Utilizing automatic quality selection scheme for multi-modal biometric fusion

With the growing understanding of quality's impact on recognition systems, a large number of quality factors with different quality measurements are available to the biometric system. In order to avoid the risk to reach the `curse of dimensionality' by increasing the number of quality measures in the multimodal fusion system, an automatic quality selection scheme is proposed in this paper to automatically select `appropriate' quality measures to effectively improve the fusion performances. The proposed quality selection scheme evaluates quality factors by examining their impacts on the corresponding matching system, and then selects certain quality factors for each modality that have major contributions to the system performance. A better solution of utilizing those selected quality measures is then proposed and implemented into quality-based fusion strategies: incorporating both gallery and probe quality measures into the multi-modal fusion systems instead of transformed `pair-wise' quality score. Experimental results demonstrate the proposed quality selection scheme can offer an optimal way to maximize the improvement of fusion performance using limited number of `appropriate' quality factors, and potentially analyze a large number of quality measures. The extensive experiments also provide a unique view to evaluate fusion performances by looking into more detailed quality groups of gallery and probe data, providing better understandings of the quality-based fusion system performance.