Research On Multispectral Palmprint Recognition Band Selection Method

Biometrics is a secure identification and authentication technology using the inherent physical characteristics of the human body and the acquired behavior characteristics. Among many biometric characteristics, research on palmprint recognition began at the late 20th century and has achieved great development in the past decades. Along with the study in depth, multispectral technology which is also increasingly studied in biometric identification has been applied,called multispectral biometrics. Since the spectral reflection and absorption character of the skin, multispectral biometrics can acquire more information to do recognition. But the effects of different spectral combinations are different, thus we study the band selection method to find more effective spectral combination. The main works are as follows.Multispectral palmprint feature level fusion algorithm: Currently, texture analysis of palmprint is the main stream of palmprint recognition. One frequently used algorithm utilizes Gabor wavelet transform to extract texture feature. We use magnitude of palmprint returned by Gabor wavelets as the weight of extracted features to fuse the texture features of the palmprint under different spectral. Experiments results show that this fusion method is effective.Multispectral palmprint collection system: we have developed a multispectral palmprint collection system which covers spectral range of 420nm~1100nm.The highest spectral resolution is 1nm. We also acquire a medium-sized multispectral palmprint database for future research.Multispectral band selection Research: We used two algorithms to investigate multispectral band selection for palmprint recognition. One is using the improved branch and bound algorithm. First, principal component analysis is used to do preliminary band selection, and then use the branch and bound algorithm for selecting the expected number of the bands. These improve the operational efficiency of the branch and band algorithm, and reach an effective band selection effect. Second, the genetic algorithm is applied to band selection. Experiments showed that the genetic algorithm in multispectral palmprint band selection has good performance.