WO2017051116A1 - Method for automatically distinguishing between a skin from a human being and an inanimate decoy of human skin - Google Patents

Abstract

This method comprises: - the estimation (102), on the basis of the reflectance of an object, of at least one parameter chosen from the group composed of: a volume fraction of melanin contained in a layer of epidermis, a volume fraction of haemoglobin contained in a layer of the dermis, a proportion of melanin present in eumelanin form, a proportion of melanin present in pheomelanin form, a proportion of haemoglobin present in the form of oxygenated haemoglobin, a proportion of haemoglobin present in the form of deoxygenated haemoglobin, a thickness of the layer of epidermis, and a thickness of the layer of the dermis; - the automatic classification (104) of this object in one or other of the following categories depending on said at least one parameter estimated: a category containing skins from a living human being and a category containing inanimate decoys of human skins.

Description

 METHOD FOR AUTOMATICALLY DISTINGUISHING A SKIN OF A HUMAN BEING FROM AN INANIMATE HUMAN SKIN LURE

[001] The invention relates to a method for automatically distinguishing a skin of a human from an inanimate lure of human skin. The invention also relates to an information recording medium and an electronic calculator for implementing this method.

 [002] In known manner, facial recognition systems or fingerprint can identify a human individual from a digital image of the face or fingerprint of this individual. This digital image can in particular be obtained from a photograph. Such systems play an important role in public safety and are a non-intrusive means of biometric identification.

 [003] A disadvantage of these methods is that it is possible for an individual to impersonate another individual ("spoofing" in English) by presenting to the system a lure specifically designed for this purpose such as a digital image, a sequence of images, a face mask, etc. For example, the wearing of a latex mask with the image of a particular person, made to measure by means of a three-dimensional printing process, allows an individual wearing this mask to be falsely recognized as being this nobody.

 [004] The patent application US2004 / 0240712 describes a biometric identification system capable of automatically identifying whether an object whose digital image is acquired, is formed of skin of a human being or if it is a question of 'an inanimate lure of human skin.

[005] In this text, the terms "skin of a human being" and "human skin" are considered as synonyms. It is a living matter. Unless stated otherwise, the term "artificial skin" refers to an inanimate lure of human skin. For example, such a lure is only formed of inert and non-living matter. This lure is not a tissue of human skin. It may be a piece of latex or silicone whose texture and / or external color are specifically chosen to mimic the appearance of human skin. Subsequently, the terms "human skin inanimate lure" and "artificial skin" are considered synonymous.

[006] The system of the US2004 / 0240712 patent application is based on measurements of the characteristics of the light waves reflected by the object to be identified at different wavelengths and at different locations. For a given wavelength, these measurements form an "image" of the object. Typically, the measured characteristic is the intensity of the reflected light waves. The distinction between human skin and artificial skin is made from the eigenvalues of a matrix composed of different images of the object measured at different lengths wave. This process works properly. However, it is desirable to further improve the reliability of this method of distinguishing between human skin and artificial skin.

 [007] From the state of the art is also known from:

- WO2007 / 084099A2,

 - Nixon K. A. and Al: "Novel spectroscopy-based technology for biometrics and liveness verfication", OPTOMECHATRONIC Micro / Nano Devices and Components III: 8-10 October 2007, Lausanne, Switzerland, vol. 5404, No. 1, 12/4/2004, pages 287-295,

 - WO2008 / 139631A1,

- ALKAWAZ MOHAMMED HAZIM and Al: "The Correlation Between Blood Oxygenation Effects and Human Emotion Towards Facial Color Skin of Virtual Human", 3D Research, 3D Display Research Center, Heidelberg, vol. 6, No. 2, 31/03/2015, pages 1-16,

 - WO2007 / 027579A2.

[008] The invention therefore relates to a method for automatically distinguishing a skin of a human from an inanimate lure of human skin according to claim 1.

 [009] The inventors have discovered that by estimating at least one parameter that corresponds to a real physiological characteristic of the skin of a human being and using it to classify an object is in the category of human skin or in the category artificial skins, the process made it possible to distinguish human skin from artificial skin with an increased success rate.

 Embodiments of the invention further have the following advantages:

- Use among the set of parameters that characterize a skin of a human being, at least the volume fractions f _me and f _hg can obtain very good results. Indeed, it was found that the volume fractions f _me and f _hg are very good discriminant to distinguish human skin an artificial skin.

 - Using the difference between the reflectance of the skin in the presence and absence of a light source to estimate the parameters of the skin can improve the accuracy of this estimate.

 According to another aspect, the invention relates to an information recording medium comprising instructions for carrying out the above method, when these instructions are executed by an electronic computer.

In another aspect, the invention relates to an electronic calculator for automatically distinguishing a human skin from an inanimate lure of human skin, the calculator being in accordance with claim 10.

The invention will be better understood on reading the description which follows, given solely by way of nonlimiting example and with reference to the drawings in which: - Figure 1 is a schematic illustration of a device for automatically distinguishing a human skin from an artificial skin;

 FIG. 2 is a schematic illustration of a computing unit of the device of FIG. 1;

 - Figure 3 is a schematic illustration of a cross section of a piece of human skin;

 FIG. 4 illustrates the evolution of the value of light absorption coefficients for different constituents of the skin of FIG. 3, as a function of the wavelength of incident light radiation;

 FIG. 5 is a flowchart of a method for automatically distinguishing human skin from artificial skin;

 FIG. 6 illustrates the evolution of the value of optical reflectance coefficients, as a function of the wavelength of incident light radiation, for different objects measured with the aid of the device of FIG. 1;

 FIG. 7 illustrates the result of a classification of a plurality of objects by means of the method of FIG. 5;

 FIG. 8 is a flow diagram of another embodiment of the method of FIG. 5;

 FIG. 9 is a flow chart of a method for automatically classifying a skin of a human being according to the age of this human being.

 In these figures, the same references are used to designate the same elements.

 In the following description, the features and functions well known to those skilled in the art are not described in detail.

FIG. 1 represents an example of a device 2 for automatically distinguishing whether an object 4 is a human skin or an artificial skin.

 The device 2 is configured to acquire optical reflectance values of the object 4 for at least five different wavelengths between 390 nm and 780 nm. These wavelengths correspond here to the field of visible light. This makes particularly simple the realization of the light source which will illuminate the object 4 during the reflectance measurements.

Preferably, here, the device 2 is capable of acquiring these values for more than 50 or 100 narrow ranges of different wavelengths within an interval I = [390 nm; 780 nm]. Each of these narrow ranges of wavelengths is generally less than 10 nm or 15 nm in width. These narrow beaches are immediately juxtaposed with each other without overlapping. Thus, the device 2 makes it possible to acquire an almost continuous optical reflectance curve over this interval I. By abuse of language, in this text, the device 2 is said to measure a value of the reflectance of the object 4 to a length given wave then that in reality it measures this reflectance over a narrow range of wavelengths containing this given wavelength.

The device 2 comprises for this purpose:

 a source 6 of visible light for illuminating at least one outer face of the object 4;

 an imaging sensor 8 for measuring the optical reflectance of said face of the object 4 when it is illuminated by the source 6;

 a calculation unit 10.

In this description, the optical reflectance R of the object 4 is defined as the ratio R = Φ _Γ / Φ, where

 - Φ, is the energy flow received by this object when it is illuminated by the source 6, and

- Φ _Γ is the energy flux radiated by this object in response to its illumination by the source 6.

 The source 6 is able to emit light radiation at several different wavelengths in the range I. In particular, the source 6 is capable of emitting light radiation for one or more specific values of these lengths. wave.

 The sensor 8 measures the reflectance for a given area of the object 4. Here, the sensor 8 measures the reflectance of the entire surface of the object 4 which has been illuminated by the source 6. The sensor 8 generally comprises several transducers arranged in rows and / or columns to form a matrix of transducers. Each transducer transforms the energy of the light radiation received on its sensitive surface at a given wavelength into an electrical quantity representative of this energy. This sensor 8 thus generates an image of the object in which each pixel corresponds to the sensitive surface of one of these transducers. In this image, each pixel is associated with several values of the reflectance obtained for different wavelengths radiated by the source 6. Such a sensor 8 is known under the term "hyper-spectral camera" and the image measured by this sensor is often called "hyper-spectral image". For example, the sensor 8 is a CCD type image sensor ("charge-coupled device" in English) or CMOS. An example embodiment of such a hyper-spectral camera is described in the following article: "Practical Spectral Photography", Ralf Habel and Michael Kudenov and Michael Wimmer, Eurographics 2012, Vol 31, Cagliari, Italy Sensor 8 is here connected to the unit 10 by a data transport link.

FIG. 2 represents an exemplary unit 10. The unit 10 comprises here:

 an information recording medium;

 a programmable electronic calculator 22;

 an interface 24 for exchanging data.

The support 20 comprises the instructions for executing the method of FIG. 5, 8 or 9. The calculator 22 reads and executes the instructions recorded on the support 20. The interface 24 makes it possible to acquire data, for example from the sensor 8. The interface 24 also makes it possible to transmit data, for example, to a human-machine interface or an electronic access control device for indicate whether the object 4 is human or artificial skin. Advantageously, the unit 10 is able to selectively control the start and stop of the source 6.

 For example, the unit 10 is a microcomputer. The support 20 comprises a non-volatile memory of Flash or EEPROM type or a magnetic recording medium. The computer 22 is here a programmable microprocessor of the 8086 family of the INTEL® company.

 When human skin is illuminated by incident light radiation, the latter undergoes several types of interaction with the skin. Generally, some of this incident radiation is reflected on the surface of the skin. Another part of the incident radiation penetrates inside the skin and is then diffused there. From there, a portion of penetrating radiation is then backscattered out of the skin while another portion is absorbed. This backscattered radiation and the reflected radiation have properties that make it possible to obtain information on characteristics of the layers of the skin situated under its outer face. The absorption of penetrating radiation is notably caused by pigments contained in the skin, such as hemoglobin and melanin, which are able to interact with light. The phenomena of reflection and backscattering are notably caused by the interaction of the light radiation with molecules or particles of the skin whose size is typically less than or equal to the wavelength of the light radiation.

These interactions can be described and quantified by a physical function that relates the reflectance of the skin to the histological characteristics of the skin. An example of a function is known by the acronym BRDF for "Bidirectional reflectance distribution function" in English. A more complete example of implementing a physical model for performing the BRDF function is described in more detail in the article by Claridge E., COTTON SD: "Developing a Predictive Model of Human Skin Coloring", SPIE Procedures , Flight. 2708 Medical Imaging 1996: Physics of Medical Imaging (1996), pp 814-825.

FIG. 3 shows in greater detail a cross-section of a sample of human skin as modeled using the BRDF model.

In this example, the human skin is modeled as comprising:

 a layer of epidermis 32, of which one face 33 is turned towards the outside of the skin, and

a layer of dermis 34 placed under the layer 32. This layer is divided into two layers which are: the upper papillary layer ("Upper Papillary Dermis" in English) and the lower papillary layer ("Lower Papillary Dermis" in English). ). In this model, the optical reflectance of the skin is a function of the absorption and diffusion coefficients ("scattering" in English) of the layers 32 and 34.

 In this model, the radiative transfer equation is expressed as follows:

{£ ^■ · V) L {X, uj) = - - Q (X Ù).

or :

- _t = o o _a o + _s is the extinction coefficient, where _a is the absorption coefficient and _{a n} the middle of the diffusion coefficient

 L (x, ω) is the irradiance at position x in the direction ω, ω being a vector,

- ρ (ω ^■ ω ') is the phase function describing the angular distribution of the light intensity scattered by particles of matter, this function being normalized as follows:

 I p {û - ω *) άω '= 1

Q (x, ω) is the emission term corresponding to the source of the radiation,

 - the symbol " . Is the scalar product operation, and

- the symbol "" designates the gradient operation.

The layers 32 and 34 are directly superimposed one above the other. When the sample 30 is illuminated by source 6, a portion of the incident light enters the interior of the layer 32 and another portion is reflected by the face 33. In this model, the layer 32 has a thickness of ^epi typically between 0.05 mm and 0.30 mm. The layer 34 has a ^derm thickness typically between 0.5 mm and 3.5 mm. The thicknesses are here measured in a direction Z perpendicular to the face 33.

 The pigments of the skin are distributed in layers 32 and 34. These pigments are mainly melanin and hemoglobin. At least 80%, and preferably at least 90% or 95%, by volume of the melanin is contained in the layer 32. At least 80%, and preferably at least 90% or 95%, by volume of the hemoglobin is contained in layer 34.

In a known manner, melanin can be in two different forms: pheomelanin or eumelanin. Note o ^pm _has the optical absorption coefficient of pheomelanin and o ^em _is the optical absorption coefficient of eumelanin. For example, in this embodiment, these coefficients o _a ^pm and σ, are given by the following relationships:

'(X) = - g. xix λ mm The hemoglobin is either in an oxygenated state or in a deoxygenated state. In the oxygenated state, hemoglobin is called "oxygenated hemoglobin". In the deoxygenated state, hemoglobin is called "deoxygenated hemoglobin". The respective absorption coefficients of oxygenated hemoglobin and oxygen deoxygen are noted o _to ^oh and o _to ^dh . For example, the values of these coefficients o _a ^oh and o _a ^dh are obtained experimentally.

+ (1 - i "*) + (1 - f_me)a σ demi The absorption coefficients of layers 32 and 34, respectively denoted o _a ^ei and o _a ^derm , are given by the following formulas:

+ (1 - i "*) + (1 - f _me ) a σ half

fhg r _oh (rf + (1-r _oh ) to ^l ) + (l - Sbg) < ^p where:

f _me is the volume fraction between the volume of melanin contained in the layer 32 and the volume of this layer 32,

 bme is the proportion of melanin present in the form of eumelanin, this proportion being between zero and one,

- (lb _me ) is the proportion of melanin present in the form of pheomelanin,

- f _hg is the volume fraction between the volume of hemoglobin contained in the layer 34 and the volume of this layer 34,

- _oh r is the proportion of hemoglobin in the form of oxygenated hemoglobin, this proportion being between zero and one,

- (1 - r ₀ ) is the proportion of hemoglobin present in the form of deoxygenated hemoglobin,

- o _a ^d is the coefficient of absorption of the depigmented skin, that is to say in the absence of melanin and hemoglobin. This absorption coefficient o _a ^d represents the contribution of skin absorption that is not caused by pigments. This absorption is for example caused by cellular structures of the skin, such as membranes or organelles. This coefficient is, for example, determined by means of the following formula: ## EQU1 ## ₍ ^d = 0.0244 + 8.53 x _θ ^{"(λ-154) / 66} ' ² (expressed in mm ¹ )

In this description, the volume fraction of a pigment present in one of the layers 32 or 34 is defined as being the ratio between the volume occupied by this pigment and the total volume of the constituents of this layer. The fractions f _me and f _hg belong here to the interval [0; 1].

Typically, the ratio r ₀ is between 0.6 and 0.8. Here, the ratio r ₀ is taken equal to 0.75. The ratio b is _me in the interval [0; 1]. FIG. 4 represents (in ordinate) the values of the absorption coefficients o _to ^pm , o _to ^em , o _to ^oh , o _to ^dh and o _to ^d of the skin for the different values of wavelength ( in asbcisse) of the visible domain. More precisely, the curves 40, 42, 44, 46 and 48 respectively correspond to the absorption coefficients o _to ^m , o _to ^em , o _to ^oh , o _to ^dh and o _to ^d . These absorption coefficients are expressed in mm ¹ . Some of these values have, for example, been obtained from the information available on the following website: http://omlc.org/spectra/.

The diffusion coefficients of the layers 32 and 34, respectively denoted o _s ^ei and o _s ^derm , are obtained by calculating, for each layer, the sum of the Rayleigh and Mie scattering coefficients.

For example, the Rayleigh coefficient of the layer 32 is calculated by means of the following function: _s ^Rayleigh = 2χ10 ^η χλ ^{~ 4} mm ¹ , where "λ" denotes the wavelength of the light radiation, expressed in nanometers and "x" denotes the operator of multiplication between two real numbers.

[0043] For example, the coefficient of Mie layer 32 is calculated by the following function: _s o ^Mie 2xl0 = ⁴ xA ^{the 5} mm ^_1.

The coefficient o _s ^derm of the layer 34 is here estimated to be 1.5 times higher than the coefficient o _s ^ei .

[0045] The absorption coefficients ^ei o _a, o _a ^derm, o o _{s s} ^derm and ^ei are connected to the total reflectance R of the skin. For example, for each layer 32, 34, the absorption and diffusion coefficients of this layer make it possible to calculate the reflectance and the transmittance of this layer by means of the following relations:

where "d" is the thickness of this layer, and the coefficients "K" and "β" are given by the following formulas:

K = y'4 <¾ + 8σ "σ _Λ - and

H y 2., + 4er _I in which o _a and o _s are respectively the absorption and diffusion coefficients of this layer.

 From the reflectance and the transmittance of each layer, it is possible to calculate the total reflectance of the skin sample. For example, the total reflectance R of the sample is expressed as follows, in particular to take account of the reflections that can occur between layers 32 and 34:

R = Ri + Ti XR ₂ X Ti + Ti XR ₂ X Ri XR ₂ X Ti ... where R, and Ti represent the reflectance and transmitivity of the ith layer, i being an integer index equal to 1 or 2, the index i = 1 denoting the layer 32 and the index i = 2 denoting the layer 34

Assuming that Ri and R ₂ are each strictly less than 1, it can be shown that this series converges to a finite limit value, which gives the following expression of the reflectance R:

R = Ri + (Ti x R ₂ x Ti) / (1-Ri x R ₂ ).

Thus, knowing the volume fractions f _me and f _hm , the proportions b _me , r ₀ and the thicknesses of ^epi and d ^derm of a piece of skin and using the mathematical model above, it is possible to estimate the total reflectance of this piece of skin. On the other hand, knowing the total reflectance R for at least five different wavelengths and using this mathematical model, it is possible to estimate the values of the volume fractions f _me and f _hm , of the proportion b _me and thicknesses ^{epi derm} and the corresponding piece of skin when the ratio r ₀ is set in advance to 0.75.

 An example of implementation of a method for automatically distinguishing whether an object is a sample of human skin or artificial skin will now be described, with reference to the flow chart of FIG. 5 and with the help of Figures 1 to 4.

In a step 98, the object 4 is presented in front of the source 6 and the sensor 8 measures the intensity of the radiation reflected by the object 4 for several tens of different wavelengths in the range I. For example, the surface of the object 4 is illuminated uniformly with the source 6 for a given wavelength value and with a given energy flow. Then, the energy flux reflected and reemitted from the object 4 is measured using the sensor 8. This operation is then repeated for a large number of different wavelengths.

In a step 100, the unit 10 acquires the measurements of the sensor 8 and calculates for each pixel the corresponding values of the reflectance. The unit 10 thus obtains a hyper-spectral image of the object 4. Here, these reflectance values are calculated for more than ten or twenty different wavelengths all contained within the interval I to obtain a practically continuous curve of reflectance.

 FIG. 6 represents the evolution, as a function of the wavelength, of a plurality of reflectance curves measured for several different objects 4. For example, in FIG. 6, the curves bearing the references P and N correspond, respectively, to human skin, to an artificial silicone skin and to a paper decoy.

Then, during a step 102, values of the fractions f _me and f _hg , of the proportion b _me and the thicknesses of ^epi and d ^derm which correspond to the measurements acquired. during step 100 are automatically estimated for each of the pixels of the hyper-spectral image. For this, the unit 10 uses the acquired reflectance curve and the mathematical model previously described. For example, the unit 10 estimates the values of the fractions f _me and f _hg , of the proportion b _me and the thicknesses of ^epi and d ^derm which minimize the quantity ε:

 λ

or :

Rmeasured (A) denotes the reflectance obtained from the measurements acquired during step 100 for a value of wavelength λ,

 modeied (A) designates the reflectance calculated by the computer 22 for this wavelength λ using the mathematical model previously described,

the operator _λ denotes the summation over all the selected wavelengths, here those of the interval I, and

 - the operator | ... | denotes the absolute value.

This minimization is carried out here using conventional mathematical optimization methods. In this example, the model is parameterized by the fractions f _me and f _hg and, advantageously, by the parameter b _me , the thickness of ^epi and the thickness d ^derm . For example, predetermined initial values of the parameters bme, ^epi , d ^derm , fme and f _hg are chosen. For each of the wavelengths of the interval I, the calculator 22 then calculates the value R _m0 deied, using the parameters b _me , ^epi , d ^derm , f _me and f _hg . Then, the operation is repeated with other values of the parameters b _me , d ^ei , d ^derm , f _me and f _hg until finding the values of the parameters which minimize the quantity ε. The values of f _me and f _hg which minimize this quantity ε are then retained and said to be the estimated values of the fractions f _me and f _hg .

Then, during a step 104, the object 4 is automatically classified, according to the estimated values of the fractions f _me and f _hg for each of the pixels of the hyper-spectral image during step 102, in one of the following categories:

 - a category containing human skin, and

 - a category containing artificial skins.

In this example, this classification is performed by means of a classifier ("classifier" in English), implemented by the computer 22. For example, the classifier is a support vector machine using the core function known as the name of "radial basis function kernel" in the English language. This classifier is previously trained from a set of training data. These learning data comprise pairs of values (f _me , fhg) each associated with an object 4 which is known beforehand to which category it belongs. As this classifier is well known, it is not described here in more detail. For example, during step 104, the estimated pair of values (f _me , f _hg ) for each pixel of the hyper-spectral image is automatically ranked in one or the other of these categories. If the number of pixels of the hyper-spectral image classified in the category of human skin is greater than a threshold S _p , then the object 4 is classified in the category of human skin. If the number of pixels of the hyper-spectral image classified in the human skin category is less than a threshold S _a , then the object 4 is classified in the category of artificial skins. By way of illustration, here the thresholds S _p and S _a are equal so that any object 4 is either classified in the category of human skin or classified in the category of artificial skin. The threshold S _p is here equal to 0.5NP or 0.7NP or 0.8NP, where NP is the number of pixels in the hyper-spectral image.

 The classification result is then advantageously provided by the device 2 to a user of the method via a human-machine interface or to an electronic access control device.

The inventors have determined that by using the histological characteristics fme and f _hg , objects 4 could be classified with a better success rate.

An application example of the method will now be described.

For example, the method was tested on a set of reflectance values measured for 170 objects and provided in the article by MJ Vrhel et al, "Measurement and analysis of spectra reflectance object", Color Research & Application, vol . 19, p. 4-9, 1994. Of these 170 objects, 30 relate to human skin. The rest concerns artificial skins made from silicone or other plastic material, textiles or plants.

 The steps 102 and 104 of the method were applied successively for each of these objects. The inventors determined that the process had made it possible to classify these objects in the two aforementioned categories with a success rate of 96.4%. This level is higher than that obtained from other known methods.

[0063] FIG. 7 represents the pairs of values (f _me , f _hg ) of these objects.

Each of the objects, and therefore pairs of values (f _me , f _hg ), is uniquely associated with a point, whose abscissa has the value f _me and for ordinate the value f _hg of this pair. This result is in the form of a two-dimensional graph whose origin is the value (f _me , fhg) = (0; 0), for abscissa the parameter f _me and for ordinate the parameter f _hg . In this figure, the points associated with the objects that are skin are represented by a circle 110. The points associated with objects that are not skin are represented by a cross 112.

It is found that most points corresponding to human skin are concentrated near the origin of the graph. Specifically, the vast majority of points that correspond to human skin are located within a Rff region. Most of the points corresponding to an object that is not skin are distributed elsewhere. This result comes from the fact that a skin model is used which takes into account the interactions between radiation and the pigments of human skin. When the object 4 is human skin, it is possible to relate the measured reflectance to the properties of these pigments and thus to find values (fme, fhg) that correspond to plausible histological characteristics for the skin. On the contrary, when the object 4 is not skin, the estimated values of the fractions f _me , f _hg are generally far from the plausible values so that it is then easy to identify an artificial skin. Typically, in the latter case, the estimated values of the fractions f _me , fg are outside the region Rff. Indeed, pigments such as hemoglobin and melanin are found only in the skin of a living human and not in inert materials that are not human skin.

 Thus, the present method obtains better results than the methods which do not take into account the interactions between the light radiation and the skin which take place in the deep layers of the skin, and not only on the surface of the skin. skin.

 Many other embodiments are possible. For example, the device 2 may be different. In particular, the source 6 can be realized in another way. For example, several monochromatic sources can be used to successively illuminate the object 4 at the desired wavelengths. Another example is to use as source 6, natural lighting as the sun.

 The sensor 8 may be different. For example, this sensor 8 comprises a single transducer and the method described above is then implemented for a single point of the skin and not for multiple points of the skin each corresponding to a pixel as described above.

In another embodiment, the ignition and, alternately, the extinction of the source 6 is controlled by the unit 10. In this embodiment, the step 98 is executed at least once then that the source 6 is off and at least once while the source 6 is on. In step 100, the unit 10 calculates a first hyper-spectral image from the measurements made only when the source 6 is off and a second hyper-spectral image from the measurements made only when the source 6 is lit. Then, again during this step 100, the unit 10 builds a differential hyper-spectral image corresponding to the difference between the first and second hyper-spectral images. For example, for this, for each pixel of the second hyper-spectral image, the unit 10 subtracts from the reflectance of the pixel of the second image, the reflectance of the pixel located at the same position in the first image. Next, the following steps of the method are the same as previously described except that the differential hyper-spectral image constructed in place of a simple hyper-spectral image is used. This eliminates the contribution of parasitic light sources also present in step 98 in addition to the source 6. This therefore improves the reliability of the process.

 The wavelengths can be chosen differently. For example, the spectral domain may cover the visible range as well as the near infrared or infrared. The sensor 8 is then a hyper-spectral sensor capable of acquiring hyper-spectral digital images. By near infrared, we designate the portion of the electromagnetic spectrum whose wavelength is between 800 nm and 2.5μίτι. The spectral domain can also cover the infrared domain. For example, the wavelengths for which measurements are acquired are then between 390 nm and 1 mm.

Alternatively, it is not necessary to acquire a curve of reflectance values for all the wavelengths of the interval I; reflectance values may be sufficient for certain wavelengths selected within this range and distant from each other by wavelength ranges greater than 10 nm in width. It is also possible to retain only the discriminating wavelengths. These are obtained beforehand by analyzing the reflectance of the skin at different wavelengths. In particular, the inventors have determined that the following wavelengths are particularly discriminating and advantageously make it possible to improve the reliability of the process: 516 nm, 540 nm, 564 nm and 576 nm. These wavelengths correspond to a particular pattern present on the optical reflectance curve, mainly caused by the evolution of the coefficient o _a ^oh .

In step 102, the number of parameters whose values are set in advance may be more important. For example, in addition to or instead of the value of the parameter r _oh , the values of the parameters ^ep and ^dderm are fixed in advance. This simplifies the estimation of the values of the other parameters. Conversely, the number of parameters whose values must be estimated may be larger. For example, in a particular embodiment, the value of the parameter r ₀ is considered to be unknown and must be estimated in step 102. Advantageously, in addition to the estimated parameters, such as the fractional volume fractions f _me and f _hg , one can use a parameter V _f defined as follows to classify an object either in the category of human skins or in the category of artificial skins. The parameter V _f is defined according to the reflectance values R of the object 4 for these wavelengths:

or : - Vi takes the Boolean value "true" if the expression "R (A = 516nm)> R (A = 540nm)" is true, and takes the value "false"otherwise;

- v ₂ takes the Boolean value "true" if the expression "R (A = 540nm) <R (A = 564nm)" is true, and takes the value "false" otherwise, and

- v ₃ takes the Boolean value "true" if the expression "R (A = 564nm)> R (A = 576nm)" is true, and takes the value "false" in the opposite case,

 - R (A = yyy nm) is the measured value of the reflectance at the wavelength yyy nm.

This parameter V _f is then taken into account during step 104 to classify the object 4. This variant was applied to the set of test previously defined, and allowed to detect artificial skin with a rate of success above 99%.

 Step 98 may be omitted when there are already reflectance values. These values have for example been acquired prior to the implementation of the method. For example, such reflectance values can be obtained from the aforementioned article by M. J. Vrhel. In this case, the source 6 and the sensor 8 can be omitted.

Step 102 can be performed differently. In particular, other models can be used to model the interactions between light and skin. For example, we use a model in which the number of layers is different to model the dermis and the epidermis. In another example, a Monte Carlo model can be used to estimate the behavior of the light radiation entering the layers 34 and 32 and thus deduce the volume fractions f _me and f _hg . The US2014 / 0213909 patent describes an example of an alternative model that can be used in place of the models described. It is also possible to combine several of these models with each other. An example of another multi-layer skin model is described in the following article: "The Recognition of Ethnic Groups Based on Histological Skin Properties", C. Malskies and E. Eibenberger and E. Angelopoulou, Vision, Modeling, and Visualization (2011).

According to another variant, the physical model can be replaced by a statistical model of deep learning ("deep learning" in English). According to this variant, during step 102, the reflectance values acquired for the object 4 are compared with known reflectance values already associated with values of the fractions f _me and f _hg . We thus estimate a pair of values (f _me , fhg) for the object 4. We can also use the model known under the name of canonical correlation analysis ("canonical correlation analysis" in English) or SVR (Support Vector Regression). In these latter cases, the model is first built automatically by learning. For example, the method then comprises a step of automatic construction by learning a statistical model that associates the acquired measurements with the volume fractions f _me and f _hg from of several sets of values measured experimentally on skins of different human beings, each set of experimental values comprising:

 the measurement acquired making it possible to calculate the reflectance of the skin of the human being, and

the values of the volume fractions f _me and f _hg measured on the same skin.

 As illustrated in FIG. 8, step 102 is then replaced by a step 202 in which the parameters are estimated by means of this statistical method.

Step 104 may be different. Other classifiers and / or core functions may be used. In particular, linear classifiers can be used, if they are able to implement the method known as the "kernel trick". In other variants, the use of a classifier is omitted. For example, the region Rff (FIG. 7) is predefined experimentally in advance. Then, in step 104, the computer checks whether the coordinate point (f _me , f _hg ) is within the region Rff. If yes, human skin is detected. Otherwise, artificial skin is detected.

Alternatively, in step 104, for the classification of the object 4, the unit 10 additionally takes into account the measurement of the difference between, on the one hand, the acquired reflectance values and on the other hand, a predefined set of reference values. For example, the difference between a measured reflectance curve for the object 4 and a predetermined reflectance curve, which is known to correspond to a sample of human skin, is measured. This predetermined reflectance can for example be obtained previously by numerical simulation. The difference is for example measured in the least squares sense. For example, the difference is defined as being equal to the following quantity: Σ _λ | Rmeasured (A) - R _ref (A) | ² , where:

 Rmeasured (A) denotes the acquired reflectance for a value of wavelength λ,

R _ref (A) denotes the predetermined reflectance, for a value of wavelength λ,

the operator _λ denotes the summation over all the selected wavelengths, here those of the interval I, and

- the operator | ... | denotes the absolute value.

 If this difference is greater than a predetermined threshold, then the object 4 is considered to be artificial skin. If not, it is determined to be human skin. Indeed, the inventors have determined that there is a difference between the theoretical optical reflectance values provided by the model and the actual values, this difference being greater for artificial skin than for human skin. In the current state of the inventors' knowledge, this difference is attributed to the fact that the numerical model used seems to overestimate the thickness of the layer 32.

[0081] The above embodiments have all been described in the particular case where the fractions f _me and f _hg are routinely used to make the distinction between human skin and artificial skin. However, regardless of the embodiment considered, other parameters of the skin model may be used in addition to or in place of the fractions f _me and f _hg to make this distinction. For example, in step 104, it is one of the following parameters or a combination of the following parameters that is used to distinguish human skin from artificial skin:

- the proportion b _me melanin in the form of eumelanin,

 the proportion of melanin present in the form of pheomelanin,

- the proportion of hemoglobin r _oh the form of oxygenated hemoglobin, - the proportion of hemoglobin present in the form of hemoglobin deoxygenated,

- the ^epi thickness of the epidermis, and

- the ^derm thickness of the dermis.

For example, in a particular embodiment, the proportion b _me and the fraction f _hg are used during step 104 but not the fraction f _me . In another variant, during step 104, it is possible to use, in addition to the fractions f _me and f _hg , other parameters for classifying the object 4 either in the category of human skin or in the category of skins. artificial. For example, at least one of the parameters b _me , d ^epi , d ^derm estimated in step 102 is also used in addition during step 104 to classify the object 4 between the human and artificial skin categories. .

According to another embodiment, the device 2 is in addition adapted to be used to classify a human skin in a age class of its owner. The computer 22 is then modified accordingly.

For example, as illustrated in FIG. 9, the method comprises:

 identical steps 298, 300 and 302, respectively, to the steps 98, 100 and 102 of the method of FIG. 5, and

 step 302 is followed by a step 304 of automatic classification of the object 4.

This step 304 is identical to the step 104 except that it also comprises an automatic classification operation of the object 4 in an age class according to the ^derm and ^epi thicknesses estimated among a plurality of classes of ages previously defined. Each age class corresponds to the age class in which the human being to which the piece of skin belongs if the object 4 has been recognized as being human skin.

 For example, the following age classes are defined: 0 to 19 years, 20 to 39 years, 40 to 59 years, 60 years or more. Thus, the method of Figure 9 also provides an estimate of the age of the human being.

In another variant, the method is capable of classifying the object 4 according to other criteria, such that, when the object 4 is a skin of a human being, the sex (male or female) of this human. The methods described above can be used to distinguish human skin on any part of the human body, such as on a face, finger, hand, foot or other.

 In the method of FIG. 9, the operation of distinguishing between human skin and artificial skin may be omitted. In this case, the method of the figure estimates only the age class to which the human being belongs without investigating whether the measures treated correspond to a piece of human skin.

The estimation of the values of the parameters of a model of the reflectance of human skin from a differential hyper-spectral image, as described above, can also be implemented in other processes. devoid of any distinction between human skin and artificial skin. More generally, the use of a differential hyper-spectral image is usable in any process where there are parasitic light sources that can disturb measurements.

Claims

A method for automatically distinguishing a skin of a human from an inanimate lure of human skin, the method comprising: a) acquiring (100) measurements for calculating values of the optical reflectance of an object for at least five different wavelengths between 390 nm and 1 mm; characterized in that this method also comprises: b) the automatic estimation (102), from the measurements acquired during step a) and using a mathematical model of the optical reflectance of the skin of a human being, of at least one parameter chosen in the group consisting of: • a volume fraction, denoted f _I, which is, when the object is the skin of a human being, the ratio between the volume of melanin contained in an epidermis layer of the skin and the volume of this layer 'epidermis, A volume fraction, denoted f _hg , which corresponds, when the object is of the skin of a living being, to the ratio between the volume of hemoglobin contained in a layer of the dermis of this skin and the volume of this layer of dermis, • proportion _me b corresponding, when the object is in the skin of a human, the proportion of melanin present in the form of eumelanin in the skin of this skin layer, • proportion _me p corresponding, when the object is in the skin of a human, the proportion of melanin as pheomelanin in the epidermis of the skin layer, • a proportion r _oh matching, when the object is the skin of a human, the proportion of hemoglobin in the form of oxygenated hemoglobin in the dermis of the skin layer, A proportion r _dh which corresponds, when the object is of the skin of a human being, to the proportion of hemoglobin present in the form of deoxygenated hemoglobin in the layer of the dermis of this skin, · A thickness of ^epi matching, when the object is the skin of a human being, the thickness of the epidermis of the skin layer, and • a thickness of ^derm matching, when the object is the skin of a human being, the thickness of the dermis layer of the skin; the mathematical model of the optical reflectance of the skin of a human being connecting said at least one parameter to be estimated at the reflectance values of this skin at said at least five different wavelengths, c) the automatic classification (104) of the object in one of the following categories according to said at least one parameter estimated in step b): a category containing the skins of a living human being, and a category containing inanimate lures of human skin. 2. Method according to claim 1, wherein said at least one parameter comprises at least the volume fractions f _me and f _hg . The method of any one of the preceding claims, wherein: the method comprises the realization (98) of measurements making it possible to calculate the values of the optical reflectance of the object for at least five different wavelengths between 390 nm and 1 mm, once when these measurements are carried out whereas the object is illuminated by a light source and once when these measurements are made in the absence of illumination of the object by the same light source, and  - In step b), only the difference between the measurements made when the object is illuminated by the light source and the measurements made in the absence of this illumination is taken into account to estimate the said at least one parameter. 4. Method according to any one of the preceding claims, wherein the mathematical model used in step b) comprises first relations which connect the measurements acquired during step a) to the optical absorption coefficients of the layers of epidermis, denoted o _has ^ei, and dermis, denoted o _has ^derm, and second relationships which connect the absorption coefficients _a o and o ^ei _has ^derm, the volume fractions f _I and f _hg and proportions b _me p _me , r _oh , r _dh , these second relations being defined by the following formulas for each of the at least five wavelengths: ° = mc (f ^ra + (i - " ^m ) + (i - i _me ) ° of ™ = Agirait + (1 - r _oh ) af) + (1 - f _b9 ) af - ^ei o _a is the optical absorption coefficient of the skin layer, - b is the proportion _me b _me of melanin present in the form of eumelanin - (l-bme) is the proportion p _me melanin as pheomelanin, - o _a ^em is the optical absorption coefficient of eumelanin, - o _a ^m is the optical absorption coefficient of pheomelanin, - o _a ^d is the optical absorption coefficient of the skin layer in the absence of melanin, - o _a ^derm is the optical absorption coefficient of the dermis layer, - r _{oh oh} r is the proportion of hemoglobin in the form of oxygenated hemoglobin, - (1- r _oh ) is the proportion r _dh of hemoglobin present in the form of deoxygenated hemoglobin, - _a ^oh o is the optical absorption coefficient of the oxygenated hemoglobin, - o _a ^dh is the optical absorption coefficient of deoxygenated hemoglobin. 5. Method according to any one of claims 1 to 3, wherein, before step b), the method comprises a step of automatic construction by learning a statistical model that associates the optical reflectance values at least said parameter estimated from several sets of values measured experimentally on skins of different human beings, each set of experimental values comprising:  the value of the reflectance measured on the skin of the human being, and the values of the said at least one estimated parameter, measured on the same skin. 6. Method according to any one of the preceding claims, wherein step a) comprises the acquisition (100) of measurements for calculating the values of the optical reflectance of the object for at least one hundred different wavelengths. between 390 nm and 1 mm. 7. The method as claimed in claim 1, in which step a) comprises the acquisition of measurements making it possible to calculate the values of the optical reflectance of the object only for wavelengths included in the visible spectrum. between 390 nm and 780 nm. The method according to any one of the preceding claims, wherein, before step a), the method comprises, for said at least five different wavelengths between 390 nm and 1 mm, the measurement (98) of the optical reflectance of the object. 9. Information recording medium (20), characterized in that it comprises instructions for the execution of a method according to any one of the preceding claims when these instructions are executed by an electronic computer. An electronic calculator (22) for automatically distinguishing a human skin from an inanimate human skin lure, which computer is programmed to: a) acquiring measurements for calculating optical reflectance values of an object for at least five different wavelengths between 390 nm and 1 mm; characterized in that this calculator is also programmed for: b) automatically estimate, from the measurements acquired in step a) and using a mathematical model of the optical reflectance of the skin of a human being, at least one parameter selected from the group consisting of : • a volume fraction, denoted f _I, which is, when the object is the skin of a human being, the ratio between the volume of melanin contained in an epidermis layer of the skin and the volume of this layer 'epidermis, A volume fraction, denoted f _hg , which corresponds, when the object is of the skin of a living being, to the ratio between the volume of hemoglobin contained in a layer of the dermis of this skin and the volume of this layer of dermis, • proportion _me b corresponding, when the object is in the skin of a human, the proportion of melanin present in the form of eumelanin in the skin of this skin layer, · Proportion _me p corresponding, when the object is in the skin of a human, the proportion of melanin as pheomelanin in the epidermis of the skin layer, • a proportion r _oh matching, when the object is the skin of a human, the proportion of hemoglobin in the form of oxygenated hemoglobin in the dermis of the skin layer, A proportion r _dh which corresponds, when the object is of the skin of a human being, to the proportion of hemoglobin present in the form of deoxygenated hemoglobin in the layer of the dermis of this skin, A thickness of ^epi which corresponds, when the object is of the skin of a human being, to the thickness of the epidermis layer of this skin, and • a thickness of ^derm matching, when the object is the skin of a human being, the thickness of the dermis layer of the skin; the mathematical model of the optical reflectance of the skin of a human being connecting said at least one parameter to be estimated at the reflectance values of this skin at said at least five different wavelengths, c) automatically classifying the object into one of the following categories based on said at least one estimated parameter in step b): a category containing the skins of a living human being, and a category containing inanimate lures of human skin.