Keywords

1 Introduction

The biometrics systems are used for access control and identification of human beings, and they are based on different physiological measures such as physical traits (PT), physiological signals, Deoxyribonucleic acid (DNA), among others. This type of identity recognition is very attractive since each person possesses different physical features that cannot be copied easily [5]. Nowadays, it is applied widely to assure computers, smartphones, communication systems, buildings, and confidential information, among others. However, the multiple techniques of individual identification had become vulnerable to falsification such as the identification system based on digital fingerprint [1, 2], which has been used for several years, but it can be falsified with different methods [4], putting at risk the legal and financial integrity of an individual.

Although some physical features are hard to imitate/duplicate, it is not impossible. Therefore, different researchers have proposed the fusion or combination of multiple physiological signals (PS) and traits with the goal of providing major sturdiness to the system [6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34]. However, the biometric is an open research area focused on the type and quantity of the data, algorithms, and functionality modes and they are classified by 3 categories of biometric modalities as follows: (i) biological, it is based on the analysis of data obtained from DNA; (ii) behavioral, this is based on the analysis of the behavior of the individual; and (iii) morphology data, they are based on specific physical features that are permanent and unique to every individual (e.g. face or fingerprints) [1].

In this paper, we discuss PS and traits applied to biometry authentication together with different combinations among them (i.e. multiple signals and multiple traits) using data fusion techniques. The review was carried out on Scopus and Web of Sciences database based on these search criteria: (i) (biometric) and (“physiological signals”); and (ii) ((“data fusion”) or (“information fusion”) and (biometric)) or (“physiological signals”). The selected papers were reported between years 2008 and 2017 in journals of quartile 1 and quartile 2 principally.

2 Physiological Signals and Traits Applied in Biometrics Systems

Nowadays, the use of PS has gone from being used only by medical diagnostics, to convert into a very important tool for security demonstrating the capability to provide characteristics that allow identified an individual with high precision.

The PS must comply with a series of criteria that is apt in biometric. The criteria are the following: (i) the signal must be able to be recollected in any person; (ii) singularity, the signal must able to distinguish different individuals; (iii) permanence, the signal must not be abruptly altered in the time; (iv) sturdiness against attacks, it must not be imitated easily [35].

Figure 1 shows a summarized taxonomy of the traits and PS reported for biometric authentication and/or identification.

Fig. 1.
figure 1

Taxonomy of physiological signals and physiological traits applied in biometric.

Full size image

Table 1 presents the results of the review regarding the use of the signals and traits including the type of modality. 56 biometric studies based on unimodal and multimodal modalities (i.e unimodal is the application only one signal or trait and multimodal is the application of two or more signals and PT). Multiple signals and traits have been studied independently (monomodal authentication or identification) such as: Electroencephalogram-EEG, Electrocardiogram-EKG, phonocardiography-PCG, electrooculography-EOG, Electromyogram-EMG, photo-plethysmography-PPG, fingerprint, palmprint, periocular, Laser Doppler Vibrometry-LDV, Speech, Finger Knuckle Print-FKP, finger vein, tongue, Iris, face, ear, lips, eyes, gait, and Knee Acceleration-knA. Other studies are based on combination or fusion of multiples signals together with multiple traits (multimodal authentication or identification) are evidenced.

Table 1. Monomodal and multimodal biometric studies
Full size table

2.1 Physiological Signals Applied in Biometrics

The EKG and PCG are noninvasive measures and they take heart information, particularly the EKG takes information about heart electrical activity. This signal is acquired situating electrodes in the thoracic zone, with the purpose to collect the signals produced by myocardium, while the PCG signals are based on the analysis of the features of the frequency of the cardiac sounds, these sounds are presented in systole (S1) and diastole (S2) [30, 31]. The condition of the atria, ventricles and heart valves among other, each of these characteristics mentioned are different for each individual when both characteristics are used at the same time we get a lot of information from the heart. In [31] is fused both types of signals obtaining a lower error rate in comparison to the error obtained with the individual signals.

The EEG signals are noninvasive too, and they are usually recorded from the surface of scalp [5]. These signals are split into five frequency bands as follows: Delta (δ) 0.5–4 Hz, Theta (θ) 4–8 Hz, Alpha (α) 8–14 Hz, Beta (β) 14–30 Hz y Gamma (γ) more of 30 Hz [6]. EEG signals can be a great option for biometric identification due to that the brain electrical activity is unique in each individual and besides closely related with the visual, mood, auditory stimuli and in general any stimulus experienced by the person. Therefore is necessary to consider that the cerebral response to any of these stimuli is different for each individual causing that EEG signals are difficult of supplant and get, therefore this signals are practical in biometrics.

Others signals have been less reported in the biometric area such as EOG signals, which consists in the registration of potential difference existing between cornea and retina for ocular movements detecting [55, 56]. Although it is possible to get relevant information about an individual, it is very difficult to implement because it is uncomfortable for the participant since the electrodes located on both sides of the eyes, above and below of these collect the potentials generated from the movement of eyeballs [56]. The EKG, EEG, EOG and PCG have very important characteristics for their use in biometrics since they are not easily accessible and also provide reliable information of the individual.

2.2 Physiological Trait Applied to Biometrics

Figure 2 shows images of the physiological traits reported in the literature, even so, have characteristics quite promising in the recognition of people. These images correspond from left to right to Finger knuckle [17], finger vein [46], lips [44], tongue [47], metacarpophalangeal [57], and gait recognition [48] respectively. The more popular trait nowadays is the fingerprint. It is highly used in personal authentication by the well-known fact that each individual has a unique fingerprint and its acquisition highly easy and cheap [58]. Fingerprint refers to the patterns located on the fingertips. On the other hand, the hands have a lot amount of folds in the knuckles which are used in recognition, this method is named finger knuckle (FKP) and it is based on capture of image around surface of phalangeal joint of the finger [18, 59], whereas, the identification based on patterns in the Metacarpophalangeal joint (MPJs), consists in the obtaining of patterns of the rear surface of the hand, given that this zone presents many lines and folds that allow discriminating a person of other. The MJP recognition offers a promising and robust alternative for authentication of identity [57], although the advantage of these methods is the high quantity of information available, such as lines, forms, and patterns, which are different for each individual, a difficulty in FKP and MJP recognition are the false rejects due to the variation of finger knuckle position on the take of the image [17].

Fig. 2.
figure 2

Physiological traits applied in biometrics

Full size image

In [46] was proposed a verification system using the finger vein as a biometric feature in response to high vulnerability registered in biometric systems based on the fingerprint. Finger vein recognition consists of locating a beam in the finger, which makes visible the veins in this zone. The patterns of the veins are considered an interesting trait, given that this patterns not easily affordable and different for each person. Blood’s temperature and volume, and the incorrect positioning of the surface to analyze affect this feature.

Recently, the gait recognition has been studied in biometric, it is based on gait biomechanics to extract own features of each person. One of its advantages more marked is the possibility of getting hundreds of samples of the cycle of gait in few minutes. This feature is significant given that the success of the design and validation of a pattern recognition system depend to a large degree of the sample size, but also has some challenges such as the clothes of the subject, sensibility to environment variations, angle camera capture and distance between the subject and the camera, which makes the gait recognition systems a difficult task to carry out in the real environment [48,49,50].

Another biometric trait is the lips, given that is an important feature of the human face. The lips features are geometric simple features based on the contour of these, they are can interpret as multiple spatial thickness and height from of mass center [44]. To be a trait captured by images it presents the same difficulty than previous features, as the incorrect positioning of the area of interest. Also, comes the question of whether it is possible to get a good rate of recognition when the lips are in movement.

The tongue is a trait, which has been very little studied but has features very interesting in biometric identification; it has used dynamic and static features obtained from the tongue. Within of the dynamic features are the texture, geometry, thickness and cracks. All obtained from the image but as is well known all system is exposed to attacks and eventually the system could not be able to differentiate image of a living person and a dead. Therefore, the researchers have raised the need for performing the detection of vitality with the last purpose of securing that the input patterns are not coming from an inanimate object. The dynamic features refer to obtaining of patterns related continuous movement and involuntary of the tongue, is an excellent dynamic firm for biometric. Thanks to all these features, the tongue is used as a physiological trait in biometric [47]. Finally, we had summarized the advantages and disadvantages of PS and traits applied to biometric in Table 2.

Table 2. Traits and physiological signals advantages and disadvantages
Full size table

3 Multimodal Systems

The human interaction is considered a natural multimodal process and contains deep physiological and psychological expressions. A multimodal biometric system combines two or more signals or traits [65], during years the authentication has been based on monomodal biometric systems, which compare only one characteristic, however, the performance of these systems vary depending on the presence of external factors, such as noise, computational cost, and the quality devices of signal acquisitions. Therefore, in the ultimate years has been introduced the multimodal biometric system with the purpose to overcome the weaknesses of the monomodal biometric system.

3.1 Architecture of Multimodal Processing Systems

In [66] is presented a description of the architecture of the multimodal system (see Fig. 3) Once has been determined the different biometric sources, the next step to follow is the selection of the architecture of the system. In general, there are two mine type of design of multimodal system, serial and parallel

Fig. 3.
figure 3

Multimodal system architecture [66]

Full size image

(i) serial: Into the serial architecture, also called as cascade architecture, the signal processing is performed in sequence. Therefore the out of first biometric feature influence the transmission second feature; (ii) parallel: Into the parallel architecture, the processing of several biometric inputs are independent of one another. Once both signals are processed separately, the results are combined.

4 Data Fusion Systems

Data fusion has notion rather fuzzy that take various interpretations with the applications and specific purposes [67]. However, in this paper the definition adopted in [68] as a set methodologies and technology that possibility the combination synergistic of heterogeneous data of several sources together with new data, content more information than the sum of each source. Through several terms have been reported in the literature such as: decision fusion [69], data combination [70], data aggregation [71] multisensor integration [72], multisensor data fusion [73], and information fusion [74]. In spite that these terms describe the same task, with some variations in terms of application and the type of data that can be difficult to differentiate, which is discussed by [75], Those who used interchangeably term data fusion and information fusion.

Data fusion is considered a very challenging task for several reasons: (i) the complexity of data; (ii) the processes depend on n variables without being all measurable; (iii) in heterogeneous data sets is hard to exploit the advantages of each set and discard the disadvantages [67]. For data fusion are used different techniques such as, the probabilistic, soft-computing, algorithm optimization, among others, whose use (characterization, estimation, aggregation, classification, compression among others) It depends on the type of application and also is necessary consider the advantages and disadvantages of techniques for a proper selection in order to get an effective performance. The Fig. 3, presents the taxonomy of the methodologies of data fusion from which can be categorized data fusion algorithms and are widely discussed in [76].

Particularly in the case of biometrics, the data fusion is highly used a level of signals fusion, level characteristics fusion and level classifiers fusion allowing to improve the performance the identification systems using several PS as reported in [77–79].

Table 3 shows the biometric signals and their respective techniques which got characterization and subsequent classification.

Table 3. Fusion methods used in biometrics
Full size table

5 Conclusion

This review described advantages, disadvantages, and shortcomings of PS and traits oriented to biometric, when they are mixed the identification error rate can be reduced. Until now, an ideal biometric signal that meets criteria of high security, easy acquisition, low computational cost and that the user feels comfortable during the process has been not achieved yet.

In general, the unimodal systems compared to the multimodal systems, these last report less percentage of error. EEG, EOG, PCG, and EKG, among others, they are considered highly promising in the biometric for different authors since they have a fairly small mistake and it ensures the identified subject this alive. Nevertheless, the accessibility to them is very restricted, but in turn involves high-tech equipment due to the complex acquisition, and eventually generating discomfort to the user. In addition, some problems must be solved how analysis of signals with pathologies, noise ratio of the signals, improving the acquisition devices in together with the development of sensors with special characteristics.

Other physiological biometric parameter or of behavior can be fused to do the authentication more dependable [9]. Although the present work shows that there is great potential in the used of PS to the biometric recognition, is important that the future analysis is performed with a grouping largest set of signals. Respect tot he PT, these have the nowadays challenges of the images recognition such as angle of capture of the image. Besides some recognized methods don’t detect if the subject is live. Therefore, studies on areas of the body in movement are necessary. Finally, We pose as challenges the study of other biometric techniques such identification bio-inspired and data fusion architectures for PS and traits processing oriented to biometric authentication.