The Research Of Palmprint And Palm Vein Fusion Algorithm Based On Feature-level

Multi-biometric identification based on information fusion becomes a hotspot in information security and biometrics research which makes up the lack of single-biometric like low security,low recognition rate and low user acceptance.The fusion of palmprint and palm vein has advantages like large amount of information,high security,simple acquisition and user acceptance.In addition,the analysis of fusion level shows that the fusion of feature level contains relatively abundant information and needs less amount of data to be processed compared with the image level,the recognition efficiency is higher than the matching level and decision level fusion,so the feature level fusion of palmprint and palm vein is researched in this thesis.Based on the summary of knowledge of information fusion and the characteristics of the palmprint and palm vein and its feature extraction algorithm,this thesis aims at the problem of the influence of noise on the fusion process of feature level and the increase of dimension after feature fusion will lead to the problem of reduced recognition rate and efficiency,the following research is carried out:(1)Aim at the influence of illumination and noise in the process of feature extract,a local directional pattern algorithm is proposed to extract the texture features of palmprint and palm veins The local direction pattern uses the 8 direction operator of Krisch template to calculate the eigenvalues of the center point,which overcomes the problem of non-uniform illumination sensitivity and anti-noise ability of traditional local binary patterns.The simulation results show that the recognition rate of palmprint recognition is average improved by 1.35%and the recognition rate of palm vein recognition is average improved by 1.50%compared with the traditional local binary pattern.Thus,the advantages of the local directional pattern extraction of palm and palm vein in terms of recognition rate are verified.(2)The improved Canonical Correlation Analysis algorithm is proposed for the palmprint and palm vein feature fusion.In the process of information fusion,the increase of feature vector dimension leads to the problem of "Dimension Disaster".Canonical Correlation Analysis can analyze the correlation of palmprint and palm vein features by analyzing a few pairs of typical variables to reduce the fusion process in the information redundancy,improve processing speed.Aim at the problem that the denominator of the traditional canonical correlation analysis method is difficult to obtain the minimum value,the correction coefficient is introduced to improve the criterion function,and the projection direction of the feature vector is adjusted to minimize the denominator of the criterion function,so that the most representative The characteristic fusion of palmprint and palm veins was performed by several typical correlation features.The experiment results show that:1)the improved Canonical Correlation Analysis improves the information fusion performance and improves the recognition rate compared with the original Canonical Correlation Analysis;2)the recognition effect of fusion the palmprint and palm vein feature based on improved Canonical Correlation Analysis has a better effect than single modal palmprint and palm vein.The recognition rate is improved by about 1.14%and 4.01%respectively.3)The receiver operating characteristic curve of this algorithm is located below the traditional Bayesian,series connection and parallel connection method ROC curve,which shows that the false rejection rate,false acceptation rate and equal error rate are of the improved Canonical Correlation Analysis lower than traditional fusion algorithm,so the fusion effect is more ideal.