Research On Fast Palmprint Recognition Based On Clustering

Palmprint recognition, as a newly emerged biometrics technology, has apromising market prospect and has been studied widely since its emergence. Afterdecades of investigation, palmprint recognition technology has become relativelymature in theory and now it’s going out of lab and stepping into its applicationstage.Nowadays, a palmprint template is matched against every enrolled palmprinttemplate in database sequentially during recognition. This method providesrecognition result fast enough when enrolled palmprint templates are of small scale.When used in large database, sequential matching can take an unacceptably longtime to produce the result, which makes it infeasible to database of large scale. Tomake palmprint recognition feasible to large database, new matching methods whichcan produce result in real-time must be provided.In this paper, we first introduce several methods that can be used to acceleraterecognition process and compare them with each other, including their advantagesand disadvantages. Then a palmprint fast recognition framework based-on clusteringis proposed. Two clustering algorithms，K-means and DBSCAN are studied in depthand improvements according to some of their drawbacks are provided. As toK-means, a new initializing method which can find initial cluster centers asseparated as possible is proposed. In order to overcome the drawback of DBSCANthat the algorithm is highly dependent on parameters Eps and MinPts, a method thatutilize Gaussian-means algorithm to estimate parameters Eps and MinPts forDBSCAN automatically is used. Experiments show that compared with thetraditional clustering algorithms, the corresponding improved ones yield betterresults. And the fast palmprint recognition system based-on K-means producesresult at least twice as fast as sequential matching and the accuracy is around97%.