Palmprint Recognition Algorithm

Palmprint is studied as an important biometric technology in the field of identity authentication recently. Comparing with other biometrics, palmprint recognition has its significant advantages: It is easy to be acceptable and implemented, and high recognition correct rate and real-time recognition can be achieved. Palmprint has rich and obvious main line features, so it doesn't require high image resolution. Palmprint Acquiring camera is cheap and palmprint recognition system can be used in many fields widely. Therefore, it has important theoretical significance and higher practical value to study palmprint recognition algorithm.Palmprint recognition algorithms including palmprint segmentation, feature extraction and feature matching are studied in the paper. The paper's concrete works are shown as follows:1) In order to achieve perfect palmprint region and reduce palm position requirement, a palmprint segmentation algorithm is proposed based on geometric features. Firstly, five feature points can be located by using geometric shape of palm, and then palm region segmentation result can be achieved by processing these points. Acquiring procedure is more convenient and humanistic because the palm doesn't need to locate in fixed position. Perfect palmprint region which is helpful to palmprint recognition can be achieved.2) There are obvious main lines and textures in detail in palm region, they belong to multi-scale features, in order to describe palmprint pattern effectively, and a plamprint line-extraction algorithm is proposed based on wavelet transform in this paper. We can transform original palm region into several scale wavelet images, and then achieve binary palmprint features by using an self-adaptive threshold. Simulation results show that the palmprint feature extraction algorithm is good.3) A new algorithm named layer by layer matching is proposed to resolve the problem of multi-scale features match. The matching algorithm will calculate similarity of low-frequency features of plamprints firstly, and then we can determine whether the similarity of high-frequency features of plamprints will be calculated or not. The match method can improve recognition accuracy and real-time performance. The experiment results show that the classification correct rate is 97.72% based on the Chinese Academy of Sciences'palmprint database.