Palmprint retrieval based on match scores and decision-level fusion

This paper proposes a new indexing mechanism for biometric databases based on match scores between the images. First, the proposed approach computes a match score to each image by comparing it against a preselected representative image. The computed match score of the image against the representative acts as its key in the database. The proposed approach uses these keys (i.e., match scores) to arrange the biometric images in sorted order like traditional database records so that a fast search is possible. During identification, this approach computes the key of the query (i.e., its match score against the representative image) and retrieves a set of images that have same key in the database as similar to it. We experimented this approach with multiple representatives and extended to multi-modal biometrics. This approach enrolls the users dynamically without disturbing the existing system. Experimental results on benchmark PolyU palmprint database show a significant performance improvement of the proposed system compared to the existing retrieval methods.