Levels of Fusion in Multimodal Biometric Systems

The previous articles on Multimodal biometrics had focused on data sources such as multi-sensor, multi-algorithm, multi-instance and multi-traits. This article will focus on the levels of fusion. The amount of information available to a biometric system drastically decreases as one proceeds from the sensor module to the decision module (matching module). In a multibiometric system, fusion can be accomplished by utilizing the information available in the sensor module, feature module, score module and decision module. The levels of fusion can be broadly classified as fusion prior to matching and fusion after matching.

Fusion prior to matching: This is the fusion of data prior to matching. The integration of information from multiple biometric sources can take place either at the sensor level or at the feature level. The raw data from the various sensors can be integrated or combined in sensor level fusion. Sensor level fusion can only be employed if the multiple sources represent samples of the same biometric characteristics obtained either using a single sensor or different compatible sensors. For example, 2D face images of an individual obtained from several cameras can be combined to form a 3D model of the face. While Feature level fusion refers to the integration of different feature sets extracted from multiple biometric sources. Biometric features used in the feature level fusion can be divided into two set, homogeneous and non-homogeneous. When the feature sets are homogeneous (e.g., multiple measurements of the same biometric characteristic such as a person’s hand geometry), a single resultant feature vector can be calculated as a weighted average of the individual feature vectors. While if the feature sets are non-homogeneous (e.g., features of different biometric modalities such as face and hand geometry), the features can be concatenated into a single feature vector. Feature selection schemes are employed to reduce the dimensionality of the ensuing feature vector. Concatenation is not possible when the feature sets are incompatible.

Fusion after matching: This is the fusion of data post matching. The integration of information after the matcher stage can be divided into four categories: dynamic classifier selection, fusion at the decision level, fusion at the rank level and fusion at the match score level. A dynamic classifier selection scheme chooses the results of that biometric source which is most likely to give the correct decision for the specific input pattern. This is also known as the winner-take-all approach within the biometrics world. Integration of information at the decision level can take place when each biometric system independently makes a decision about the identity of the user (in an identification system) or determines if the claimed identity is true or not (in a verification system). Since most commercial biometric systems provide access to only the final decision output by the system, fusion at the decision level is often the only viable option. When the output of each biometric system is a subset of possible matches (i.e., identities) sorted in decreasing order of confidence, the fusion can be done at the rank level. This is relevant in an identification system where a rank may be assigned to the top matching identities. When each biometric system outputs a match score indicating the proximity of the input data to a template, integration can be done at the match score level.

More information on the implementation of biometrics based solutions can be requested from info@namibiabiometricsystems.com.