[Paper Review] Feature Level Clustering of Large Biometric Database

Hunny Mehrotra, Dakshina Ranjan Kisku+3 more|arXiv (Cornell University)|Feb 2, 2010

Face and Expression Recognition7 references3 citations

TL;DR

This paper proposes a fuzzy clustering approach at the feature level to efficiently partition large offline signature databases for biometric identification. By combining global and local descriptors with fuzzy c-means clustering, the method reduces search space more effectively than traditional k-means, achieving lower bin-miss rate in experiments.

ABSTRACT

This paper proposes an efficient technique for partitioning large biometric database during identification. In this technique feature vector which comprises of global and local descriptors extracted from offline signature are used by fuzzy clustering technique to partition the database. As biometric features posses no natural order of sorting, thus it is difficult to index them alphabetically or numerically. Hence, some supervised criteria is required to partition the search space. At the time of identification the fuzziness criterion is introduced to find the nearest clusters for declaring the identity of query sample. The system is tested using bin-miss rate and performs better in comparison to traditional k-means approach.

Motivation & Objective

• Address the challenge of efficiently indexing large biometric databases where features lack natural ordering.
• Reduce search space during identification by partitioning biometric data using clustering instead of sequential search.
• Improve identification accuracy by leveraging fuzzy clustering to handle uncertainty in feature similarity.
• Develop a scalable solution for large-scale biometric systems using feature-level clustering with global and local descriptors.
• Demonstrate superior performance over conventional k-means clustering in terms of bin-miss rate.

Proposed method

• Extract global and local descriptors from offline signature images to form a comprehensive feature vector.
• Apply fuzzy c-means clustering to group feature vectors into clusters based on similarity, allowing partial membership.
• Use the fuzziness criterion during identification to determine the most likely clusters for a query sample.
• Partition the database into clusters at the feature level to reduce the search space during identification.
• Optimize the clustering process to handle the high dimensionality and non-ordered nature of biometric features.
• Evaluate performance using bin-miss rate as the primary metric to compare with traditional k-means.

Experimental results

Research questions

• RQ1 How can large biometric databases be efficiently partitioned when features lack natural ordering?
• RQ2 Can fuzzy clustering at the feature level improve identification performance compared to traditional k-means?
• RQ3 What is the impact of using both global and local descriptors on clustering accuracy and search efficiency?
• RQ4 How does the fuzziness criterion enhance the identification process in ambiguous or overlapping feature regions?
• RQ5 To what extent does feature-level clustering reduce the bin-miss rate in large-scale biometric systems?

Key findings

• The proposed feature-level fuzzy clustering approach achieves a lower bin-miss rate than the traditional k-means method.
• Fuzzy clustering effectively handles the uncertainty and overlap in biometric feature space, improving identification robustness.
• The integration of global and local descriptors enhances the discriminative power of the feature vectors used in clustering.
• The clustering-based partitioning significantly reduces the search space during identification, improving scalability.
• The system demonstrates improved performance in large-scale biometric applications due to better cluster separation and membership assignment.
• The results confirm that fuzzy clustering is more suitable than crisp k-means for indexing biometric data with inherent feature ambiguity.