EP1747526B1 - Method of masking a digital datum, such as a biometric print, and use thereof for securing a security document - Google Patents

Abstract

The invention relates to a method of masking a plain datum b having n bits. The inventive method is characterised in that a masked datum m is produced using the following masking function: (I), wherein p is a prime number, bi is the bit at position i of plain datum b, and qi is the prime number at position i in a set of prime numbers (q1, . . . , qn). The invention also relates to a method of masking a biometric print, consisting in: determining a set of s real minutiae which are characteristic of the print; mixing and arranging the real minutiae with t false minutiae; and forming a mixed biometric datum b having n=s+t bits, such that, for any i: bi=1 if position i corresponds to a real minutia, and bi=0 if position i corresponds to a false minutia. The invention can be used to secure a security document such as a bank cheque. m = ∏ i = 1 n ⁢ ⁢ q i b i ⁢ mod ⁢ ⁢ p ( I )

Description

The invention relates to identification systems and / or biometric authentication. These systems manipulate biometric data of all types such as, for example, fingerprints, digital fingerprints of the eye, skin, face, or even the voice.

The use of biometric fingerprints is increasingly being considered to supplement user passwords or a manual signature, especially for applications requiring a high level of security. Indeed, the use of a biometric imprint is a good complement to a password or a manual signature, insofar as a biometric imprint can hardly be stolen from its real owner and can not be imitated, copied . In return for this security and to the extent that a biometric fingerprint can not be replaced, it is essential to prevent direct access to this fingerprint to ensure the safety of people and the reliability of the footprint.

For this, one can for example consider known masking methods to hide the biometric fingerprint to secure. The masked fingerprint can then be used instead of the imprint, to sign a message, authenticate a person's identity, and so on. The interest of such methods is the use of hash functions, which are one-way functions, that is, they can not be inverted. In other words, knowing a fingerprint masked by a hash function from a clear fingerprint, it is not possible to find the fingerprint, even knowing all the parameters of the hash function.

It is well known that it is not possible to take two strictly identical biometric fingerprints of the same individual at different times. Firstly because it is very difficult to position, in exactly the same manner but at different times, the same measuring instrument adapted to raise said biometric fingerprint. Secondly, that the environment (temperature, humidity, etc.) and the general state of health (stress, skin disease, etc.) of the individual at the moment when the impression is taken can disturb the result of the survey. .

However, with known hashing functions, starting from two slightly different initial data, the corresponding masked data are very different and decorrelated, so that it is not possible, by comparing them, to deduce whether the initial data are identical to a few errors or not. It is therefore not possible to use known hash functions to hide a biometric fingerprint.

The document US 6,697,947 B1 (JR MATYAS, STEPHEN, MICHAEL ET AL, February 24, 2004) ) describes a biometric data masking scheme that uses a low-sensitivity function based on a modular exponentiation operation.

A first object of the invention is to propose a masking method using a new hash function, better adapted to hide biometric fingerprints than known hash functions. In a preferred mode of implementation, the masking method according to the invention is used to secure a biometric fingerprint.

Finally, a second object of the invention is a use of the masking method of the invention for securing a security document such as for example a bank check.

où p est un nombre premier, bi est le bit de rang i de la donnée claire b, qi est le nombre premier de rang i d'un ensemble de nombres premiers (q₁, ..., q_n) . De préférence, p est un nombre premier de grande taille et les éléments de l'ensemble de nombres premiers sont de petite taille.The first object of the invention is achieved by a method of masking a clear data b of n bits, characterized in that a masked data m is produced by using the following hash function: $\mathrm{m}\mathrm{=}\underset{\mathrm{i}\mathrm{=}\mathrm{1}}{\overset{\mathrm{not}}{\mathrm{\Pi }}}{{\mathrm{q}}_{\mathrm{i}}}^{{\mathrm{b}}_{\mathrm{i}}}\mathrm{mod\; p}\mathrm{,}$

where p is a prime number, bi is the bit of rank i of the light data b, qi is the prime number of rank i of a set of prime numbers (q ₁ , ..., q _n ). Preferably, p is a prime number of large size and the elements of the set of prime numbers are small.

a pour principal intérêt d'être tolérante aux erreurs, comme on le verra mieux par la suite, elle est de ce fait particulièrement bien adaptée pour masquer des données biométriques.Compared to known hash functions, the function $\mathrm{m}\mathrm{=}\underset{\mathrm{i}\mathrm{=}\mathrm{1}}{\overset{\mathrm{not}}{\mathrm{\Pi }}}{{\mathrm{q}}_{\mathrm{i}}}^{{\mathrm{b}}_{\mathrm{i}}}\mathrm{mod\; p}$

Its main interest is to be error tolerant, as will be seen later, so it is particularly well suited to hide biometric data.

• b_i = 1 si le rang i correspond à une minutie réelle et
• b_i = 0 si le rang i correspond à une fausse minutie.

et on applique la fonction de hachage selon l'invention à cette donnée mélangée pour produire une donnée masquée.In a preferred embodiment, the above masking method is applied to a biometric imprint. For this purpose, during the process, a set of actual minutiae characteristics of the said imprint are determined, the actual minutiae are mixed and stored with t false minutiae, a mixed data b of n = s + t bits is formed such that , for everything i:

• b _i = 1 if the rank i corresponds to a real minutia and
• b _i = 0 if the rank i corresponds to a false minutia.

and applying the hash function according to the invention to this mixed data to produce masked data.

Preferably, the actual minutiae and the false minutiae are mixed randomly.

• on mémorise la dite donnée de référence sur ou dans le document de sécurité, ou
• on associe à la dite donnée de référence un code - barre que l'on mémorise sur ou dans le document de sécurité, la donnée de référence et le code - barre étant également mémorisés dans une table.

The second subject of the invention relates to a method for securing a security document, for example a bank check, in which, after obtaining a reference data by masking a biometric fingerprint according to a process as described above,

• storing said reference data on or in the security document, or
• the reference datum is associated with a bar code which is stored on or in the security document, the reference datum and the barcode being also stored in a table.

Preferred examples of implementation of the invention are described below.

We will first detail the masking process according to the invention. To mask a clear data b = (b _n , ..., b ₁ ) of n bits, we use the following hash function: $$\mathrm{m}\mathrm{=}\underset{\mathrm{i}\mathrm{=}\mathrm{1}}{\overset{\mathrm{not}}{\mathrm{\Pi }}}{{\mathrm{q}}_{\mathrm{i}}}^{{\mathrm{b}}_{\mathrm{i}}}\mathrm{mod\; p}$$

The function uses as parameters a set (q _n , ..., q ₁ ) of small prime numbers, for example integers of about 60 bits. The function also uses a parameter p, which is a large integer, for example about 1024 bits. p is preferably chosen such that 2 * q _n ^ 2t <p <4 * q _n ^ 2t, where t is a number of accepted errors.

Contrary to the known hash functions, the function according to the invention is not very sensitive to errors, that is to say that, knowing two data m, μ masked by this function, it is possible to say whether the clear data of origin b, β corresponding are identical, with at most t errors.

Indeed, m, μ are obtained by the relations: $$\mathrm{m}\mathrm{=}\underset{\mathrm{i}\mathrm{=}\mathrm{1}}{\overset{\mathrm{not}}{\mathrm{\Pi }}}{{\mathrm{q}}_{\mathrm{i}}}^{{\mathrm{b}}_{\mathrm{i}}}\mathrm{mod\; p\; and}\mathrm{\mu }\mathrm{=}\underset{\mathrm{i}\mathrm{=}\mathrm{1}}{\overset{\mathrm{not}}{\mathrm{\Pi }}}{{\mathrm{q}}_{\mathrm{i}}}^{{\mathrm{\beta }}_{\mathrm{i}}}\mathrm{mod\; p},$$

avec $$a\mathrm{=}\underset{i\in {\mathrm{\Delta }}_i}{\mathrm{\Pi }}{q}_i\mathrm{mod}p\mathrm{et\; \alpha }=\underset{i\in {\mathrm{\Gamma }}_i}{\mathrm{\Pi }}{q}_i\mathrm{mod}p$$

où Δ_i est l'ensemble des indices i compris entre 1 et n pour lesquels b_i = 1 et β_i = 0, et où Γ_i est l'ensemble des indices i compris entre 1 et n pour lesquels b_i = 0 et β_i = 1. La somme des tailles des ensembles Δ_i et Γ_i est au plus égale à t, t étant le nombre de bits de β différents des bits de b de même rang, correspondant au nombre maximum d'erreurs acceptées.We further define: $$\mathrm{\lambda }\mathrm{=}\frac{\mathrm{m}}{\mathrm{\mu }}\mathrm{mod}\left(\mathrm{p}\right)\mathrm{=}\frac{\mathrm{at}}{\mathrm{\alpha }}\mathrm{mod\; p}$$

with $$\mathrm{at}\mathrm{=}\underset{i\in {\mathrm{\Delta }}_i}{\mathrm{\Pi }}{q}_i\mathrm{mod}p\mathrm{and\; \alpha }=\underset{i\in {\mathrm{\Gamma }}_i}{\mathrm{\Pi }}{q}_i\mathrm{mod}p$$

where Δ _i is the set of indices i between 1 and n for which b _i = 1 and β _i = 0, and where Γ _i is the set of indices i between 1 and n for which b _i = 0 and β _i = 1. The sum of the sizes of the sets Δ _i and Γ _i is at most equal to t, t being the number of bits of β different from the bits of b of the same rank, corresponding to the maximum number of accepted errors.

a and α, which are products of small prime numbers q _i , are also small numbers, which further satisfy the relation: a * λ = α mod p. From this last equality and the number λ, it is then possible to find the numbers a and α. A decomposition of a and α into prime numbers finally makes it possible to factor a and α. Decomposition is facilitated by taking advantage of the fact that a and α decompose in principle into small prime numbers. If a and α decompose on the set (q _n , ..., q ₁ ), then we deduce that the original data b and β are identical, with at most t errors.

We will now describe a preferred embodiment of the masking method using the masking function described above, for masking a biometric fingerprint.

In the example below, the physical biometric fingerprint that is to be masked is a fingerprint characterized by a predefined number of actual minutiae. A real minutia is a detail of a footprint at a given point in the physical footprint, such as a break in a line, a fork on a line, and so on. Numerically, a minutia can be translated by a string of characters including information on the position and the form of the minutia.

According to the invention, in order to mask the physical footprint, a set of t false minutiae, also defined by a string of characters but which do not correspond to a real minutia of the physical footprint. In one example, a false minutia is completely randomly defined, and a set of t = 80 false minutiae is added to a set of s = 20 real minutiae.

• b_i = 1 si le rang i correspond à une minutie réelle et
• b_i = 0 si le rang i correspond à une fausse minutie.

The order of the real minutiae and the false minutiae is mixed, for example in a random manner, then a mixed datum b = (b _n , ..., b ₁ ) of n = s + t bits is created such that, for all i:

• b _i = 1 if the rank i corresponds to a real minutia and
• b _i = 0 if the rank i corresponds to a false minutia.

The mixed data item b is then masked by the masking method according to the invention to produce a masked data item m such that: $$\mathrm{m}\mathrm{=}\underset{\mathrm{i}\mathrm{=}\mathrm{1}}{\overset{\mathrm{not}}{\mathrm{\Pi }}}{{\mathrm{q}}_{\mathrm{i}}}^{{\mathrm{b}}_{\mathrm{i}}}\mathrm{mod\; p}$$

The masked data m can then be stored in a database, on an identity card, in a memory of a smart card, etc. The masked data m can be used as reference data, for example to verify the identity of a person, as follows.

• de relever une nouvelle empreinte biométrique physique sur la personne puis de calculer l'ensemble de s minuties réelles correspondant,
• d'ajouter t fausses minuties, de mélanger les fausses minuties et les vraies minuties, de déterminer la donnée mélangée β associée à la nouvelle empreinte biométrique, puis de masquer β par la fonction $\mathrm{\mu }\mathrm{=}\underset{\mathrm{i}\mathrm{=}\mathrm{1}}{\overset{\mathrm{n}}{\mathrm{\Pi }}}{{\mathrm{q}}_{\mathrm{i}}}^{{\mathrm{\beta }}_{\mathrm{i}}}\mathrm{mod\; p}$
  
  pour obtenir une nouvelle donnée masquée µ,
• de déterminer s'il y a concordance entre la donnée référence m précédemment mémorisée et la donnée µ masquée obtenue à partir de la nouvelle empreinte réelle qui vient d'être relevée.

To verify the identity of a person, simply:

• to take a new physical biometric imprint on the person and then calculate the set of corresponding real minutiae,
• to add t false minutiae, to mix the false minutiae and the real minutiae, to determine the mixed data β associated with the new biometric imprint, then to mask β by the function $\mathrm{\mu }\mathrm{=}\underset{\mathrm{i}\mathrm{=}\mathrm{1}}{\overset{\mathrm{not}}{\mathrm{\Pi }}}{{\mathrm{q}}_{\mathrm{i}}}^{{\mathrm{\beta }}_{\mathrm{i}}}\mathrm{mod\; p}$
  
  to obtain a new hidden data μ,
• to determine whether there is a match between the previously memorized reference datum m and the masked μ datum obtained from the new real fingerprint that has just been read.

• on calcule $\mathrm{\lambda }\mathrm{=}\frac{\mathrm{m}}{\mathrm{\mu }}\mathrm{mod\; p}\mathrm{=}\frac{\mathrm{a}}{\mathrm{\alpha }}\mathrm{mod\; p},$
  
  puis a et α à partir de la relation a*λ = α mod p, avec a et α petits devant l'entier p, par l'algorithme des fractions continues par exemple.
• on décompose ensuite a et α en facteurs premiers, puis
  • il y a concordance si a et α se décomposent sur au plus t éléments de l'ensemble des nombres premiers (q_n , ..., q_l),
  • il n'y a pas concordance sinon

To determine if there is a match between m and μ:

• we calculate $\mathrm{\lambda }\mathrm{=}\frac{\mathrm{m}}{\mathrm{\mu }}\mathrm{mod\; p}\mathrm{=}\frac{\mathrm{at}}{\mathrm{\alpha }}\mathrm{mod\; p},$
  
  then a and α from the relation a * λ = α mod p, with a and α small in front of the integer p, by the algorithm of continuous fractions for example.
• we then break down a and α in prime factors, then
  • there is a concordance if a and α are decomposed on at most t elements of the set of prime numbers (q _n , ..., q _l ),
  • there is no concordance otherwise

An intended application of the masking method according to the invention is to secure a security document such as for example a bank check. For this, according to the invention, a biometric fingerprint of the owner of the security document is masked by a masking method as described above, to produce a reference data.

According to a first embodiment, the reference data is stored on or in the security document, for example by printing.

According to a second embodiment, the reference datum is associated with a barcode, the associated datum / barcode pair is stored in a database, and the barcode is stored, for example by printing, on the security document.

It then suffices, when the security document is transmitted for example, to simultaneously take with the security document a biometric fingerprint of the person who transmits the document and then check that the biometric fingerprint of the person who transmits the document corresponds well. the imprint included in the reference data stored in the document or associated with the bar code stored on the document.

In the first embodiment, the verification can be done by any person, the data of reference being stored directly on the document. In the second embodiment, the verification may be done by anyone having access to the database, and who is not necessarily the person who receives the document.

The bar code is produced according to known techniques, for example a one-dimensional barcode consisting of a series of vertical bars of varying thickness and spacing may be used. The choice of the form of the bar code is in practice a function of the number of reference data to be memorized, each reference datum corresponding to different persons.

Preferably, the database in which the reference datum pairs / associated bar code are stored is accessible for verification only to a limited number of people, depending on the desired level of security: the access can for example be authorized for any person to receive security documents or, to a lesser extent, only to a certifying authority.

In a practical example, the security document is a bank check and the fingerprint of its owner is stored on the check in the form of a bar code. A merchant has a device for reading and masking a fingerprint with means for reading a fingerprint, hide it and then print the associated masked data. The bank issuing the check only has the right of access to the database in which are stored the masked reference data (corresponding to the initial masked print) and the associated bar code; this access allows him to verify that the impression left by the person who presented the check to the merchant, and that the latter has hidden and printed on the check, corresponds to that of the owner of the check.

Claims (7)

1. Method of masking a plain datum b having n bits, characterised in that a masked datum m is produced using the following hash function: $\mathit{m}\mathit{=}\underset{\mathit{i}\mathit{=}1}{\overset{\mathit{n}}{\mathrm{\Pi }}}{{\mathit{q}}_{\mathit{i}}}^{{\mathit{b}}_{\mathit{i}}}\mathrm{mod}\mathit{p},$

   

   where p is a prime number, b_i is the bit at position i of plain datum b, and q_i is the prime number at position i in a set of prime numbers (q_i...,q_n) .
2. Masking method according to claim 1, in which p is a large prime number and the components of the set of prime numbers are small.
3. Masking method according to either one of the claims 1 or 2, applied to a biometric print, characterised in that it consists in determining a set of s real minutiae, which are characteristic of said print, mixing and arranging the real minutiae with t false minutiae, and forming a mixed biometric datum b having n = s + t bits, such that, for any i:

   b_i = 1 if position i corresponds to a real minutia and

   b_i = 0 if position i corresponds to a false minutia

   and the hash function according to the invention is applied to this mixed datum in order to produce a masked datum.
4. Method according to claim 3, in which the real minutiae and the false minutiae are mixed in a random fashion.
5. Method of securing a security document, for example a bank cheque, during which, after having obtained a reference datum by masking a biometric print according to a method according to claim 3 and 4,

   - said reference datum is stored on or in the security document, or

   - a barcode is associated with said reference datum, which is stored on or in the security document, the reference datum and the barcode also being stored in a table.
6. Method of verifying a security document secured by a method according to claim 5, verification method during which;

   - a physical biometric print of a person presenting the security document is digitised,

   - the digitised print is masked using a method according to one of the claims from 3 to 4, to produce a masked datum,

   - the masked datum is compared with a reference datum, and then

   - the security document is accepted if the masked datum and the reference datum are identical with a predefined rate of error, and the document is refused otherwise.
7. Method according to claim 6, in which, during the comparison step, if a barcode associated with the reference datum is stored on the security document, then:

   - the barcode is read and the reference datum associated with the barcode is searched in a table, and then

   - the reference datum is compared with the masked datum.