I am not convinced that ECG or EEG signals can be used for a reliable high-end biometric systems. The signals are i) just to easily influenced by non-obvious parameters, ii) to difficult to get reliable/reproducible access to and iii) to cumbersome to acquire in huge numbers for a data base....

ad i) just think even with a clinically attached multi-lead ECG on the variations in heartbeat during a day...

ad ii) Influences are e.g. mood, health status, electrode contact, positioning.... All these parameters are influencing the measurements. Completely different from a finger print!

ad iii) a data base must contain a huge number of recognizable/labeled data sets. I somehow would not very happily spend hours with providing EEG signals under many different conditions....

Please enlighten me, if you have a way to overcome this reliability issues.

There is certainly a lot of potential in using EEG and ECG for biometric purposes. The main issues are in the cumbersome data collection and on the stability issues. For example, fingerprints can be easily obtained using a finger-scanner and the finger scans don't change over time.

EEG and ECG bio metrics are consider for high end applications only............

I am not convinced that ECG or EEG signals can be used for a reliable high-end biometric systems. The signals are i) just to easily influenced by non-obvious parameters, ii) to difficult to get reliable/reproducible access to and iii) to cumbersome to acquire in huge numbers for a data base....

ad i) just think even with a clinically attached multi-lead ECG on the variations in heartbeat during a day...

ad ii) Influences are e.g. mood, health status, electrode contact, positioning.... All these parameters are influencing the measurements. Completely different from a finger print!

ad iii) a data base must contain a huge number of recognizable/labeled data sets. I somehow would not very happily spend hours with providing EEG signals under many different conditions....

Please enlighten me, if you have a way to overcome this reliability issues.

I have been thinking a lot about the articles floating around that show, 1) retina scan, 2) fingerprint scan, 3) ECG-related, 4) voice, ...

Here is my feeling: individually, every one of these will have a detection failure rate, or, conversely, DETECTION ACCURACY. The only way to guarantee good accuracy is to use MULTIPLE metrics, such as, voice+fingerprint. Because probabilistically, if the accuracy of the first one is p1 percent (e.g., 90%), and the other one is p2 percent (e.g., 80%), the failure rate goes from 10% (i.e., 1-90%) or 20% (i.e., 1-80%) for individual detections down to 2% (i.e., 10% * 20%) for a MULTIPLE detection. Intuitively, it is very hard to MISDETECT somebody's voice AND fingerprint simultaneously. I am doing research in this area.

I am convinced that, ECG signals have a) person-specific, and b) generic (i.e., non-person-specific) components ... just like other bio-signals ... This is not different than voice. A person's voice is used for SPEECH RECOGNITION (i.e., non-person-specific), and SPEAKER RECOGNITION (i.e., person-specific). Same can be done for ECG with proper modeling, since everybody's heartbeat's will have some unique signature (just my feeling, not referring to any research).

The real question is: When will we come up with an ALGORITHM to model this ?

Easy sensors for EEG-ECG are being developed like the Enobio sensor (http://www.neuroelectrics.com/enobio), which we have used for testing our multimodal EEG-ECG system. For an exemple of such systems using soft data fusion operators see:

Fusion operators for multi-modal biometric authentication based on physiological signals

A Soria-Frisch, A Riera, S Dunne

http://www.researchgate.net/publication/221358973_Fusion_operators_for_multi-modal_biometric_authentication_based_on_physiological_signals/file/79e415023d1cd8d873.pdf

We have recently improved the performance of the system reaching Equal Error Rate of the multi-modal system around 3%. So I would say the potential usage of such modalities is big.

Conference Paper Fusion operators for multi-modal biometric authentication ba...

I agree with Ulrich. Variations of physiological signals with time and a person's state are a non-trivial challenge. Thanks Aureli for your interesting work. However, if I understand correctly, you compare the ability of ECG and EEG to differ between subjects on one occasion. It would be interesting how the informative characteristics of these sensor modalities vary over time, i.e. how stable they are over changes in context. Are you aware of any such research that looks at the person-specific characteristics of ECG/EEG over different recording sessions?