Iris Recognition Using Curvelet Transform Based on Principal Component Analysis and Linear Discriminant Analysis

The iris texture curve features play an important role in iris recognition. Although better performance in terms of recognition effectiveness can be attained using the recognition approach based on the wavelet transform, the iris curve singularity cannot be sparsely represented by wavelet coefficients. In view of the better approximation accuracy and sparse representation ability of the Curvelet transform, an iris recognition method based on Curvelet, Principal Component Analysis (PCA) and Linear Discriminant Analysis(LDA) is proposed. First, the iris image is preprocessed including iris localization, elimination of the eyelash shading and iris normalization. Next, the preprocessed iris image is decomposed into the N layers Curvelet coefficients by the Curvelet transform. The highest frequency coefficients are filtered directly because they mainly contain the false information caused by some environmental noises. And then the iris Curvelet features are mapped by PCA and LDA to extract the further features, in which we use the Curvelet coefficients of the former N-1 layers and only the first layer coefficients as the iris Curvelet features respectively. At last, the nearest neighbor classifier is adopted for iris recognition. Experimental results show that the iris recognition methods using the former N-1 layers Curvelet coefficients and only using the first layer Curvelet coefficients both can recognize iris effectively and get the higher recognition rate with minor difference. But the method only using the first layer Curvelet coefficients can effectively reduce the feature dimension and improve the recognition speed.