Fingerprintbased cryptobiometric system for network security
 Subhas Barman^{1},
 Debasis Samanta^{2}Email author and
 Samiran Chattopadhyay^{3}
https://doi.org/10.1186/s1363501500201
© Barman et al.; licensee Springer. 2015
Received: 15 May 2014
Accepted: 11 March 2015
Published: 3 April 2015
Abstract
To ensure the secure transmission of data, cryptography is treated as the most effective solution. Cryptographic key is an important entity in this process. In general, randomly generated cryptographic key (of 256 bits) is difficult to remember. However, such a key needs to be stored in a protected place or transported through a shared communication line which, in fact, poses another threat to security. As an alternative to this, researchers advocate the generation of cryptographic key using the biometric traits of both sender and receiver during the sessions of communication, thus avoiding key storing and at the same time without compromising the strength in security. Nevertheless, the biometricbased cryptographic key generation has some difficulties: privacy of biometrics, sharing of biometric data between both communicating parties (i.e., sender and receiver), and generating revocable key from irrevocable biometric. This work addresses the abovementioned concerns. We propose an approach to generate cryptographic key from cancelable fingerprint template of both communicating parties. Cancelable fingerprint templates of both sender and receiver are securely transmitted to each other using a keybased steganography. Both templates are combined with concatenation based feature level fusion technique and generate a combined template. Elements of combined template are shuffled using shuffle key and hash of the shuffled template generates a unique session key. In this approach, revocable key for symmetric cryptography is generated from irrevocable fingerprint and privacy of the fingerprints is protected by the cancelable transformation of fingerprint template. Our experimental results show that minimum, average, and maximum Hamming distances between genuine key and impostor’s key are 80, 128, and 168 bits, respectively, with 256bit cryptographic key. This fingerprintbased cryptographic key can be applied in symmetric cryptography where session based unique key is required.
Keywords
1 Introduction
Information security and a secure transmission of data become very important in information and communication technology. A third party can trap data or steal important data stored in a computer. To prevent this, it is advocated to encrypt the messages to provide information security. This type of protection is usually provided using cryptography. In cryptography, a key (K _{1}) is used to encrypt a message (called plaintext P) with encryption algorithm (E) into ciphertext (C). The ciphertext is converted into plaintext using a key (K _{2}) and decryption algorithm (D). There are two types of cryptography: symmetric cryptography and asymmetric cryptography. In symmetric cryptography (e.g., Data Encryption Standard (DES) [1], Advanced Encryption Standard (AES) [2]), same key (i.e., K _{1}=K _{2}=K) is used in encryption (C=E _{ K }(P)) and decryption algorithm (P=D _{ K }(C)). In asymmetric cryptography (e.g., RivestShamirAdleman (RSA) algorithm [1]), two different keys are used (i.e., K _{1}≠K _{2}), the public key is used to encrypt a message (\(C=E_{K_{1}}(P)\)), and private key is used to decrypt the ciphertext into plaintext (\(P=D_{K_{2}}(C)\)) [1].
Strength of cryptography in respect to security depends on the strength of keys used in encryption and decryption algorithms. A key is said to be strong if it is not easily guessed and not feasible to break within a real time. So, the issue that arises is the selection of cryptographic key. If the key is simple or very short in length, then for the attacker it is easy to guess the key. If the key is very long (128, 192, or 256 bits, for example, in AES algorithm [2]), then it is very difficult to memorize the key by a user. As a consequence, user should store it in a smart card or hardware token which can be misplaced or stolen out by an attacker. Moreover, the token or smart card is protected by passwordbased authentication mechanism to control the access of cryptographic key. Nevertheless, password can be forgotten or guessed by social engineering [3] and dictionary attack [4]. Both knowledgebased (e.g., password) and possession based (e.g., token) authentication systems are unable to assure nonrepudiation property in traditional cryptography.
Of late, biometric is being integrated with cryptography (called cryptobiometric system) to alleviate the limitations of the abovementioned systems [5,6]. Biometric is the unique measure of the identity of individuals with their behavioral and physiological traits like face, fingerprint, iris, retina, palmprint, speech, etc. [7]. Many researchers are trying to use biometric traits in the authentication component of cryptography to remove the requirement of passwordbased authentication. The integration of biometric with cryptography, deals with either cryptographic key release [810] or cryptographic key generation [6,1117], is promising in many aspects. As biometric is directly linked with the owner, it removes the problem to memorize the cryptographic key and confirms the nonrepudiation of users.
Cryptobiometric system, however, has some issues. Any biometric system needs to provide biometric template protection which confirms the privacy and security of biometric data [18]. The biometric data used in a biometric system should not leak any information about the biometric features. It is also required to provide revocability to the irrevocable biometric data. In passwordbased authentication systems or tokenbased authentication systems, passwords or tokens are easy to change while it is compromised. But, biometric traits are inherent and fixed forever, that is, the biometric data is irrevocable [7]. The owner of biometric traits is not able to revoke her biometric when it is compromised. As a result, the biometric data become useless forever [6]. To overcome this problem, it demands a cancelable transformation [18,19] of biometric template to provide revocability to the irrevocable biometric. Simultaneously, it would ensure the privacy of biometric data [5], so that the transformed template does not leak any information about the original template. Moreover, biometric data is required to be transmitted over nonsecure communication channels for remote use. Therefore, there is a need to generate cryptographic key, which is revocable and noninvertible from the biometrics of two different users without compromising the privacy and security of the biometrics involved in key generation process.
This work aims to address the abovementioned concerns and proposes a solution to develop a cryptobiometric system. Our proposed solution includes the following: 1) how to generate cancelable fingerprint template so that biometric features of neither communicators are never disclosed to anyone, 2) how to generate a unique cryptographic key for encryption (decryption) of messages using the cancelable fingerprint templates of both sender and receiver, and 3) how to generate revocable session key from irrevocable biometric traits prior to each session. In this paper, we propose an approach to generate, share, and update cryptographic key for symmetric cryptography from the fingerprints of sender and receiver at their sites for encryption and decryption, respectively. Initially, sender shares two secret keys namely stego key (K _{ g }) and shuffle key (K _{shuf}) with receiver. Stego key is generated from a password (pwd) by sender and receiver using pseudo random number generator (PRNG) [1]. Shuffle key (K _{shuf}) is generated randomly, which is a binary stream of bits and stored in token. In this work, sender shares K _{shuf} and pwd with receiver using public key cryptography. With our proposed approach, asymmetric cryptography is proposed to exchange an initial shuffle key K _{shuf} and a password pwd between sender and receiver. For session keys, we propose biometricbased cryptographic key generation to establish a link of users biometric with cryptographic key. In our approach, biometrics of both communicating parties are integrated to generate cryptographic keys so that we can avoid the complex random number generation and alleviate the issue of storing the random cryptographic keys in the custody of sender and receiver. Moreover, revocable key generation in every session and protecting the privacy of biometric templates are the challenge which has been addressed in this work. Both sender and receiver exchange their cancelable fingerprint template with each other using keybased steganography. Both cancelable templates are then merged together using concatenationbased feature level fusion technique [20] to generate a combined template. Shuffle key is used to randomize the elements of the combined template. Finally, cryptographic key is generated from this shuffled template using a hash function. In our approach, fingerprint identity of sender is not disclosed to receiver and vice versa as cancelable template is exchanged between them to derive cryptographic key. Moreover, in our approach, cryptographic key or fingerprint template or both can be revoked easily if required. The revocability is provided to the cryptographic key with cancelable template and or with updated shuffle key.
The rest of the paper is organized as follows. A brief review of related research is given in Section 2. The proposed approach for cryptographic key generation from the fingerprints of sender and receiver is given in Section 3. The experimental results and security analysis are discussed in Sections 4 and 5, respectively. Finally, the paper is concluded in Section 6.
2 Literature survey
Our work consists of mainly three subtasks: i) transformation of biometric template, ii) secure transmission of biometric data, and iii) cryptobiometric system. There exist few work in the literature related to each subtask, which are discussed in this section.
2.1 Biometric template transformation
Biometric systems require a transformation of biometric template to ensure privacy, security, and revocability of biometric data. The technique which can meet this requirement is called cancelable or revocable biometric. This privacy enhancement problem is identified, and conceptual frameworks of biometric templates are presented in [21,22]. Ratha et al. [23] formally defined the problem of cancelable biometric. Uludag et al. [6] provides a comprehensive review on privacy and revocability of biometrics with some corrective measures. Recently, Ratha et al. [19] proposed three practical solutions to cancelable biometrics and generate cancelable fingerprint templates. These three template transformation approaches are Cartesian, polar, and functional transformations on feature domain. In Cartesian transformation approach, the minutiae space is divided into rectangular cells which are numbered with sequence. A userspecific transformation key (i.e., matrix) is used to shift the cells to a new location, and the minutiae points are relocated to the new cells. In polar transformation, coordinate space is divided into polar sectors that are numbered in sequence. The sector position is changed with the help of translation key, and it changes the minutiae location also. In functional transformation, Ratha et al. [19] model the translation using a vectorvalued function \(\overrightarrow {F}\)(x,y) which is an electric potential field parameterized by a random distribution of charges. The phase angle of the resulting vector decides the direction of translation and the magnitude \(\overrightarrow {F}\) of this vector function parameterizes the extent of movement. In an alternate formulation, Ratha et al. [19] use the gradient of a mixture of Gaussian kernels to determine the direction of movement and the extent of movement is determined by the scaled value of the mixture. Some researchers proposed shufflingbased transformation to generate cancelable templates using a userspecified random key [24,25]. In these works, the iris code is divided into blocks and then the blocks are shuffled with a userspecified random shuffling key to generate cancelable iris template.
Jain et al. [18] reviewed the existing work of fingerprint template protection such as encryption, template transformation, and cryptobiometric systems. They analyzed the practical issues involved in applying these techniques for biometric template protection. They compared the existing solutions of template protection on the basis of template security and matching accuracy of biometrics in transformed domain.
2.2 Biometric data transmission
There are many work reported in the current literature where the biometric data is transmitted over communication channels for the purpose of remote authentication. Existing work [2628] consider hiding of biometric data within another media called cover media using data hiding technique. Different types of data hiding techniques are used for secure transmission of biometric data using steganography. Minutiae points of fingerprint are hidden within face or synthetic fingerprint using watermarking technique and sent to other user via insecure communication channel [26]. Similarly, fingerprint is also used as cover media to hide other biometric data (i.e., face) in watermarking technology and used as carrier image for secure transmission of biometric data [27]. Note that in the data hiding concept, use of real biometric as cover media is risky as it reveals sender’s biometric identity to receiver. Agrawal and Savvides [29] propose a biometric data hiding approach where a biometric (iris and fingerprint) data is encrypted with a key and the encrypted biometric data is encoded with error correcting code. The encoded biometric data is embedded bit by bit using the sign of discrete cosine transform (DCT) coefficients of a random cover image.
2.3 Cryptobiometric systems
Biometricbased cryptosystems are classified into two types, namely key release and key generation. In the first approach, a randomly created cryptographic key is protected from unauthorized access with users’ biometric data. Fuzzy vault [810] and fuzzy commitment scheme [24,25,30] fall under this category. In fuzzy vault scheme, biometric data (e.g., minutiae points) is considered as an unordered set s ^{ E }=x _{1},x _{2},…x _{ r } of r elements. The secret key (k _{ v }) of k bits is transformed into a polynomial of degree k. All elements of s ^{ E } are evaluated on the polynomial and the polynomial evaluation value P(x _{ i }) and x _{ i }, that is, (x _{ i },P(x _{ i })) points are secured with some randomly generated chaff points (a _{ j },b _{ j })^{ q } _{ j=1} which do not lie on the polynomial P (i.e., b _{ j }≠P(a _{ j }) and a _{ j }∉s ^{ E }, ∀j=1,2,…,q). The genuine ((x _{ i },P(x _{ i }))) and chaff (a _{ j },b _{ j }) points constitute the fuzzy vault. The security of this vault depends on the computational difficulties of solving polynomial reconstruction problem. Now, the secret key is released only when the query biometric is close to the set s ^{ E }. Most of the existing fingerprintbased fuzzy vault use the (x,y) coordinate values of minutiae points [9,10] whereas, Nandakumar et al. [8] propose a fingerprintbased fuzzy vault where both (x,y) coordinates and orientation (θ) of minutiae points are used. On the other side, in fuzzy commitment scheme, biometric data is represented in a binary vector b ^{ E } and the vector is locked by a random secret key of less or equal bits of b ^{ E } with XOR operation. In [24,25], iris code is combined with a random key using XOR operation and using the query iris code, the secret key is extracted from the combined iris code.
Few approaches have been proposed to generate cryptographic key from the biometric traits [1416,31,32]. Monrose et al. [14] propose an approach to generate cryptographic key from user’s voice while speaking a passphrase. Feng et al. [15] propose a cryptosystem, that is, BioPKI, where user’s online signature is used to generate a private key. In [16,32], face biometric is used to extract a suitable length cryptographic key. Iris is also used for cryptographic key generation from iris texture [12,13,31]. Rathgeb et al. [31] analyze the iris feature vector and detected the most stable or reliable bits in the binary iris code to construct cryptographic key. Fingerprint, the most universal and acceptable biometric, is also used to derive a cryptographic key from cancelable fingerprint template [17,33,34]. Main problem of the approaches [17,33,34] is that it is not able to generate revocable key for session based communication.
In recent research, multimodal or multiple biometrics are used in cryptobiometric systems [1113]. A Jagadeesan et al. proposed a method [12] where multimodal biometrics (fingerprint and iris) are used. They applied the feature level fusion of minutiae points and texture properties of iris to generate the multimodal biometric templates and the key is generated from this template. In another work [13], Jagadeesan et al. use the same biometrics (fingerprint and iris) but different method to generate the transformed template. The exponentiation operation is performed where iris texture values are used as base numbers and minutiae coordinates are used as exponent. Then, the next prime number is calculated for each exponentiation result and multimodal template is generated using multiplication of two resultants prime numbers to generate a key of 256 bits. This approach is, however, not free from key sharing problem of traditional symmetric cryptography.
Dutta et al. reported a method of fingerprintbased cryptography and network security [11]. In this method, they work with the fingerprints of sender and receiver. The fingerprint of receiver is transmitted to sender, and it is merged with sender’s fingerprint to generate cryptographic key (of 128 bits) using standard hash function (MD5). In their approach, cryptographic key along with a random vector are watermarked into the genuine fingerprint and watermarked image is sent to the recipient. This method is not secure as genuine biometric is used as the cover image for data hiding. As the key and random sequence vector are transmitted over insecure channel with data hiding technique, it causes security threats to the message transmission if the fingerprint of the user is compromised to a third party, anyway. In this approach, fingerprint of receiver is sent to sender and fingerprint of sender is used as cover image and the watermarked image containing the master key and random vector is sent to receiver. Thus, fingerprint of sender is known to receiver and vice versa. A third party (maninmiddle) can generate the key using cryptographic hash function and with the knowledge of fingerprints of sender and receiver. Further, this approach is silent about the revocability issue.
3 Proposed methodology
3.1 Feature extraction from fingerprint image

F _{ S } = Set of minutiae points extracted from sender’s fingerprint.
\(= \left [{m^{s}_{1}}, {m^{s}_{2}},\ldots,m^{s}_{N_{s}}\right ]\); where \({m^{s}_{i}} = (x_{i},y_{i})\), \({m^{s}_{i}}\) is the ith minutiae points of sender’s fingerprint, i= 1 to N _{ s } and N _{ s } is the size of F _{ S }

F _{ R } = Set of minutiae points extracted from receiver’s fingerprint.
\(= \left [{m^{r}_{1}},{m^{r}_{2}},\ldots,m^{r}_{N_{r}}\right ]\); where \({m^{r}_{i}} = (x_{i},y_{i})\), \({m^{r}_{i}}\) is the ith minutiae points of receiver’s fingerprint, i= 1 to N _{ r } and N _{ r } is the size of F _{ R }.
Note that, for all fingerprint images, it is a general observation that number of minutiae points for a person lies within 50. However, in case, if there are more than 50 minutiae points (i.e., N _{ s },N _{ r }≥50), then first 50 minutiae points according to their quality value would be selected and the rest be discarded.
3.2 Cancelable template generation
 1.The coordinate system is divided into N cells of same size h×w, where h is the height and w is the width of the cells. The total number of cells N can be calculated with equation given below$$ N = {(H \times W) \over (h \times w)} $$(1)
where H and W are the height and width of the fingerprint image.
 2.Each cell is denoted by C _{ i,j } where i = 1 to n (n cells in each row) and j = 1 to m (m cells in each column). The cells can be represented in a onedimensional vector, and cell C _{ i,j } can be represented by c _{ t } and the value of t can be computed in the following way.$$ t = \{(j  1) \times n\} + i $$(2)
where for any value of i (1≤i≤n) and j (1≤j≤m), there will be a unique t such as 1≤t≤N _{cell}. Each cell contains either no minutiae or a set of minutiae points. It depends on the cell size, distribution of minutiae points in fingerprint image. For sender’s fingerprint, if the tth cell c _{ t } contains n _{mt} number of minutiae then the ith minutiae point of tth cell of senders fingerprint can be represented as \(m^{\text {st}}_{i}\) and the value of i is defined as 1≤i≤n _{mt}.
For an illustration, we divide the image space into n vertical cells and m horizontal cells, and the cells are shown in Figure 2 where the twodimensional cells are shown in left and the sequence number is shown in right side table.  3.Generate a userspecific (0,1) matrix (M) of size N×N (where N=n×m) to map the cells with their new positions as per the following equation.$$ C'_{i'j'} = C_{i,j} M $$(3)
where C _{ i,j } is the original cell which is replaced by cell C i ^{′},j ^{′}′. Here, one cell can replace zero or multiple cells. The number of minutiae points (say n _{mt}) in the new cell can be the same or different.
 4.Modify the (x,y) coordinates according to their new cell locations, and cancelable template T _{ C } is generated as follows.$$ T_{C} = F \left(C_{i,j},C'_{i',j'}\right) $$(4)
The (x,y) coordinate values of all minutiae points belong to C i ^{′},j ^{′}′ will be placed to the cell C _{ i,j } according to the mapping function (F).
For the replacement of cell C _{ i,j } by C i ^{′},j ^{′}′ (i.e., C _{ i,j }←C i ^{′},j ^{′}′), the (x,y) coordinate value of minutiae point of cell C i ^{′},j ^{′}′ can be computed for new location as follows. (a)
If i ^{′}=i then \({x^{d}_{i}}=x_{i}\)
 (b)
If i ^{′}>i then \({x^{d}_{i}}=x_{i} + (i'i)*w\)
 (c)
If i ^{′}<i then \({x^{d}_{i}}=x_{i}  (ii')*w\)
where x _{ i } is the x coordinate value of a minutiae points and \({x^{d}_{i}}\) is the displaced x coordinate value of the same minutiae points and w is the width of a cell of size h×w. Similarly, depending on the value of j ^{′}, the y _{ j } will be changed to \(y_{j}^{d}\).
 (a)
After transformation of fingerprint template using this transformation, the transformed template (T _{ C }), also known as cancelable template, contains modified minutiae points denoted by \(m^{s'}_{i}\), which represents ith modified minutiae point or ith elements of the cancelable template. In the cancelable template, we consider that it contains 50 modified minutiae points. If it exceeds 50, then we consider only first 50 elements and if it contains less than 50 elements, then we augment sufficient numbers of zero elements at the end to make it of intended size. In the subsequent discussion, the cancelable templates of sender and receiver are denoted by T _{CS} and T _{CR}, respectively, where T _{CS} = \(\left \{{m^{s'}_{1}, m^{s'}_{2}, \ldots, m^{s'}_{N_{\text {TCS}}}}  N_{\text {TCS}} = T_{\text {CS}}\right \}\) and \(T_{\text {CR}} = \left \{ m^{r'}_{1}, m^{r'}_{2}, \ldots, m^{r'}_{N_{\text {TCR}}} N_{\text {TCR}} = T_{\text {CR}}\right \}\).
3.3 Steganographic encoding
In this work, cancelable template of one party (say sender) needs to be sent through a shared communication channel to other party (say receiver) and vice versa. Sender (and receiver) uses steganographybased data hiding technique to hide the cancelable template data into a cover image (I) (of size M _{ I } pixels, say). The cancelable template is converted into binary stream (s _{1},s _{2},s _{3},…s _{ L }), where L is the number of bits in cancelable template after the conversion, which is to be hidden into cover image using LSB steganography [35]. A secret key, called stego key K _{ g }, is generated from a password (pwd) using pseudo random number generator where the password is used as the seed value. The stego key (\(\phantom {\dot {i}\!}K_{g}=\;[k_{1}, k_{2}, \ldots k_{N_{\text {kg}}}]\); where L≤N _{kg}≤M _{ I }] is used to select the pseudo random path of pixel locations in cover image (I) to hide the cancelable template bit by bit in cover image. The stego image (I _{stego}) is sent to the recipient from sender and vice versa.
3.4 Steganographic decoding
In this phase, hidden data are extracted from the stego image (I _{stego}) using the decoding function. The stego key (K _{ g }) is used to locate the pixels where data embedding take place. The extracted binary stream is then used to reconstruct the cancelable template.
3.5 Merging cancelable templates T _{ CS } and T _{ CR }
After receiving the cancelable template of the counter partner, receiving party merges his own cancelable template with the received cancelable template. Say, sender has its own cancelable template T _{CS} and received the cancelable template T _{CR} from receiver. Both T _{CS} and T _{CR} consist of modified minutiae points (\(m^{s'}_{i}\), \(m^{r'}_{i}\)) of sender and receiver, respectively. These two cancelable templates are fused using the feature level fusion [20,36] of modified minutiae features.
This way, sender and receiver both derive the same secret key which establishes a secure communication between the sender and receiver for a session. For a new session, a new session key can be generated from the same fingerprints using an updated shuffle key, which is discussed in the next subsection.
3.6 Shuffle key update
 1.
Both sender and receiver share their cancelable fingerprint data (Sections 3.3 and 3.4) and generate F _{code} following the method discussed in Section 3.5.
 2.
Shuffle key (K _{shuf}) is XORed with the first K _{shuf} bits of F _{code} to obtain a new shuffle key (K _{SS}), that is, K _{SS}=K _{shuf}⊕F _{code}.
 3.
Sender computes the hash value of the new shuffle key (h(K _{SS})) and sends it along with update request to receiver. We have used XORbased hash function (h), as discussed in Section 3.5.
 4.
Similarly, receiver also generates a new shuffle key (K _{SR}=K _{shuf}⊕F _{code}) and computes the hash of the new shuffle key (h(K _{SR})) using the same one way hash function (h) which is used in sender side.
 5.
Receiver compares the computed hash (h(K _{SR})) with received hash (h(K _{SS})), if both are same, then receiver replaces old shuffle key (K _{shuf}) with new one (K _{SR}) and sends success message to the sender.
 6.
On the basis of receiver’s report, sender also replaces the old shuffle key (K _{shuf}) with new shuffle key (K _{SS}).
In this way, both sender and receiver are able to update their shuffle key. In every session, a unique shuffle key is generated from the fingerprint data of sender and receiver. After the session is over, old shuffle key is destroyed and modified one is used for next session of communication. In this way, our approach can provide the diversity of cryptographic key of fingerprintbased symmetric cryptography.
4 Experiment and experimental results
There are mainly two objectives in our experiment. First, we investigate the impact of data encoding into cover images (i.e., synthetic fingerprints from DB4 of each database) and accuracy of data decoding from stego images. In the next part of our experiment, we measure the randomness of cryptographic key generated from fingerprints of genuine users with respect to the key generated from impostor’s fingerprints. In this regard, the Hamming distances between the genuine and impostor’s keys are measured and corresponding histograms are plotted. Here, we consider all possible cases of attacks when different entities are compromised and the Hamming distances are computed for each case.
4.1 Database
Fingerprints used in our experiments
FVC2000 (sensors, image size)  FVC2002 (sensors, image size)  FVC2004 (sensors, image size)  

DB1  Optical (KeyTronic), 300×300  Optical (Identix) (CrossMatch V300), 388×374  Optical sensor, 640×480 
DB2  Capacitive (ST Microelectronics), 256×364  Optical (Biometrika), 296×560  Optical Sensor (Digital Persona U.are.U 4000), 328×364 
DB3  Optical (Identicator Technology), 448×478  Capacitive (Precise Biometrics), 300×300  Thermal sweeping sensor (Atmel FingerChip), 300×480 
DB4  SFinGe v2.0, 240×320  SFinGe v2.51, 288×384  SFinGe v3.0, 288×384 
4.2 Experimental setup
A unique pair of fingerprints is taken (as fingerprint of sender and receiver) from a subset (i.e., DB1 or DB2 or DB3 or DB4) of a specific FVC to generate a genuine key. All remaining pairs of fingerprints in the same subset of the same FVC database are taken to generate impostor keys with respect to that genuine key. This way, all pairs of genuine fingerprints and corresponding pairs of impostor fingerprints are chosen. Initially, all unique combination of two person’s fingerprints from 110 person’s fingerprints of each subset of the fingerprint database is computed. Thus, the total number of pairs for genuine users is \(\frac {110}{2} =\) 55 for each subset (i.e., DB1, DB2, DB3, and DB4 for both sets A and B). For each FVC database (considering all four subsets of set (A + B), the total number of genuine keys is 4×55= 220. Similarly, for every genuine key, remaining 54 keys are impostor keys. In our experiment, the duplicate pairs of genuine and impostor fingerprints are avoided carefully. As a result, we get 220×3=660 impostor’s keys per FVC database and a total number of \(3\times 4 \times \frac {55\times 54}{2}=\) 17,820 impostor’s keys.
Now, the minutiae points from fingerprint images are extracted using NBIS software (MINDTCT) [41]. The MINDTCT tool takes fingerprint image as input and returns minutiae points set in the format of (x,y,θ,q) where q is the quality of that minuitae point. Average number of minutiae points is found as 50. We consider only (x,y) coordinates of first 50 minutiae points according to the quality of minutiae reported by MINDTCT in our experiment. The minutiae points are transformed into cancelable template with a userspecified transformation key (Section 3.2). We have divided the fingerprint images into 64 cells. The cell size is computed by dividing the height (H) and width (W) with 8 (i.e., cell size =h×w where \(h=\frac {H}{8}\) and \(w=\frac {W}{8}\)). The cancelable template converted into binary stream (Section 3.5). Most of the sizes of fingerprint images are within the range of 256 to 511. The maximum values of x and y coordinate points thus can be represented with 9 bits binary numbers. However, as exceptions, the y coordinate value of the fingerprint images in some FVC2002 database (it is A and B sets in DB2), and the x coordinate values of fingerprint images in FVC2004 database (e.g. A and B sets in DB1) exceed the range of 511. As the x and y coordinate values of minutiae points in almost all fingerprint images are less than 512, we represent them by 9 bits (maximum decimal value with 9 bits is 2^{9}−1=511). Of course, the coordinates exceeding 511 are with a less approximation, which do not affect the results adversely. The values of x and y coordinate of modified minutiae points (i.e., the elements of cancelable templates) also lie within this range. Therefore, each element of the cancelable template is converted into 18 bits binary number (9 bits for x coordinate and 9 bits for y coordinate) and binary conversion of all elements of cancelable template produces a binary bit stream of 18×50= 900 bits.
We propose to generate the transformation keys (S _{ k }, R _{ k }) randomly. In our experiment, these keys are generated using pseudo random number generator (PRNG) available in MATLAB. We have used a unique transformation key to generate a cancelable template from fingerprint of a specific person. In our approach, stego key (K _{ g }) is also generated randomly (using PRNG in MATLAB) from a password (pwd) and a unique stego key is assigned to each cover image of the synthetic fingerprints (i.e., the fingerprints in DB4). A unique shuffle key (pseudo random number) is used for a unique pair of fingerprint images.
We use synthetic fingerprint [40] from DB4 database (of FVC2000, FVC2002, FVC2004) as the cover image to hide cancelable fingerprint template of genuine users using LSB steganography. The cover image is picked up at random by both sender and receiver. A stego key (K _{ g }) is used to locate the pixels of cover image where the cancelable template (C _{TS}, C _{TR}) bits are hidden during steganographic encoding (Section 3.3). Similarly, the same stego key is used to decode the cancelable template bits from stego image during steganographic decoding (Section 3.4).
Note that in our experiment, both sender and receiver exchange their own cancelable template between them using the same stego key (K _{ g }) but different cover images (say I _{ S }, I _{ R }).
4.3 Experimental results
The results of our experiment are stated below with respect to different scenarios.
4.4 Case 1: Impact of steganography
In our experiment, effect of data encoding over cover image is investigated and we follow the evaluation method as given in [26,27]. The stego key K _{ g } is fixed for each unique pair of users and it differs when the pair of sender and receiver is changed.
Effect of steganography
Cover pixel  Stego pixel  Overall pixel  Absolute pixel 

average  average  change  change 
173.26  173.48  0.47%  50.45% 
4.5 Case 2: Both fingerprints and shuffle keys are unknown
4.6 Case 3: Shuffle key is known
Now, we consider that the shuffle key (K _{shuf}) is compromised by the attacker and an attacker tries to generate the same cryptographic key using this shuffle key from the fingerprints other than genuine fingerprints. In this case, same shuffle key (K _{shuf}) is used to generate genuine and impostor keys. The transformation keys (S _{ k } and R _{ k }) are distinct for each fingerprint. In our experiment, Hamming distances are computed to measure the similarity or dissimilarity of the genuine keys and impostor’s keys. The observation is shown using histogram in Figure 8b. In this case, the Hamming distances are distributed from 37.50% (minimum) to 61.33% (maximum) with a mean of 50% and standard division of 0.0309. Maximum impostors’ keys (i.e., 99.85% impostors’ keys) differ from genuine key with the range of 40% to 60% Hamming distances. Even when the similarity of the impostor’s key is maximum (i.e., 62.5% bits are similar), the attacker needs to guess 96 bits (i.e., 2^{96} trials in brute force attacks) to crack the genuine key.
4.7 Case 4: Only one genuine fingerprint/cancelable template is known
In the proposed approach, we also consider that one of the genuine fingerprints (either the fingerprint of sender or receiver) along with transformation key is compromised by the impostor and cryptographic key is generated using one genuine cancelable template and one impostor’s cancelable fingerprint template. In this case, shuffle key K _{shuf} is unknown to the attacker and a different K _{shuf} is used to generate impostor keys. Whereas, we have used one fixed transformation key (S _{ k }) to generate genuine and impostor templates. The dissimilarity of the trial key with respect to genuine key is shown in Figure 8c. In this case, maximum 67.19% bits of the genuine key is unchanged for impostor’s key and up to 65.63% bits of the genuine key is changed in impostor’s key. Whereas, mean value of the Hamming distances and standard deviation of the distributions are almost similar as in other cases.
4.8 Case 5: One genuine fingerprint and shuffle key are known
In this case, we consider that the impostor knows the cancelable template of either sender or receiver and the shuffle key (K _{shuf}) to be used in the key generation. In this case, one fingerprint along with transformation key (either S _{ k } or R _{ k }) and K _{shuf} are common in genuine and impostor keys. Hamming distances are computed and plotted using histogram. The histogram in Figure 8d shows that the knowledge of fingerprint and shuffle key helps the attacker slightly. In this case, minimum 31.25% bits of the genuine key is changed with respect to impostor’s key and maximum 65.23% bits of the impostor’s key are different from genuine key.
4.9 Case 6: Both genuine fingerprints are known but shuffle key is unknown
Let us consider that attacker has complete knowledge about both genuine fingerprints or cancelable templates but no knowledge about shuffle key K _{shuf}. In this case, we have used same cancelable templates (i.e., same fingerprints with same S _{ k } and R _{ k }) with ten different shuffle keys to generate ten impostors keys and compare the similarity between genuine key and these impostors keys with respect to Hamming distance. The histogram in Figure 8e shows that an attacker is not able to compromise the cryptographic key even when both fingerprints/templates are known but the shuffle key is unknown. The average Hamming distance between a geneuine and impostor key is 49.99% of the length of cryptographic key. The most of the impostor’s keys (99.86% of impostor’s keys) are reported a dissimilarity between the range of 40% to 60% with respect to genuine key.
From our experimental results, we may conclude that an attacker is not able to generate the cryptographic key without the complete knowledge about both shuffle key and fingerprints of sender and receiver. According to our experiment, for all cases, minimum Hamming distance is 31.25% (case 5), that is, 80 bits of the impostor’s key are mismatched and maximum Hamming distance is about 65.63% (case 4) that is, 168 bits of the impostor’s key are needed to correct to break the genuine key. It is also observed that average Hamming distance is about 50% which means 128 bits among 256 bits of the genuine key are dissimilar on an average case. If we consider the average case, then at least 2^{128} trials are required in brute force attack to break the cryptographic key.
Computation time of our approach
Operations  Time (in sec) ^{a} 

Feature extraction  0.05 
Cancelable template generation  0.002 
Steganographic encoding and decoding  0.15 
Cryptographic key generation  0.002 
Total time  0.204 
5 Security analysis
In this section, security efficacy of the proposed approach is analyzed under different conditions. In our work, both communicating parties exchange their fingerprint data after transformation and mutually agree on two secrets like stego key (K _{ g }) and shuffle key (K _{shuf}).
5.1 Privacy of fingerprints
We use least significant bit (LSB)based steganography to exchange fingerprint data between sender and receiver. In our proposed approach, stego key (K _{ g }) is used to hide fingerprint data in the pixels which are randomly chosen from cover image. For each communication, a unique synthetic fingerprint image (I) is used as cover image. An eavesdropper does not suspect the existence of the genuine fingerprint data (i.e., cancelable template) due to high imperceptible stego image (I _{stego}). The cover image is not required to decode the hidden data from stego image. An adversary, with sufficient knowledge of decoding methods, is not able to extract the correct fingerprint data from the stego images \(I^{S}_{\text {stego}}\) and \(I^{R}_{\text {stego}}\) of sender and receiver, without possessing the K _{ g }. Our experimental result shows that only 0.47% pixels of cover image is modified, which assures that the conventional Targeted Steganalysis is not able to detect the existence and meaning of the hidden data [42]. The statistical steganalysislike histogram attack [43] detects only sequential embeddings but it does not detect the random embedding of small size message (i.e., size <50% of available LSB of I). Another steganalysislike sample pair analysis [44] is able to detect message embedding of size up to 5% of the available embedding space (LSB) of cover image while RS analysis [45] can detect embedding of message of size 2% to 5%.

According to the nature of transformation key, some cells may replace multiple cells and some cells may not replace any cell but they may be replaced by other cells. For example, say, there are four cells (1, 2, 3, 4) and they are replaced by (3, 2, 4, 3). Now it is found that cell 3 replaces cells 1 and 4, and cell 1 does not replace any cell. If transformation key and cancelable template both are compromised, then attacker may know the minutiae points belongs to cell 2, 3, and 4 but not of the cell 1.

In fact, an attacker needs to know the complete information about the fingerprint image size, that is, height and width of the image to compute the minutiae points accurately.
The cancelable templates (T _{CS}, T _{CR}) are fused, and the resultant template (T _{ f }) of fusion is shuffled and hashed to generate cryptographic key (K). Therefore, the cryptographic key is noninvertible, which confirms that the cryptographic key does not leak any information about the fingerprints of users.
5.2 Security of cryptographic key
In our approach, the cryptographic key is generated from the combination of two fingerprints of sender and receiver. The key is not shared by them but generated at their end separately. There is no need to store the key for the use in decryption. So the key is secured from any attack. An attacker needs to compromise either (stego key K _{ g }, sender’s stego image \(I^{S}_{\text {stego}}\), receiver’s stego image \(I^{R}_{\text {stego}}\), shuffle key K _{shuf}) or (sender’s fingerprint F _{ S }, receiver’s fingerprint F _{ R }, transformation key of sender S _{ K }, transformation key of receiver R _{ K }, K _{shuf}) to generate the genuine cryptographic key (K). Otherwise, the following conditions arise for different attacks.
5.2.1 5.2.1 Known stego key attack
It means that the stego key K _{ g } is compromised by eavesdropper who eavesdrops both stego images \(I^{S}_{\text {stego}}\) and \(I^{R}_{\text {stego}}\). Next, as the shuffle key K _{shuf} is unknown, he has to guess the K _{shuf} which is of 200 bits. It requires a trial of 2^{200} to break the K _{shuf} using brute force attacks. Indeed, without K _{shuf}, it is almost impossible to compute the original cryptographic key even if both fingerprints data are known.
5.2.2 5.2.2 Known shuffle key attack
It happens when the token is stolen or lost and the attacker gets access to the shuffle key (K _{shuf}) but he has no knowledge about the fingerprint data of genuine sender and receiver. In this case, the attacker has to guess 50+50=100 minutiae points for two fingerprints. Otherwise, he has to compute the key using impostor’s fingerprints which may be available to him. As the fingerprint is unique for a person, there is a rare chance to get the same fingerprint from impostor’s fingerprint database. Even in our experiment, there is no collision between genuine cryptographic key and impostor’s cryptographic key when the K _{shuf} is known. The shuffle key is updated in every session, which guarantees that it is impractical to compute the K _{shuf} of previous session or the same of future sessions.
5.2.3 5.2.3 Known fingerprint attack
In this attack, there are two scenarios that may occur. In the first scenario, only the fingerprints of both parties are compromised but transformation key is not known by attacker. In the second scenario, the attacker has complete knowledge about the fingerprints along with transformation key, that is, the attacker is able to generate cancelable templates from fingerprints (of both sender and receiver) using the transformation keys and algorithm of transformation.
where K is the cryptographic key and ‘?’ marks represent those parameters which are unknown to the adversary.
only one parameter (K _{shuf}) which is unknown to the adversary. Then, the attacker has to break the K _{shuf} which needs 2^{200} trials in brute force attack.
5.2.4 5.2.4 Known key attack
In this case, we consider that the shuffle key K _{shuf} and or cryptographic key (K) are compromised by the attacker. Our proposed approach assures that compromise of (\(\phantom {\dot {i}\!}K_{\text {shuf}_{i}}\) or K _{ i }) of ith session does not affect the previous or future session as \(\phantom {\dot {i}\!}K_{\text {shuf}_{i}} \neq K_{\text {shuf}_{i+1}} \neq K_{\text {shuf}_{i1}}\) or \(\phantom {\dot {i}\!}K_{i} \neq K_{i+1} \neq K_{i1}\). In our approach, both K _{shuf} and K are updated session wise. Initially, K _{shuf} is randomly generated and time to time it is updated with the help of F _{code}. Cryptographic key (K) is revocable and is used as a session key for symmetric cryptography. When the session is over, the current session key is destroyed.
5.2.5 5.2.5 Resists replay attack
Our approach can prevent replay attack using session key. In every session of communication, a unique session key is used to establish a secure communication between sender and receiver and the session key is destroyed after the session. Our proposed approach is able to generate 100!×100! different cryptographic keys from two fingerprint biometric traits. If an eavesdropper wants to make replay attack using a message previously transmitted by legal users, then it will make no sense to the legal user as the cryptographic key is changed. Even when stego image is used for replay attack, eavesdropper is not able to decode the stego image without the stego key.
5.2.6 5.2.6 Resists maninmiddle attack
In our approach, fingerprints of communicating parties are transmitted over communication channel using data hiding scheme. If the maninmiddle (MiM) eavesdrops the stego image and is able to decode the hidden data by any means, then the MiM requires the perfect knowledge of secret shuffle key (K _{shuf}). Otherwise, he is not able to generate the genuine key (K) and, as a result, is not able to decrypt the ciphertext sent by genuine sender.
where D _{ s } is the data unhiding function, TCS′ and TCR′ are the cancelable templates of sender and receiver, respectively, extracted by MiM, whereas Kshuf′ is the estimated shuffle key by the MiM. If K _{shuf}≠Kshuf′, then the estimated cryptographic key K _{est} does not match with the genuine key K.
where the cancelable template of MiM is T _{mim} and K ^{ S }, K ^{ R } are the keys generated by sender and receiver, respectively, and used in encryption. \(K^{S}_{\text {est}}\) and \(K^{R}_{\text {est}}\) are the estimated key by the MiM and used for decryption.
In the estimation of cryptographic keys \(\left (K^{S}_{\text {est}}, K^{R}_{\text {est}}\right)\), MiM trials the K _{shuf} which produces different keys, \(\left (\text {i.e.,}~~ \left (K^{S}_{\text {est}} \neq K^{S}\right), ~~\text {and} ~~\left (K^{S}_{\text {est}} \neq K^{R}\right)\right)\), which do not decrypt the ciphertexts correctly (i.e., \(\text {message}_{S} \neq \text {message}^{m}_{S}\) and \(\text {message}_{R} \neq \text {message}^{m}_{R}\)). In this way, our approach resists maninmiddle attack.
6 Conclusions
Cryptographic key generation and subsequently its maintenance are the two important issues in traditional cryptography. A cryptographic key should be generated in such a way that it is hard enough to guess and then it should be managed without any overhead of users. This work addresses these issues and propose a novel approach to generate random cryptographic key using fingerprint biometric of sender and receiver.
In our works, the privacy and security of fingerprint data are provided with cancelable template. Also, we propose a protocol with which key can be revoked thus addressing the limitation of irrevocability property of biometric trait. More significantly, there is no need to store the key, prior to communication. In fact, our protocol adds more security allowing to generate different keys in different sessions. The proposed cryptobiometric system is resilient to many attacks such as known key attacks, replay attack, maninmiddle attacks, etc. Our proposed approach thus provides an effective solutions where we need a sessionbased cryptographic key during message transmission over an insecure network channel.
Declarations
Authors’ Affiliations
References
 W Stallings, Cryptography and Network Security: Principles and Practice, 5e (Prentice Hall, 2010).Google Scholar
 Advance Encryption Standard (AES), Federal Information Processing Standards Publication 197 (United States National Institute of Standards and Technology (NIST), November 26, 2001).Google Scholar
 K Mitnick, W Simon, S Wozniak, The Art of Deception: Controlling the Human Element of Security (Wiley, New York, 2002).Google Scholar
 DV Klein, in Proceedings of the 2nd USENIX Security Workshop (Portland). Foiling the cracker: A survey of, and improvements to, password security, (1990), pp. 5–14.Google Scholar
 F Hao, R Anderson, J Daugman, Combining Crypto with Biometrics Effectively. IEEE Trans. Comput. 55(9), 1081–1088 (2006).View ArticleGoogle Scholar
 U Uludag, S Pankanti, S Prabhakar, AK Jain, Biometric Cryptosystems: Issues and Challenges. Proc. IEEE. 92(6), 948–960 (2004).View ArticleGoogle Scholar
 D Maltoni, D Maio, AK Jain, S Prabhakar, Handbook of Fingerprint Recognition (Springer Verlag, New York, 2003).MATHGoogle Scholar
 K Nandakumar, A Jain, S Pankanti, Fingerprintbased fuzzy vault: Implementation and performance. IEEE Trans. Inf. Forensics Secur. 2(4), 744–757 (2007).View ArticleGoogle Scholar
 N CT Charles, DJ Kiyavash, in Proceedings of the 2003 ACM SIGMM workshop on Biometrics methods and applications. Lin, Secure smartcardbased fingerprint authentication (ACMNew York, NY, USA, 2003), pp. 45–52.Google Scholar
 S Yang, I Verbauwhede, in Proceedings (ICASSP’05). IEEE International Conference on Acoustics, Speech, and Signal Processing, Vol. 5. Automatic secure fingerprint verification system based on fuzzy vault scheme (IEEEPhiladelphia, Pennsylvania, USA, 2005), pp. v/609–v/612.Google Scholar
 S Dutta, A Kar, BN Chatterji, NC Mahanti, in Proc. Adv. Concepts Intell. Vis. Syst., LNCS 5259. Network Security Using Biometric And Cryptography (Springer BerlinHeidelberg, 2008), pp. 38–44.View ArticleGoogle Scholar
 A Jagadeesan, K Duraiswamy, Secured Cryptographic Key Generation from Multimodal Biometrics: Feature Level Fusion of Fingerprint and Iris. Int. J. Comput. Sci. Inform. Secur. 7(2), 28–37 (2010).Google Scholar
 A Jagadeesan, T Thillaikkarasi, K Duraiswamy, Cryptographic Key Generation from Multiple Biometrics Modalities: Fusing Minutiae with Iris Feature. Int. J. Comput. Appl. 2(6), 16–26 (2010).Google Scholar
 F Monrose, MK Reiter, Q Li, S Wetzel, in Proceedings of IEEE Symposium on Security and Privacy. Cryptographic key generation from voice (IEEE Computer SocietyWashington, DC USA, 2001), pp. 202–213.Google Scholar
 H Feng, CC Wah, Private key generation from online handwritten signatures. Inform Manag. Comput. Secur. 10(4), 159–164 (2002).Google Scholar
 B Chen, V Chandran, in Proceedings of 9th Biennial Conference of the Australian Pattern Recognition Society on Digital Image Computing Techniques and Applications. Biometric Based Cryptographic Key Generation from Faces (GlenelgAustralia, 2007), pp. 394–401.Google Scholar
 SVK Gaddam, M Lal, Efficient Cancellable Biometric Key Generation Scheme for Cryptography. Int. J. Netw. Secur. 11(2), 57–65 (2010).Google Scholar
 AK Jain, K Nandakumar, A Nagar, in Security and privacy in biometrics. Fingerprint Template Protection: From Theory to Practice (Springer London, 2013), pp. 187–214.Google Scholar
 NK Ratha, S Chikkerur, JH Connell, RM Bolle, Generating Cancellable Fingerprint Templates. IEEE Trans. Pattern Anal. Mach. Intell. 29(4), 561–572 (2007).View ArticleGoogle Scholar
 A Ross, AK Jain, Information fusion in biometrics. Pattern Recognit. Lett. 24, 2115–2125 (2003).View ArticleGoogle Scholar
 A Bodo, Method for Producing a Digital Signature with Aid of Biometric Feature. German Patent DE 4243908A1 (1994).Google Scholar
 JH RM Bolle, S Connell, NK Pankanti, AW Ratha, Senior, Guide to Biometrics (SpringerVerlag, New York, 2003).Google Scholar
 NK Ratha, JH Connell, R Bolle, Enhancing Security and Privacy in BiometricBased Authentication System. IBM Syst. J. 40(3), 614–634 (2001).View ArticleGoogle Scholar
 S Kanade, D Camara, E Krichen, D PetrovskaDelacre~taz, B Dorizzi, Evry F, in Proceedings of 6th Biometrics Symposium (BSYM 2008). Three Factor Scheme for BiometricBased Cryptographic Key Regeneration Using Iris (Tampa, Florida, USA, 2008), pp. 59–64.Google Scholar
 S Kanade, D PetrovskaDelacre~taz, B Dorizzi, in Proceedings of Fourth IEEE International Conference on Biometrics: Theory Applications and Systems (BTAS), Washington, DC, USA, 2010. Generating and sharing biometrics based session keys for secure cryptographic applications, (2010), pp. 1–7.Google Scholar
 A Jain, U Uludag, Hiding Fingerprint Minutiae in Images. in Proceedings of Third Workshop on Automatic Identification Advanced Technologies (AutoID), (Tarrytown, New York. USA, 97–102 (2002).Google Scholar
 A Jain, U Uludag, in Proceedings of 16th International Conference on Pattern Recognition, vol. 3. Hiding a Face in a Fingerprint Image (Canada, 2002), pp. 756–759.Google Scholar
 A Jain, U Uludag, Hiding Biometric Data. IEEE Trans. Pattern Anal. Mach. Intell. 25, 1494–1498 (2003).View ArticleGoogle Scholar
 N Agrawal, M Savvides, in Proceedings of IEEE Computer Society Conference on Computer Vision and Pattern Recognition. Biometric data hiding: A 3 factor authentication approach to verify identity with a single image using steganography, encryption and matching (Miami Beach Florida, 2009), pp. 85–92.Google Scholar
 A Juels, M Wattenberg, in Proc. 6th ACM Conf. Computer and Communications Security, ed. by G Tsudik. A fuzzy commitment scheme (ACMNew York, NY, USA, 1999), pp. 28–36.Google Scholar
 C Rathgeb, A Uhl.Contextbased biometric key generation for Iris. IET Comput. Vis. 5(6), 389–397 (2011).View ArticleGoogle Scholar
 C YaoJen, W Zhang, T Chen, in Proceedings of IEEE International Conference on Multimedia and Expo (ICME’04), Taipei, 2004, Vol. 3. Biometricsbased cryptographic key generation (IEEE, 2004), pp. 2203– 2206.Google Scholar
 N Lalithamani, KP Soman, in The 2nd International Conference on Computer Science and Information Technology. Towards Generating Irrevocable Key for Cryptography from Cancelable Fingerprints (Beijing China, 2009), pp. 563–568.Google Scholar
 N Lalithamani, KP Soman, Irrevocable Cryptographic Key Generation from Cancelable Fingerprint Templates: An Enhanced and Effective Scheme. Eur. J. Sci. Res. 31(3), 372–387 (2009).Google Scholar
 V Lokeswara Reddy, A Subramanyam, P Chenna Reddy, Implementation of LSB Steganography and its Evaluation for Various File Formats. Int. J. Adv. Netw. Appl. 02(05), 868–872 (2011).Google Scholar
 A Ross, K Nandakumar, AK Jain, Handbook of Multibiometrics (SpringerVerlag, Berlin, Germany, 2006).Google Scholar
 Fingerprint Verification Competition FVC2000, [Online]. Available: http://bias.csr.unibo.it/fvc2000.
 Fingerprint Verification Competition FVC2002, [Online]. Available: http://bias.csr.unibo.it/fvc2002.
 Fingerprint Verification Competition FVC2004, [Online]. Available: http://biometrics.cse.msu.edu/fvc04db/index.html.
 R Cappelli, A Erol, D Maio, D Maltoni, in Proceedings of 15th International Conference on Pattern Recognition, 2000, vol. 3. Synthetic Fingerprint Image Generation, (2000), pp. 475–478.Google Scholar
 C Watson, M Garris, E Tabassi, C Wilson, M McCabe, S Janet, Ko K, User’s Guide to NIST Biometric Image Software (NBIS) (National Institute of Standards and Technology, Gaithersburg, MD, 2007).Google Scholar
 A Rocha, W Scheirer, T Boult, S Goldenstein, Vision of the unseen: Current trends and challenges in digital image and video forensics. ACM Comput. Surv. 43(4), 26:1–26:42 (2011).View ArticleGoogle Scholar
 A Westfeld, A Pfitzmann, in Proc. Information Hiding, 3rd Int’,l Workshop. Attacks on Steganographic Systems (Springer Verlag, 1999), pp. 61–76.Google Scholar
 S Dumitrescu, Wu Xiaolin, N Memon, in International Conference on Image Processing, (ICIP 2002), vol.3, no., 2428. On steganalysis of random LSB embedding in continuoustone images (Rochester, New York, USA, 2002), pp. 641–644.Google Scholar
 J Fridrich, M Goljan, Du Rui, Detecting LSB steganography in color, and grayscale images. MultiMedia IEEE. 8(4), 22–28 (2001).View ArticleGoogle Scholar
Copyright
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited.