The finger vein recognition based on curvelet

Multiscale analysis is a focus of image processing in recent years, among which wavelet has received a great deal of progress and achievement, but the singularity information extracted using wavelet is point singularity, and the representation based on wavelet has too much redundancy. Therefore, in the comparative experiments we propose two types of algorithm using ridgelet, which can show the line singularity of finger vein image, the recognition performance is satisfactory. But there always exist a larger number of curves in ordinary image, in this case, we try using curvelet for the feature extraction. In the second section, two discrete algorithms based on USFFT and Wrapping for curvelet is introduced, the former has a higher reliability and the latter is easier to be implemented. Considering that the finger vein image is still filled with burr and truncation after the preprocessing, so we choose USFFT coefficients in different scale to represent different types of image information. In this paper we use corresponding processing algorithm for the feature in different scales, finally all the modified features are linked to an integrated feature. The classifiers are SVM and NN respectively, the recognition performance is fairly well.