An Adaptive Biometric System Based on Palm Texture Feature and LVQ Neural Network

摘要

We propose an adaptive biometric system based on the palm texture feature and LVQ2 neural network. The user's palm image is acquired by a scanner and preprocessed to be a labeled palm contour in the binary image format. Then, the positions of 12 feature points are identified speedily and roughly on the contour and refined to be more precise with a proposed correction mechanism. By referring the positions of feature points, six subimages of five fingers and the palm are obtained and transformed into six feature vectors with a modified texture descriptor of LFP (local fuzzy pattern). We employ the LVQ2 to learn the prototypes of feature vectors of each user. Therefore, an unknown user's palm feature vector is compared with prototypes to identify or verify his identity.