Iris Recognition Based on DLDA

Abstract

Iris feature extraction is very important for an iris recognition system. This paper focuses on iris feature extraction. In this paper we propose direct linear discriminant analysis (DLDA) which combines with wavelet transform to extract iris feature. In our method, firstly, we apply wavelet decomposition to the normalized iris image whose size is 64times256 and just choose the coefficients of the approximation part of the second level wavelet decomposition to represent the iris image because this part contains main feature of the original iris image but the size of this part is only 16times64. And then make use of DLDA to extract the iris feature from this approximation part. During classification, the Euclidean distance is applied to measure the similarity degree of two iris classes. In the end of this paper, the proposed method was tested on the second version CASIA iris database. We evaluate the performance by equal error rate (EER) which is the point that the false match rate (FMR) is equal to false non-match rate (FNMR) in valve. The experiment shows that the EER of our method is about 1.44% which is lower than other methods such as principle component analysis (PCA) and independent component analysis (ICA) etc