Two dimensional Fisher linear discriminantanalysis(2D-FLDA) is a very effective method for palmprintrecognition. However, it cannot be used when each objecthas only one training sample because the within-class scattermatrices cannot be calculated. In this paper, a novel methodis developed to solve this problem. Using the blocksegmentation, wavelet transform, and sampling methods, anew training set containing three training samples in eachclass can be obtained. Then the 2D-FLDA can be applied toextract the discriminant palmprint feature vectors. Finallythe pattern classification can be implemented by the nearestneighbor classifier. Experimental results on the PolyUpalmprint database show that the proposed method isefficient and it has better recognition performance thanmany existing schemes.
展开▼