Finger Vein Indexing using Unsupervised Clustering
MetadataShow full item record
Finger vein recognition systems have gained more popularity recently, and it is one of the most reliable biometric modality. Finger veins are present inside the finger, and its pattern is not known which makes it hard to spoof. In addition, it’s known to produce high accuracy, permanence and reliable for person identification/ recognition. Because of these advantages, finger vein recognition has already been implemented in financial sectors of Brazil, China, Japan and Poland. With the increased use of Biometric systems for large scale application like Unique ID project (AADHAAR) in India and border control projects, there is a need for indexing methods to narrow down the search space in large scale databases. So, It is essential to have a good indexing scheme for finger vein as widely used in real life scenarios. As of now, there is only one method based on the local sensitive hashing (LSH) to search in large-scale databases. The main limitations of LSH are, it depends on the choice of hash functions and memory required to store the hash tables will grow according to the size of the database. It is not a feasible solution to use for real life scenarios where, database scaling is required. In this work, we presented finger vein indexing and retrieval schemes based on unsupervised clustering. To this extent, we investigated three different clustering schemes namely K-means, K-medoids with binary & real-valued features and self organizing maps (SOM) neural networks with real-valued features. The proposed schemes are experimentally verified with the large scale heterogeneous finger vein database comprised of 2850 unique identities constructed using seven different finger vein databases. The obtained results demonstrated the efficacy of the proposed schemes for large scale finger vein applications in real life scenarios.