Using brain prints as new biometric feature for human recognition
Introduction
Nowadays, person's identification and verification are increasingly used in different fields and different applications depending on the required level of security [27], [35], [40]. Among the most commonly used biometric modalities, one can obviously mention fingerprints [11], [18], palmprint [26], [52], [53], iris recognition [7], [13], [15], [25] and face recognition [10], [33]. These common modalities have been widely integrated in numerous systems and devices, which make them useful, but vulnerable regarding potential attacks. Within this context, the biometric community is very active to provide robust solutions in order to prevent identity spoofing. For example, some common spoofing concerns: fake fingerprints for which numerous publications have been dedicated for this purpose [6], [9], [17], [24]. One can also cite other attacked modalities such as: iris recognition, 2D/3D facial recognition, palm-print (including vein biometrics using near infrared). Even though, many interesting and promising advanced solutions have been proposed to overcome this issue, it becomes interesting to explore another type of biometric.
Recently, emerging biometric category is explored. Called hidden biometrics, the purpose is the fact that identification and verification processes are performed by extracting features from any part of human body; which is not directly accessible nor visible by naked eye [1], [38], [39]. Within this context, hidden biometrics may require some specific devices or equipment's that are commonly employed in the clinical and the medical field. For example, hidden biometrics can include bio-signals, such as EEG (Electroencephalogram), [5], [32], ECG (Electrocardiogram) [8], X-ray imaging, [30], [31], and MRI, etc.
In this paper we are exploring human brain images acquired from MRI. The purpose is to extract a unique brainprint from each 3D brain volumetric image to be used either for verification or identification. These images contain folds information, including a set of cortical and subcortical structures. Compared to some common biometric modalities, the proposed approach has a major advantage, which makes attacks/spoofing a difficult task to consider “No one can modify the features of his own brain”.
Based on some studies, it becomes interesting to deliver some answers to the following questions:
- •
Are cortical folding patterns unique to each individual?
- •
Are brains really asymmetric?
- •
Is the shape of individual brain structures heritable?
- •
Are cortical folds and brain shape stable?
Actually, it is known that human brain consists of white matter, gray matter and cerebrospinal fluid. The cerebral cortex is the outer layer of neural tissue of the cerebrum of the human brain. It is separated into the left and the right hemispheres by the longitudinal fissure. The cerebral cortex is folded, giving a much greater surface area in the confined volume of the skull. A fold or ridge in the cortex is called a gyrus (plural gyri) and a groove or fissure is called a sulcus (plural sulci). In the human brain, more than two-thirds of the cerebral cortex is buried in the sulci. The gyri and sulci called sulco-gyral patterns form the brainprint.
The growth of the brain in the skull provides a unique signature. In this context, many theories have addressed the issue of brain folds origin [50]. The arguments were classified schematically into two types:
- 1.
The first theory considers that only micro external mechanical forces form the brain folds. Indeed, the brain as a resilient organ expands within the skull which is rigid organ and limited in size. The resulting mechanical tensions allow stationary, stable and unique sulco-gyral patterns.
- 2.
The second theory assumes the existence of a relationship between the morphology and the final structure of the brain, on one hand, and its cyto-architectonic and functional organization, on the other hand.
One can conclude that both mechanical tension (occurring during brain growth) and experience that we learn allows the cortical and sub-cortical patterns to develop in distinctly different ways [42], [50]. This shows that brain folds are discernibly different among individuals. Furthermore, recent studies in genetic and neuroscience show that the jumping genes, which provide identical twins being different, may also influence the brain folds and shape [22]. All these studies show that the human brain morphology is unique. Moreover, studies of brain morphometric and some work on the cerebral asymmetry have shown that even a number of brain sulci (main sulci) are present in all individuals. These differ in their positions, shapes and numbers of components. Indeed, the same sulcus met in several individuals may be more or less profound, more or less long, more or less sinuous and can also be broken and composed of multiple distinct furrows [43]. Although, both right and left hemispheres are very similar in size, shape and weight, but the distribution of brain tissues (white and gray matter) and the outline of brain folds are not the same and are different from one hemisphere to the other. Hence, both hemispheres of the same individual are not symmetrical. Thus, two brains will never have the same aspect, even in identical twins [23], [48], [49]. Consequently, brain folds and the sulco-gyral patterns are specific to each individual, which makes all human brains, anatomically different.
Different studies on the anatomical variability of the brain related to aging and cerebral diseases have shown a change in the brain gray matter, but have confirmed its stability and steadiness in adulthood [21], [34]. In addition, this variability affects mainly morphometry and the brain's shape like volume, and the weight of cortical structures but it does not affect the sulco-gyral patterns. Therefore, the main question that arises within the context of our work is as follows: is it possible to use brain folds as biometric traits for verification/identification routines?
Few research attempts have been considered in the feasibility of using brain images for security biometrics. Objectively, one can distinguish two types of approaches; namely, those based on texture analysis and those based on shape and structure analysis. Regarding the first approach, Aloui et al. [2], [4], used a single slice from an MR volumetric image. In particular, they extracted textural features, from 2D images containing cortical folds, using 1D Log-Gabor transform. Using a similar algorithm as the one employed for iris recognition [16], a binary template called Braincode was generated. Afterward, Hamming distance was used for template matching. The major drawback of this approach is that a single MR slice acquired at a given distance is used for verification. Considering a single MR slice may provide less reliable descriptors.
Regarding the second approach, shape information extracted from a set of cortical and subcortical structures has been considered. Aloui et al. [3], [4], used a single slice from an MR volumetric image. In particular, brain shape descriptors are extracted from of an ellipse that circumscribes the brain. Afterward, features vector describing the brain shape is generated. For the matching phase, a Range-Normalized Euclidean Distance (RND) has been used for similarity measurement. In another work, Chen et al. [12] implemented brain segmentation algorithm in order to extract gray matter from an input brain image. Then, an alignment based matching algorithm was developed for brain matching. Takao et al. [47] performed brain recognition using voxel-based morphometric approach for image normalization. Principal Component Analysis (PCA) was used for features extraction. For the matching phase, Euclidean distance between image pairs projected into the subspace is calculated.
Different studies show that, performances of shape-based methods are highly sensitive to segmentation quality and also to the Region of Interest (ROI) analysis techniques. Furthermore, it appears that considering shape representation has the disadvantage of being very sensitive to the inter-subject variabilities of the shape and location of the brain structures, and also to the changes that may be caused by different scanners and acquisition protocols.
To prevent the disadvantages of shape-based methods evoked above, we investigate in this work, a new representation cerebral cortex from which textural features are extracted and used as brainprints. In particular, our approach considers curvilinear slices that allow planar views of the full cortex, including its cortical folds. This new characterization differs from the previous single planar slice that is limited to only a small part of the cortex.
This paper is organized as follows: In Section 2, we present a general scheme of the proposed modality. Brainprint feature extraction and characterization is detailed in Section 3. In Section 4, preliminary results over MR Datasets are presented, analyzed and objective performance evaluation is reported. Finally, a conclusion of this research is provided in Section 4.
Section snippets
General methodology
Our approach requires six processing phases. A conceptual framework of the present work is shown in Fig. 1. For some phases, technical details will be provided in the next sections.
- 1.
Brain image acquisition: due to its high resolution and non-requirement of radiative contrast medium, structural brain MR images are used.
- 2.
Curvilinear slices extraction: using 3D volumetric brain scans, curvilinear slices are extracted. After being projected as two-dimensional images, a planar view of the cortex,
Enrolment and cortical brain characterization
In this section we explain in detail the main stages of our approach given by the conceptual framework shown in Fig. 1.
Experimental results
To evaluate our new biometric approach based human brainprint recognition, experiments are conducted on OASIS data sets (Open Access Series of Imaging Studies) [36] providing volumetric brain MR images. We have chosen OASIS database because it offers MR brain images for healthy and adult individuals and for each we have 4 different acquisitions. We note that limited available databases propose MR brain images for healthy individuals, including multiple acquisitions. OASIS database is composed
Conclusion
Through this research, we introduced a new approach to extract a robust biometric signature (i.e. brainprint), from MR brain images. As we saw, the brainprint is computed by mapping curvilinear slices (that include the cerebral cortex), into 2D images. This new brain folds representation produces symmetrical views of the cortex and allows low inter-variabilities. Feature extraction was achieved through Gabor wavelet transform from projected curvilinear slices. As result, we achieved a correct
Acknowledgments
The authors gratefully acknowledge the support of University of Tunis El Manar and University of Paris-Est. We acknowledge the use of OASIS database provided by the Washington University Alzheimer's Disease Research Center, Dr. Randy Buckner at the Howard Hughes Medical Institute (HHMI) at Harvard University, the Neuroinformatics Research Group (NRG) at Washington University School of Medicine, and the Biomedical Informatics Research Network (BIRN).
References (54)
- et al.
Iris recognition by local extremum points of multiscale Taylor expansion
Pattern Recognit.
(2009) - et al.
Fusion of local normalization and Gabor entropy weighted features for face identification
Pattern Recognit.
(2014) - et al.
Designing Gabor filters for optimal texture separability
Pattern Recognit.
(2000) - et al.
Vulnerabilities of fingerprint reader to fake fingerprints attacks
Forensic Sci. Int.
(2011) - et al.
Fingerprint matching using ridges
Pattern Recognit.
(2006) Mean value coordinates
Comput. Aided Geom. Des.
(2003)- et al.
Lateral asymmetry of neurotransmitters in human brain
Brain Res.
(1982) - et al.
Palmprint verification based on principal lines
Pattern Recognit.
(2008) - et al.
Fronts propagating with curvature-dependent speed: algorithms based on Hamilton–Jacobi formulations
J. Comput. Phys.
(1988) - et al.
Brain morphology is individual-specific information
Magn. Reson. Imaging
(2015)
Robust palmprint verification using 2D and 3D features
Pattern Recognit.
Image contrast enhancement by constrained local histogram equalization
Comput. Vision Image Understanding
Caractérisation Du Cerveau humain : Application à La Biométrie (Ph.D. thesis)
A novel approach based brain biometrics: Some preliminary results for individual identification
New biometric approach based on geometrical humain brain patterns recognition: Some preliminary results
A new useful biometrics tool based on 3D brain human geometrical characterizations
J. Signal Inf. Process.
Fusion of physiological measures for multimodal biometric systems
Multimed. Tools Appl.
Fake fingerprint detection by odor analysis
A novel biometric authentication approach using ECG and EMG signals
J. Med. Eng. Technol.
Security evaluation of biometric authentication systems under real spoofing attacks
IET Biom.
Hierarchical minutiae matching for fingerprint and palmprint identification
IEEE Trans. Image Process.
The potential of using brain images for authentication
Sci. World J.
Non-orthogonal view iris recognition system
IEEE Trans. Circuits Syst. Video Technol.
How iris recognition works
IEEE Trans. Circuits Syst. Video Technol.
Uncertainty relation for resolution in space, spatial frequency, and orientation optimized by two-dimensional visual cortical filters
J. Opt. Soc. Am. A
Surface parameterization: a tutorial and survey
Attachment stability from infancy to adulthood: meta-analysis and dynamic modeling of developmental mechanisms
Pers. Soc. Psychol. Rev.
Cited by (11)
Sharing individualised template MRI data for MEG source reconstruction: A solution for open data while keeping subject confidentiality
2022, NeuroImageCitation Excerpt :However, though the face is easily the most identifiable feature, it is not the only feature of an MRI that can be used to identify individuals (Amico and Goñi, 2018; Byrge and Kennedy, 2019). The brain's anatomy is unique to each person: the folding pattern of the cortex is identifiable, just like a person's fingerprint, and could therefore be used to deanonymise research participants if that brain fingerprint could be linked to another database (Aloui et al., 2018). Sharing the anatomical MRIs for MEG source analysis is subject to the ethical consideration of sharing person sensitive information and the trade-off between replicability and participant confidentiality: on one end, share all data with the original MRI—on the other end, share only the MEG data and a template head model or leadfield.
A comprehensive survey on the biometric recognition systems based on physiological and behavioral modalities
2020, Expert Systems with ApplicationsCitation Excerpt :Table 14 presents the major findings in a summarized way with the Ear based biometric recognition system. Aloui, Ali, and Naceur (2018) presented a novel and successful concept in the biometric recognition systemon the brain print trait that depends upon the sulcogyral patterns and brain-fold of an individual which was used for the age identification and then the classification of the gender. They developed a new non-invasive technique that explored a brain cortical region of MRI (Medical Resonance Imaging) based on a dataset of 220 persons.
A novel octopus based Parkinson's disease and gender recognition method using vowels
2019, Applied AcousticsCitation Excerpt :Finally, matching or machine learning methods are used to identify the people [4]. The widely used physical and behavioral features used in biometric systems are fingerprint, face, iris, retina, palm, voice, gait, signature [3,5–8]. Voice recognition is one of the widely used system because voice based biometric systems are cheap and effective.
Biometric Recognition in 3D Medical Images: A Survey
2023, IEEE AccessAutomated gender-Parkinson's disease detection at the same time via a hybrid deep model using human voice
2022, Concurrency and Computation: Practice and Experience