1. Introduction

In the last decade, optical technologies have largely contributed to improve security systems in terms of proposing new methods for information encryption and designing identification (ID) tags for object surveillance or tracking, among other aspects.

Regarding cryptology, special attention has been paid to develop new encryption methods that ensure information confidentiality during the complete process of host encryption, transmission and client decryption. Since the optical encryption scheme based on double random phase keys was proposed in [1] some modifications of this initial proposal [2–6] and a number of other ciphering techniques have been developed [7–12]. Recently, a report on cryptanalysis [13] has shown certain vulnerability of double random phase based ciphering techniques to some given attacks due to the linearity of the encryption scheme. For this reason, thorough analysis of ciphering techniques and further development of encryption methods are required to consider a system to be secure.

The common objective of the aforementioned contributions is to keep a piece of information in secrecy. Other research papers have pointed out the need of encoding multiple primary images either for holographic optical storage [2,4,14–17] or, very recently, for increasing the reliability of the security system [18]. Focusing on this latter contribution, different categories of identity signals or factors are combined to produce a multifactor authentication that only gives positive verification when the whole set of signals are identified [18,19] achieving an attractive proposal for high-security applications with strict identification requirements.

In a close field, optical identification (ID) tags [20] have been introduced for robust, realtime and remote identification to enable surveillance or tracking of moving objects, such as vehicles or parcels on a conveyor belt. From the development of the first proposal, specific designs for distortion-invariant ID tags were presented [21,22] to allow remote information readout under the effects of scale variations or/and in-plane rotations. A review of the ID tag design as well as the influence of different sources of noise can be found in [22,23].

None of the optical techniques introduced in the literature so far handles the situation as a whole, but partial aspects. As it has been mentioned, some contributions focus on the encryption procedure to keep information confidential, others on the object surveillance at a distance. This paper suggests a new scenario in securing techniques involving remote identification. We propose a new combination of the multifactor encryption procedure and optical ID tags to take full advantage of both techniques. Moreover, the reliability of the security system can be increased by using infrared techniques working together with the previous ones. We design a novel compact technique for encryption-verification that relates for the first time the following four elements: multifactor encryption, distortion-invariant ID tag, near infrared (NIR) readout, and optical processor. A highly-reliable security system is obtained by joining the advantages of these four elements as it is described in detail in Section 2. In addition to this, the paper also aims to show the remarkable characteristics that the designed NIR ID tag exhibits, namely distortion-invariance, easy and economical tag building and increased robustness. They are pointed out by the verification results provided in this work.

This security system is addressed to tackle situations such as that illustrated in Fig. 1, which we consider a representative case of others with similar level of complexity and requirements. Let us consider the surveillance and tracking of classified parcels that have been confiscated and which are located on a conveyor belt for inspection. This situation may require the control of a number of elements: the person who is responsible for delivering the parcels, their origin, their content, and also the destination (place and/or time). Only by assuring the whole control of all these features, one can be sure that there has not been any unauthorized manipulation. In this identification task, the four factors to verify belong to both, the person and the parcel involved in the control process. Regarding the person, a biometric signal such a retinal scanning can be used as a very stable signature. Information about the origin and content of the parcel can be stored either in a database or a card. At the same time, one of the factors of the encryption method can be used as a random key that, for instance, controls the access of one person to a given area or for a given period of time.

 

Fig. 1. (1.4 Mb) Movie of classified parcel surveillance by using multifactor authentication and NIR distortion-invariant optical ID tags (4.4 Mb version). [Media 1]

Download Full Size | PDF

The information corresponding to the set of chosen factors or primary images has to be previously encrypted and then, introduced into a NIR ID tag so that it is not recognizable at the naked eye or using conventional cameras in the visible spectrum. The distortion-invariant NIR ID tag will permit the readout of the information from different distances or points of view, due to the fact that parcels may have different sizes and orientations when passing through the control machine. In the case shown in Fig. 1, the verification task implies the multiple AND-validation of the set of input images with respect to the set of primary images obtained from the ID tag deciphering. The set of input images consists of a biometric, some signals about the parcel and a random key and they are captured in situ from the person, the card and a database. Only the satisfactory validation of the whole set of input images will provide a positive verification. As shown in Fig. 1, tolerance to scratches or other sources of noise needs to be considered to a certain extent. On the other hand, a system for an automatic destruction of the ID tag must be applied to prevent tags and parcels from unauthorized manipulation or tampering.

2. Principles of NIR multifactor ID tag

The main aspects of the compact encryption-verification procedure are presented in detail in this section. Multifactor encryption permits the simultaneous verification of a number of factors. This technique has no a priori constraint on the piece of information taken into account. The images on which the identification is based can be of different category (e.g. biometric, signatures, matrix bar codes, and alphanumeric sign) and/or imaged in different spectral bands (e.g. Visible, IR, UV) by using the proper cameras. Remote identification is achieved using distortion-invariant ID tags. The processor gathers information from different sources such as from the NIR region. A similar principle inspires the multidimensional optical sensor described in [24]. The optical processor is based on joint transform pattern recognition by optical correlation to verify the information. The novel NIR ID tag is experimentally built and read out. By using commonly available materials, distortion-invariant NIR ID tags are built so that their content is visible just in the NIR region. It cannot be either copied, scanned, or captured by any conventional device. Only a receiver with sensitivity to NIR can read out the NIR ID tag and send the information acquired for decoding, decryption and verification.

Figure 2 depicts a schematic diagram of the procedure steps for the sake of clarity. The multifactor encryption method is firstly introduced [Fig. 2(a)]; secondly, the distortion-invariant NIR ID tag design is described; and finally, the information readout and verification process is presented [Fig. 2(b)].

2.1 Multifactor encryption

The selection of authenticators is a crucial step because the identification of an element (object or person or both) is based on them. They must uniquely represent the element whose identity is to be validated on a basis of signal recognition. Frequently, the authenticators are images such as logotypes, bar codes, alphanumerical signs, signatures, biometric information, and random sequences. Combining information from different sources reinforces system security. The principles and the mathematics of the method for a four-factor authentication were described in [18]. Considering the situation we described in Section 1, our aim is to simultaneously control and verify classified parcels along with the person responsible for their delivering. In our system, we choose up to four signatures or factors that contain relevant information to identify both the person and the parcel, regarding its origin, content and destination. The primary images to be considered as distinct features to identify a given parcel and its holder are the following [Fig. 3(a)]:

 

Fig. 2. (a) Encryption of multiple primary images and codification into an invisible sign (NIR ID tag). (b) Readout and verification process.

Download Full Size | PDF

First, a person biometric, such as fingerprint, face, hand, iris, retinal scan, which are considered stable biometric signals as they are relatively constant in time. However, handwriting, voice, gait, are easily alterable. Among all these biometric signals, retina scan is considered one of the most reliable biometrics, very stable and difficult to alter, so it is taken into account in our example as one of the identifying factors used to verify the identity of the person [function s(x) in Fig. 3(a)]. This piece of information can be obtained in situ from the person using a retinal camera in order to check its coincidence with the corresponding biometric encrypted in the ID tag. Only the most relevant features of the retina scan, which correspond to the spatial distribution of vessels, are considered in a binary image.

Secondly, two factors may refer to the parcel. One signal to be considered is its content [r(x) in Fig. 3(a)]. Hazardous substances, explosives, narcotics, fragile items and so on must be precisely labeled and carefully handled. The origin of the parcel is another data [n(x) in Fig. 3(a)] to consider, for instance, whether it has been confiscated by the police, at customs, or it has been produced in a specific center, etc. This parcel information can be supplied by the parcel holder using either a card or a database.

Finally, a key code [b(x) in Fig. 3(a)], which is a random phase sequence, is needed to give the final noisy appearance of the encrypted distribution. It can be used to control the destination, that is, the unlimited or restricted access to a place on set days, and it is provided from a database.

Let the four s(x), r(x), b(x) and n(x) be the set of reference primary images, in one-dimensional notation for simplicity, to be encrypted in the ID tag as the secret information for multifactor verification. All the four reference primary images are normalized positive functions distributed in [0,1]. The complex-amplitude encrypted function ψ(x) containing the multifactor authenticators is given by the equation

where t_f(x) = exp{jπf(x)} defines the phase-encoded function of image f(x) (being f = s,r,b,n or lineal combinations of them), F^-1 indicates inverse Fourier transform, and ^∗ the convolution operation. The computed complex-valued encrypted distribution ψ(x) accomplishes the main features for an encrypted function to be reliable in terms of security: the set of four signals [Fig. 3(a)] are scrambled together and they remain invisible in the resulting distribution [Fig. 3(b)], the encrypted function is easy to be computed and reproduced, and at the same time it is difficult to fake. The ciphered function ψ(x) will be organized in an ID tag that is to be attached to the item (parcel) as we will explain in the following subsection.

 

Fig. 3. (a). Signatures and key code to be encrypted; (b) Magnitude and phase distribution of the encrypted multifactor function.

Download Full Size | PDF

2.2 Distortion-invariant NIR ID tag

We aim to achieve detection and verification of the encrypted distribution ψ(x), which will be included in an ID tag, even when the receiver captures it from an unexpected location or orientation. To do so, the proposed ID tag, which will be attached to the item under surveillance, should be invariant to distortions such as scale variations and rotations. Reference [22] offers a deeper insight of the main characteristics that meet the need for more useful and robust optical ID tags.

Distortion-invariance is achieved by both multiplexing the information included in the ID tag and taking advantage of the ID tag topology, without unnecessarily increasing the system complexity [21–22]. The complex-valued function ψ(x) obtained from Eq. (1) is to be fully grayscale encoded. It is convenient to print the phase content of ψ(x) in grayscale variations rather than in phase. Otherwise, the phase content of the encrypted distribution could be easily neutralized and the ID tag sabotaged if an adhesive transparent tape were stuck on it. For this reason it is useful to further encode the phase content of the signal in intensity variations. Thus, we consider encoding both the magnitude and phase in grayscale values [22]. Let us consider the encrypted function ψ(x) in array notation ψ(t) = |ψ(t)|exp{jϕ_ψ(t)} where t = 1,2,…N and N is the total number of pixels of the ciphered function. We build two real valued vectors that are to be encoded grayscale: the magnitude vector |ψ(t)| and the phase vector ψ_ψ(t). The information included in the ID tag can be distributed either in a single [21], or in two optical codes [22]. In the latter, the real-valued magnitude function and the phase distribution of ψ(t) are separately reproduced in the final ID tag. This second representation makes the authentication process more robust.

The information included in the ID tag is distributed in two circles. Different possibilities can be considered to rearrange the information contained in these two circles. Fig. 4(a) shows a possible arrangement of both circles fully reproduced in amplitude. One of them corresponds to the magnitude of the encrypted function [left circle in Fig. 4(a)]. The other contains its phase distribution [right circle in Fig. 4(a)]. In both circles the information is distributed similarly to the structure of a wedge-ring detector. One half of each circle [upper semicircles in Fig. 4(a)] includes either the magnitude of the phase distribution written in a radial direction and repeated angularly so that rotation-invariance can be achieved. The other semicircle of both circles [bottom semicircles in Fig. 4(a)] contains either the magnitude or the phase written circularly and repeated in concentric rings. Therefore, the information of a given pixel of the encrypted function will correspond to an angular sector in the optical code. Thus, the readout of the ciphered information will be tolerant to variations in scale. For encrypted functions with a large number of pixels, such as our example in Fig. 4(a), information of the scale-invariant ID tag have to be distributed by using different concentric semicircles to assure a minimum number of pixels for each sector to recover the information properly. Consequently, the tolerance to scale variations will be affected in accordance to the number of concentric circles used in the ID tag.

In an alternative rearrangement, one circle allows the rotation-invariant readout of both magnitude and phase of the encrypted function and the other circle the scale-invariant readout of the same signal [23]. The choice of a particular distribution of the signal information depends on practical considerations of a given problem. A mathematical description of the distortion-invariant ID tag can be found in [22].

As an additional degree of security we aim to increase the system robustness to counterfeiting by gathering the data of the ID tag from the NIR region of the spectrum. In such a way, data is no longer visible at naked eye and only by using the adequate sensor it is possible to grab the correct information.

 

Fig. 4. Optical distortion-invariant ID tag (rotation angle 7 degrees) experimentally captured by using (a) a NIR XEVA camera and (b) a visible SONY 9100P camera. (c) Reference isosceles triangle of white spots.

Download Full Size | PDF

The NIR ID tag is built by printing the ID tag gray level distribution with a common laser printer on a black cardboard. In the visible spectrum, the whole information is completely hidden to either the naked eye or common cameras operating in the visible region of the spectrum [Fig. 4(b)]. In such a way it is not possible to know neither the kind of information included in the ID tag nor the exact position of this ID tag over the item under surveillance. Only NIR InGaAs cameras or conventional monochrome CCD cameras without the IR cut-off filter are able to detect the appropriate information to decode and verify the multifactor information. Figure 4(a) shows the distortion-invariant ID tag as it is experimentally captured by a camera sensitive to the NIR spectrum. Scale and rotation invariant regions are clearly detected. Incandescent illumination was used to capture the images shown in Fig. 4.

Looking again at the movie of Fig. 1, there is a parcel whose ID tag shows some scratches due to common friction. The redundant information multiplexed in the tag allows signal verification despite the scratches. On the other hand, there is a different case for which the parcel suffered an attempt of tampering and the ID tag was cut. In such a case, there is an auto-destruction mechanism of the ID tag. For example, we have designed a reservoir of black ink (black in terms of NIR illumination) under the ID tag. When the tag is cut, the ink is spread throughout it, the tag cannot be properly read, and the processor gives an alarm. The sequence of images displayed in Fig. 1 shows the last two cases.

2.3 Information readout

The encrypted information is recovered by the procedure described in Fig. 2(b). First, the NIR ID tag is captured by a NIR sensitive device. Since the receiver resolution is not generally known a priori, the image of the triangular-shaped pattern consisting of three white spots [Fig. 4(c)] can be used as a reference to know if the receiver has enough resolution to read the encrypted information. Since the triangle pattern can provide information about scale and rotation, therefore, one could object that there is no need to codify the encrypted distribution in the distortion-invariant ID tag. However, we must take into account that if the encrypted function, written in a matrix array, is affected by rotation and/or scale variations, then it needs to be sampled again and rearranged into the matrix form before decryption. This operation entails interpolations that can produce errors such as aliasing. For this reason we consider that the distortion-invariant ID tag, provided it is correctly built, allows more accurate readouts of the encrypted information under rotation and/or scale variations.

The triangular-shaped pattern permits to allocate the two circles [Fig. 4(c)]. The base of the isosceles triangle determines the border line between the two sectors of each circle. The vertex of the triangle distinguishes between the semiplanes where rotation invariance [upper semiplane in Fig. 4(c)] and scale invariance [bottom semiplane in Fig. 4(c)] are achieved. The ciphered information in vector notation ψ(t) can be decoded by reading out the information of the optical code from either the rotation-invariant or the scale-invariant areas. The magnitude and phase of the encrypted function are extracted from both circles of the ID tag. From the semicircles corresponding to the rotation-invariant areas, the magnitude and the phase could be read out by using a linear array detector placed in any radial direction of the imaged semicircles [22]. Not only is a single pixel value read along a unique radius, but a set of different radial codes are read and their median value is computed to increase noise robustness. From the semicircles corresponding to the scale-invariant areas, the magnitude and phase in vector notation are recovered by reading out the pixels of the ID tag in semicircular rings. To minimize errors in the reading process, the median value of a set of pixels located in neighbor concentric rings in the radial direction are considered. The retrieved information of the pixels should be written back into matrix notation prior to verify the multifactor authentication. Following this procedure, the encrypted distribution will be recovered when the ID tag is captured in its original orientation and size or in rotated or scaled formats.

Previous works have shown that information redundancy in the design of the ID tag allows a certain tolerance to the presence of additive noise [22, 23] in the capturing process.

2.4 Verification

Once the encrypted multifactor distribution is retrieved, the verification step is carried out. Let q(x), p(x), d(x) and m(x) denote the positive and normalized input images to compare with the set of reference primary images obtained after deciphering the ID tag. Simultaneous verification of the complete set of images is done by an optical processor that combines a nonlinear JTC and a classical 4f-correlator as it is described in Ref. [18]. A variety of nonlinear techniques [25–28] can be applied during this verification step so that the system discrimination capability can be adjusted and noise resistance improved, among other properties. The primary and input images are appropriately introduced into different planes of the optical processor according to the description of Ref. [18]. When the signatures to compare coincide with the information included in the ID tag, that is, the multiple AND condition s(x) = q(x) AND r(x) = p(x) AND b(x) = d(x) AND n(x) = m(x) is fulfilled, a positive validation occurs. The term of interest for the multifactor application in the output plane of the optical processor corresponds to the cross-correlation of autocorrelation (AC) signals given by

provided that the phase extraction technique [25] is applied as a nonlinearity. In Eq. (2) the symbol ⋆ denotes cross-correlation, subindices CMF (classical matched filter), POF (phase-only filter), PPC (pure phase correlation) indicate the sort of filter involved in the autocorrelation signal. A capital T_f stands for the Fourier transform of the function in small t_f. A sharp and intense multifactor autocorrelation peak is obtained in the output plane provided the system is free of noise and distortions. Consequently, the information contained in Eq. (2) allows reinforced security verification by simultaneous multifactor authentication.

If any of the authenticator signals do not coincide with the corresponding reference primary image, that is s(x) ≠ q(x) or r(x) ≠ p(x) or b(x) ≠ d(x) or n(x) ≠ m(x), the output contains a cross correlation signal that is, in general, broader and less intense than the multifactor autocorrelation peak of Eq. (2). Furthermore, the key code known by the processor, b (x), plays an important role in optical security as an additional authenticator with the properties described in Refs. [1, 3, 29–30].

3. Experimental multifactor validation

3.1 Experimental capturing conditions

To show the feasibility of the proposed compact encryption-verification system, some experimental results were obtained and analyzed. For practical reasons, we have computed the encrypted distribution ψ (x) using techniques for computer generated holograms (CGH) and then, also by computer, we have rearranged the information to redistribute it into the two discs of amplitude and phase that compose the ID tag. A distortion-invariant ID tag containing the multifactor encrypted information was produced by printing the ID tag using a common Hewlett Packard laser printer on a black cardboard. The printed ID tag was uniformly illuminated by ordinary incandescent light bulbs and grabbed lately by a NIR InGaAs camera [Fig. 4(a)], with sensitivity in the NIR region (900–1700nm). This result shows a feasible way to obtain NIR ID tags using common materials. Just for comparison the ID tag was also registered using a monochrome camera sensitive in the visible region of the spectrum [Fig. 4(b)] to show that its content cannot be perceived at naked eye.

3.2 Identification results

The ID tag was experimentally captured, and read out to obtain the encrypted distribution in matrix notation. The registered ID tag was rotated 7 degrees from horizontal position [Fig. 4(a)]. The scrambled four factors (primary images s(x), r(x), b(x), n(x)) were decrypted and introduced as a reference for the validation of the set of input images q(x), p(x), d(x), m(x). Figure 5 shows the output planes of the processor for different situations that correspond to the most relevant identification results obtained in the experiment. The maximum intensity value of the output planes, normalized to the case where a satisfactory verification is achieved, are summarized in Table 1. In both, Fig. 5 and Table 1, the predicted ideal results are shown along with the experimental ones for comparison. A high and sharp multifactor autocorrelation peak is essential for the method to be reliable. This situation occurs when the system is free of noise and distortions. In practice, image noise particularly arises when using coherent illumination. It is well known that varieties of non-linear techniques, useful to adjust the discrimination capabilities of a system and to improve its noise robustness, do exist. In the present proposal, we have used phase extraction [25–26] as it has been mentioned in Section 2.4.

The first analyzed case [Fig. 5(a)] corresponds to a positive validation, for which the four input images coincide with the primary images included in the ID tag. As a result, an intense and sharp intensity peak projects over a low background on the output plane of the optical processor. This peak stands for the correct identification of an authorized person and parcel. The maximum intensity value of this output plane is normalized to one for the sake of comparison with the following verification results. Also for comparison, the output correlation signal of the processor is depicted for an ideally captured ID tag and on its side for the experimentally captured ID tag [Fig. 4(a)].

 

Fig. 5. Experimental and simulated results for the verification system by using distorted NIR multifactor ID tags: (a) Positive validation when the four identifying factors coincide with the information included in the ID tag; Negative results obtained when one (b) or more factors (c) do not coincide with the set of primary images; (d) Positive validation with a partially scratched ID tag. In all cases, verification outputs are normalized to the positive validation (a).

Download Full Size | PDF

When the parcel origin does not match the one included in the ID tag [Fig. 5(b)], the obtained output plane shows an insignificant intensity peak that hardly projects over the background. An appropriate threshold value will indicate that the verification process is not satisfied due to the fact that one factor corresponds to an unauthorized signal.

If another signal among the four factors (biometric, parcel content or key code) or even the whole set of input images does not correspond to the set introduced into the ID tag, the resulting output planes are very similar to the one plotted in Fig. 5(c). As an example, we show the result obtained for four unauthorized signals. A low energy background is produced at the system output.

Finally, if the ID tag is slightly scratched due to friction [Fig. 5(d)], a positive verification result is still obtained when the whole set of input images coincides with the authorized factors included in the ID tag as it is proved by the sharp and intense peak that projects on the output plane.

Table 1 summarizes the numerical results obtained for the analyzed authentication tasks. A positive validation is only obtained when the four input images match the primary images contained in the ID tag. If one or more input images do not coincide with the set of primary images, a negative result is obtained in the verification process.

In all the analyzed examples, there is a satisfactory agreement between the experimental and the predicted verification results.

Table 1. Experimental and simulated results for the multifactor verification system. The maximum intensity peak value of the output plane of the verification system is given for different situations. Values are normalized to the positive validation which is when the four factors coincide with the authorized factors included in the captured ID tag.

View Table

4. Conclusions

This paper suggests a new scenario in securing techniques involving remote identification. A compact encryption-verification technique has been proposed for highly-secure identification systems. It relies on the combination of a multifactor encryption method, NIR ID tags and a joint pattern recognition-based optical processor. The distortion-invariant NIR ID tag, which has a non visible signal, can be built by using commonly available materials so that it is only captured in the NIR spectral region. ID tag designing is smart: distortion-invariant, NIR ID tag printing on a black cardboard just using a laser printer, and increased robustness are the remarkable characteristics this kind of tag exhibits. Experimental verification results have been shown to demonstrate the feasibility of the proposal and their agreement with the ideally predicted results obtained by numerical simulation. We have shown that our system is robust against scratches produced by common friction or handling, though it can be sensitive to cuts of the ID tag caused by non-authorized manipulation.

The proposed system is an attractive tool for highly secure authentication tasks. The surveillance and tracking of classified parcels along with the identification of the person responsible for transferring the parcel has been shown as an example of a high-demanding security task. But this is not the only case. Other situations can also benefit from this encryption-verification system. For instance, this proposal can be also applied to the control of vehicle access to restricted areas. If there is a strict need of security, it would be necessary to identify both the vehicle as well as its driver for a given day and place. Several pieces of information such as a biometric of the driver along with the plate number and/or the tire pattern could be verified to increase security. The encrypted multifactor information would be placed in an ID tag. Distortions could appear if the ID tag readout is made at a distance. Thus, ID tags used to identify vehicles should also be read even though they were captured under some type of scale variation or rotation.