Abstract: Aiming at the problems that palmprint acquisition was influenced by external factors and noise, palmprint recognition rate of traditional method was low and robustness was poor, we proposed a new palmprint recognition algorithm based on feature weighted and kernel principal component analysis. Firstly, palmprint image was decomposed by Curvelet transform, contour coefficients of different scales and angles were obtained, and Curvelet coefficients were weighted by the fusion operation. Secondly, feature extraction was realized by using kernel principal component analysis to reduce dimension of palmprint feature. Finally, relevance vector machine was used to realize palmprint matching, and the performance of the algorithm was tested by using PolyU palmprint image. The test results show that, compared with other palmprint recognition algorithms, the proposed algorithm has higher palmprint recognition rate, and palmprint matching time is the shortest, which can meet the requirements of realtime palmprint recognition.

Key words: relevance vector machine, palmprint recognition, feature dimension reduction, kernel principal component analysis, Curvelet transform