Mitsubishi Electric Corporation has developed a multi-modal biometrics authentication system with convenience and flexibility that can recognize a person by freely combining several biometrics, e.g., fingerprints, face, signatures, and so on. We have also developed a high-speed ID-less data-searching algorithm that enables personal identification using only biometrics without having to input an ID number.

Background and Outline of Development
As social consciousness towards information security has increased recently, personal authentication systems using passwords are at risk of having them stolen.
Biometrics methods such as fingerprints have attracted attention as a way of personal confirmation able to simultaneously improve security and convenience.

However, it has been difficult for a biometrics method to achieve an authentication accuracy of 100% by itself. Therefore, many biometrics methods have been accessorily combined with the conventional password authentication procedure to avoid such cases. In addition, each biometrics method has adequate applications according to its accuracy and other features. Because of this, it has been necessary to select a kind of biometrics for each application, and to construct a personal authentication server for each application.

To date, there have been systems performing personal authentication by combining multiple biometrics methods. However, biometrics methods and combined judgment operations have been fixed for these systems, and there have not been systems capable of flexibly changing between/among biometrics methods and combined judgment operations.

Our developed multi-modal biometrics authentication system overcomes these problems. With one personal authentication server, it is possible to perform personal authentication by freely combining multiple types of biometrics methods and combined judgment operations for each authentication.

Main Features and Results
a. Multi-modal biometrics authentication technology able to freely select biometrics methods and combined judgment operations

This technology makes it possible to freely select the biometrics methods and combined judgment operations according to the preferences of the user and the necessary authentication accuracy. The technology eliminates the need to construct a personal authentication server for every application and enables flexible correspondence for the authentication demands from various clients.

b. High-speed ID-less data searching algorithm applicable to various biometrics methods

To find the data of a person without the use of his/her ID number, it has typically been necessary to verify N/2 times on average for a registration count of N. However, when this high-speed ID-less data searching algorithm is used for fingerprints (as an example), it is possible to detect the data of a registered person that agrees with the input data by verifying an average of 45 times from among 2000 fingerprints.

Application Fields
The multi-modal biometrics authentication technology is broadly applicable to physical access control systems in buildings, information security systems such as the Internet and intranets.

Future Developments
On the basis of the developed technology, we will continue the technical development of combined judgment operations to further improve highly convenient and authentication accuracy. In addition, we plan on advancing technical development in order to proceed to apply the technology to information security systems and human identification systems on the Internet and intranets.

Contents of Development
The technical features are described as follows:
(1) Multi-modal biometrics authentication technology can freely select biometrics methods and combined judgment operations
At present, various personal authentication technologies via biometrics are being developed. There are various characteristics in every biometrics method and each biometrics method has fitting applications. Because of this, it has been necessary from a conventional viewpoint to install a personal authentication server for each application, and to separately carry out authentication processing. To address this problem, we have integrated multiple biometrics methods into one personal authentication server this time. In addition, we have developed a multi-modal biometrics authentication technology able to carry out personal authentication by freely combining multiple biometrics methods and combined judgment operations.

With this multi-modal biometrics authentication system, it is possible to perform various combined judgment operations besides AND/OR for multiple biometrics methods. With this system, it is possible to list up the high-possibility candidates for each biometrics method, and to perform order-based combined judgments to comprehensively judge the candidate lists in the combined judgment operation stage. For example, there are few cases where the authentication results obtained from fingerprint and face data differ, when high-speed ID-less data searching is carried out for each set of biometric data, respectively, and combined judgment operations are performed from the obtained results. In this way, various combined judgment operations are possible. Besides these, it is also possible to add various combined judgment operations easily.

By using this multi-modal biometrics authentication technology, an authentication client selects the ideal single or multiple biometrics method(s), and requests the user to input the necessary biometrics data. The authentication client sends to the personal authentication server the obtained biometrics data and the combined judgment operation equations selected by the client. The personal authentication server carries out authentication processing with the combined judgment operations specified from the client, and returns the authentication results.

(2) High-speed ID-less data searching algorithm applicable to various biometrics methods

In the past, the general practice was for a user to input his/her ID number and then input his/her biometrics data via a biometrics sensor. This time, we have developed a high-speed ID-less data-searching algorithm that finds the data of a person in an instant from a high volume of registered data, without the input of an ID number.

For example, a linear search method that checks the registered data in order to find data that agrees with the input biometrics data requires verifying of N/2 times on average from among N biometrics data. Classification methods have been studied to classify all the data into several groups in order to reduce N. With classification methods, however, there have been a number of problems. For instance, it has been necessary to newly develop a classification strategy for every biometrics method and there have been an increasing number of searches with an increasing amount of registered data due to the use of linear search within the groups.

Accordingly, this time, we have developed a data-searching algorithm that can be applied to various biometrics methods and that can find the data of a person in an instant without performing classification. Assuming evaluation with fingerprint data, the algorithm makes it possible to attain the data of a person with an extremely small N value, i.e., 45 times on average (an acceleration of 22 times compared with a linear search method) from registered data including 2000 fingerprints. (The processing time for fingerprint data is about 200 milliseconds with a 600MHz Pentium II3 *1PC.)
*1: Pentium II3 is a registered trademark of Intel Corporation, U.S.A.

With this data-searching algorithm, reference matching scores with all other registered data are calculated during the registration of biometrics data. This algorithm achieves a fast speed searching for the data of the target person by referring to the degrees of similarity in the series of the reference matching scores with the data of the other people. The algorithm is applicable to various biometrics methods because it is a method that accelerates the search process by only using reference matching scores between biometrics data.

Now, it is supposed 10 data is registered. In the enrollment phase, matching scores with all the combination of 10 registered data are calculated, and 10*10 reference matching score table can be obtained. Here, matching score means the index value that indicates the similarity of the two biometrics data. A matching score takes between 0 and 100, and 0 and 100 means that two biometrics data are absolutely different and identical respectively.
Patents
Received one in Japan; submitted applications for eight in Japan and one overseas.

back