resolution of the image is effected and new calculations of equations 2 to 5 are performed.
In this way N normalized variances are obtained, denoted: —1 σο σο σο which form a characteristics vector.
The initial image 100 depends on the acquisition device 402 and the ambient photographing conditions. The variances o0,...,o„are therefore dependent on these conditions and the normalization of the varianceso1,...,a„ by σθ makes it possible to dispense with the acquisition device 402 and the ambient conditions.
In order to implement the detection method, a learning phase makes it possible, from a series of test images, to teach said method which image has cast shadows and which image does not.
To this end, a linear classifier, the role of which is to classify, in one class among several, an image that has properties similar to said class, is installed. The use of the classifier requires a learning mechanism during which the series of test images is distributed into “cast shadows” or “no cast shadows” classes.
A linear discriminant analysis (LDA) is carried out on each image in the series of test images in order to determine the characteristics of the “cast shadows” class and to predict whether each initial image 100 captured in the “cast shadows” belongs or not.
Use of LDA makes it possible to obtain the vectors of the decomposition of the LDA and in particular a vector particular to this decomposition associated with the “cast shadows” class.
In order to determine whether an initial image 100 has cast shadows, it is checked whether this initial image 100 belongs to the “cast shadows” class. For this purpose, the scalar product of the characteristics vector and the eigenvector is produced. The result of this scalar product constitutes a score that can be compared with a reference threshold value. If the score is higher than the reference threshold value, the initial image 100 is then considered to exhibit cast shadows and classified in the appropriate “cast shadows” class, while if the score is lower than the reference threshold value, the initial image is then considered not to exhibit a cast shadow and is not classified in the “cast shadows” class.
Other classification means, such as for example support vector machines, can be used. These are based on the concept of support vectors that represent the closest samples belonging to distinct classes. From these the separating hyperplane that maximises the margin is constructed. This is the distance separating the boundaries of the samples. This hyperplane will then be used to determine belonging to such and such a class.
Thus, by analysing the score, only the initial images 100 actually exhibiting cast shadows will be processed by cast-shadow processing software. This offers a saving in time since only the initial images 100 actually exhibiting cast shadows are processed by the cast shadow processing software, and furthermore the initial images 100 not exhibiting a cast shadow are not degraded by the cast-shadow processing software.
Fig. 3 shows an algorithm of a method 300 for detecting cast shadows on the image of size “0” (100) according to the invention. The detection method 300 comprises:
- a capture step 302 during which the acquisition device 402 captures and optionally resizes an image of size “0” (100) of the face of an individual so that the image of size “0” has a definition of de n*n, with “p” varying 0 to N-l,
- an initial division step 304 during which the image of size “p” (100, 200) is divided into np *np pixels where np = 2^ ^ ,
- a first step 306 of normalization of the image of size “p” (100, 200) during which the value Xp of each pixel of the image of size “p” (100, 200) is normalized in the form of a value .ri’, for example by applying one of the norms LI, L2 or infinite L,
- a computing step 308 during which the variance σ p of the image of size “p” in accordance with the appropriate formula (1) or (4) is computed.
- a reduction step 310 during which the resolution of the image of size “p” is reduced by a factor of 2 to an image of size “p+1”, where the values Xp+1 of the pixels of the image of size “p+1” (200) are computed with respect to the normalized values xp of the pixels of the image of size “p” using more specifically formula (2), which is written Xf1 = , k=0 1=0
- an incrementation step 312 during which the index “p” is incremented by
1,
- a test step 313 during which the value of the index “p” is compared with N,
- if the result of the test step is negative, that is to say if the index “p” is different from N, a looping step 314 during which the detection method loops onto the division step 304,
- a second normalization step 316 during which the varianceo^ of each reduced image of size “p” where p is an integer and varies from 1 to N-l is divided by the initial variance σθ of the image of size “0”,
- a vectorization step 318 during which a characteristics vector is constructed from all the variances thus normalized —, σο σο σο
- a multiplication step 320 during which the scalar product of the characteristics vector and the predetermined eigenvector issuing from the LDA is calculated,
- a comparison step 322 during which the scalar product thus calculated is compared with a reference threshold value, and
- a classification step 324 during which the image of size “0” (100) is classified in the “cast shadows” class if the score is higher than the reference threshold value, and is not classified in the “cast shadows” class if the score is below the reference threshold value.
To implement the method 300, the invention proposes a detection device 400 that is depicted in Fig. 4 and comprises:
- the acquisition device 402 intended to capture and optionally resize an image of size “0” (100) of the face an individual so that the image of size “0” has a definition of n*n where n = 2N and where N is a natural integer greater than or equal to 2,
- division means 404 intended to divide an image of size “p” (100, 200) into np *np pixels where np = 2^1^ ,
- first normalization means 406 intended to normalize the value Xp. of each xp pixel of the image of size “p” (100, 200) in the form of a value x? =
- computing means 408 intended to compute the variance <5 p of the image i=l 7=1 of size “p”,
- reduction means 410 intended to create an image of size “p+1” by reducing the resolution of the image of size “p” by a factor of 2, where the values Xf+1 of the pixels of the image of size “p+1” (200) are computed with respect to the normalized values des pixels of the image of size “p”, the formula being more 1 1 specifically formula (2) that is written X Γ'=ΣΣ<,μ,.
k-0 1=0
- incrementation means 412 intended to increment the index “p” by 1,
- test means 414 intended to test the value of the index “p” with respect to N,
- second normalization means 416 intended to divide the variance σ p of each reduced image of size “p” where p varies from 1 to N-l by the variance σθ of the image of size “0”,
- vectorisation means 418 intended to construct a characteristics vector from all the normalized variances —, SL·. °v 1 , σο σο σο
- multiplication means 420 intended to calculate the scalar product of the characteristics vector and the predetermined eigenvector issuing from the LDA,
- comparison means 422 intended to compare the scalar product with a reference threshold value, and
- classification means 424 intended to classify the image of size “0” (100) in the “cast shadows” class if the score is greater than the reference threshold value, and not to classify the image of size “0” (100) in the “cast shadows” class if the score is less than the reference threshold value.
The division means 404, the first normalization means 406, the computing means 408, the reduction means 410, the incrementation means 412, the test means 414, the second normalization means 416, the vectorization means 418, the multiplication means 420, the comparison means 422 and the classification means 424 are combined in the form of a processing unit taking the form of a computer.
Naturally the present invention is not limited to the examples and embodiments described and depicted but is capable of numerous variants accessible to persons skilled in the art.
For example, in the context of such devices when several images of the same person are required, it is possible to sort the images in order to keep only those that do not exhibit cast shadows.
2013269822 27 Jun2018

Claims (3)

  1. CLAIMS:
    1. Method for detecting shadows cast on an image of size “0”, said method comprising:
    - a learning phase during which a linear discriminant analysis is carried out on each image of a series of test images to determine the characteristics of a “cast shadows” class;
    - a capture step during which an acquisition device captures and optionally resizes the image of size “0” of the face of an individual so that the size “0” image has a definition of n*n where n = 2N and where N is a natural integer greater than or equal to 2, for “p” varying from 0 to N-l
    - a division step during which the image of size “p” is divided into np *np pixels where np = 2(Λ,_/,),
    - a first step of normalization of the image of size “p” , during which the value X,'· of each pixel of the image of size “p” is normalized in the form of a value xp = ‘7 x;
    ΣΣ(χν
    V <=1 7=1
    - a computing step during which the variance σρ of the image of size “p” is computed,
    - a reduction step during which the resolution of the image of size “p” is reduced by a factor of 2 into an image of size “p+1”, where the values X,''1 of the pixels of the image of size “p+1” are calculated with respect to the normalized values xp of the pixels of the image of size “p”,
    - an incrementation step during which the index “p” is incremented by 1,
    - a test step during which the value of the index “p” is compared with N,
    - if the result of the test step is negative, a looping step during which the detection method loops onto the division step,
    - a second normalization step during which the variance σρ of each reduced image of size “p” where p varies from 1 to N-l is divided by the variance σ0 of the image of size “0”,
    2013269822 27 Jun2018
    - a vectorisation step during which a characteristics vector is constructed from all the variances thus normalized , σο σο
    - a multiplication step during which the scalar product of the characteristics vector and the predetermined eigenvector issuing from the LDA is calculated,
    - a comparison step during which the scalar product thus calculated is compared with a reference threshold value, and
    - a classification step during which the image of size “0” is classified in the “cast shadows” class if the score is greater than the reference threshold value, and is not classified in the “cast shadows” class if the score is less than the reference threshold value.
  2. 2. Detection method according to claim 1, wherein the value Xf+1 of each pixel is a sum of the normalized values xf of the corresponding four pixels of the image of size “p” and given by the formula χΡ , yl yl rP-l , , Aij zJ/=o xi+ft,j+l _ γ γ / ? Zj Zj Xi+kT/+7 .fc=0 7=0
  3. 3. Detection device suitable for implementing the method of either claim 1 or 2, said detection device comprising:
    - learning means carrying out a linear discriminant analysis on each image of a series of test images to determine the characteristics of a “cast shadows” class;
    - an acquisition device intended to capture and optionally resize an image of size “0” of the face of an individual so that the image of size “0” has a definition of n*n where n = 2N and where N is a natural integer greater than or equal to 2,
    -dividing means intended to divide an image of size “p” into np*np pixels where np = 2(N~P),
    2013269822 27 Jun2018
    ΣΣ(χΑ <=i >1
    - first normalization means intended to normalize the value X,'- of each pixel of the
    Xp image of size “p” in the form of a value x? =
    - computing means intended to compute the variance σρ of the image of size “p”,
    - reduction means intended to create an image of size “p+1” by reducing the resolution of the image of size “p” by a factor of 2, where the values Xf+1 of the pixels of the image of size “p+1” (200) are computed with respect to the normalized values Xf of the pixels of the image of size “p”,
    - incrementation means intended to increment the index “p” by 1,
    - test means intended to test the value of the index “p” with respect to N,
    - second normalization means intended to divide the variance σροΐ each reduced image of size “p” where p varies from 1 to N-l by the variance σα of the image of size “0”,
    - vectorization means intended to construct a characteristics vector from all the normalized variances —, —,...1 , σο σ{} σ{}
    - multiplication means intended to calculate the scalar product of the characteristics vector and the predetermined eigenvector issuing from the LDA,
    - comparison means intended to compare the scalar product with reference threshold value, and
    - classification means intended to classify the image of size “0” in the “cast shadows” class if the score is above the reference threshold value, and not to classify the image of size “0” in the “cast shadows” class if the score is below the reference threshold value.
    MORPHO
    WATERMARK INTELLECTUAL PROPERTY PTY LTD P39799AU00
    PL. 1/2
    400
    PL. 2/2
AU2013269822A 2012-05-31 2013-05-22 Method for detecting drop shadows on an initial image Active AU2013269822B2 (en)

Applications Claiming Priority (3)

Application Number Priority Date Filing Date Title
FR1255018A FR2991479B1 (en) 2012-05-31 2012-05-31 METHOD FOR DETECTING SHADOWS ON AN INITIAL IMAGE
FR12/55018 2012-05-31
PCT/EP2013/060443 WO2013178510A1 (en) 2012-05-31 2013-05-22 Method for detecting drop shadows on an initial image

Publications (2)

Publication Number Publication Date
AU2013269822A1 AU2013269822A1 (en) 2014-12-18
AU2013269822B2 true AU2013269822B2 (en) 2018-07-19

Family

ID=47191835

Family Applications (1)

Application Number Title Priority Date Filing Date
AU2013269822A Active AU2013269822B2 (en) 2012-05-31 2013-05-22 Method for detecting drop shadows on an initial image

Country Status (5)

Country Link
EP (1) EP2856391B1 (en)
AU (1) AU2013269822B2 (en)
FR (1) FR2991479B1 (en)
NZ (1) NZ702402A (en)
WO (1) WO2013178510A1 (en)

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20100232705A1 (en) * 2009-03-12 2010-09-16 Ricoh Company, Ltd. Device and method for detecting shadow in image

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20100232705A1 (en) * 2009-03-12 2010-09-16 Ricoh Company, Ltd. Device and method for detecting shadow in image

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
Nicolas Morizet, 'Revue des algorihmes PCA LDA et EBGM utilisés en reconnaissance 2D du visage pour la biométrie' (2006-11-24); PCA LDA et EBGM utilisés en reconnaissance 2D du visage pour la biométrie (2013-13-15) *

Also Published As

Publication number Publication date
FR2991479A1 (en) 2013-12-06
EP2856391A1 (en) 2015-04-08
FR2991479B1 (en) 2014-07-25
EP2856391B1 (en) 2019-03-27
WO2013178510A1 (en) 2013-12-05
NZ702402A (en) 2017-07-28
AU2013269822A1 (en) 2014-12-18

Similar Documents

Publication Publication Date Title
US10635946B2 (en) Eyeglass positioning method, apparatus and storage medium
US8457406B2 (en) Identifying descriptor for person and object in an image
WO2019033572A1 (en) Method for detecting whether face is blocked, device and storage medium
Peng et al. Face presentation attack detection using guided scale texture
Alvarez-Betancourt et al. A keypoints-based feature extraction method for iris recognition under variable image quality conditions
CN102938065B (en) Face feature extraction method and face identification method based on large-scale image data
JP4479478B2 (en) Pattern recognition method and apparatus
US8934679B2 (en) Apparatus for real-time face recognition
JP6351243B2 (en) Image processing apparatus and image processing method
JP5936561B2 (en) Object classification based on appearance and context in images
Fradi et al. Crowd density analysis using subspace learning on local binary pattern
CN109376717A (en) Personal identification method, device, electronic equipment and the storage medium of face comparison
Sudhakar et al. An accurate facial component detection using gabor filter
Özseven et al. Face recognition by distance and slope between facial landmarks
KR20140107946A (en) Authentication apparatus and authentication method thereof
US20140341436A1 (en) Classifying materials using texture
AU2013269822B2 (en) Method for detecting drop shadows on an initial image
Kuo et al. Fast pedestrian detection system with a two layer cascade of classifiers
Hassan et al. Facial image detection based on the Viola-Jones algorithm for gender recognition
Das et al. Pedestrian detection using dense LDB descriptor combined with HOG
Hussain et al. Face recognition using multiscale and spatially enhanced weber law descriptor
Hashim et al. Local and semi-global feature-correlative techniques for face recognition
Anila et al. Global and local classifiers for face recognition
Tan et al. A novel weighted hausdorff distance for face localization
Dahmane et al. Local phase-context for face recognition under varying conditions

Legal Events

Date Code Title Description
FGA Letters patent sealed or granted (standard patent)
HB Alteration of name in register

Owner name: IDEMIA IDENTITY & SECURITY FRANCE

Free format text: FORMER NAME(S): MORPHO