Design And Implementation Of Fingerprint Recognition System Under Massive Dataset By Distributed Computing

Currently, fingerprint recognition technology has been greatly improved. Excellent fingerprint recognition algorithm was published constantly. Most of them only concerned about how to improve the matching accuracy while ignoring the cost of matching time. However, in daily life we often need to do the fingerprint recognition in massive data. Because most of the existing technology did not in considering the issue of massive data, this led to that the real-time identification with massive fingerprints is a challenging task. The main problems are: 1) We need a solution to build index with fingerprint features in order to speed up fingerprint recognition; 2) The computing power with single computer is limited, which can’t satisfy the situation of searching with massive fingerprints; 3) If we use the distributed implements, how to do the recognition fully concurrently thus returning result fast is a key point.In consideration of these problems, this thesis refers to the existing generic content-based image retrieval solutions and comes up with a scheme which deals with massive fingerprint images and realizes the processes how to construct fingerprint index and retrieval rapidly. Firstly this thesis introduces the existing fingerprint feature descriptors and discusses the merit and demerit of them. Then it describes and accomplishes the Minutia-Cylinder-Code(MCC) algorithm. The fingerprint feature descriptor built with this algorithm is a 1280 bit binary code, which has the best recognition performance in the existing technologies. Then it introduces the solution how to construct index with binary-valued features. After comparison of the traditional LSH based method and k-means tree based method, we chose the multi hierarchical k-means tree algorithm to build index with binary-valued features. This algorithm performs well in both retrieval accuracy and retrieval time. Lastly we deploy the whole solution under distributed environment based on MapReduce processes. This solution realizes parallel computing with multi-nodes and provides real-time searching interface which improves the efficiency of fingerprint searching observably.Experimental results show that the performance of the MCC feature descriptor is pretty good, and the multi hierarchical k-means tree algorithm has better retrieving time and efficiency compared with traditional LSH algorithm. At the same time, the introduction of distributed computing model solves the problem of building index with massive fingerprints and real-time searching.