Contactless palmprint recognition based on convolutional neural network

Abstract

Palmprint recognition has become the novel and effective technology in biological recognition fields. The contactless palmprint recognition is the mainstream way with its unique advantages. However, hand pose variations, rotations, translations, complicated backgrounds are the common problems in contactless palmprint recognition. To solve these problems, this paper proposes a palmprint recognition approach based the convolutional neural network (CNN), which consists of three main parts, namely, image preprocessing, CNN feature extraction and matching. First palmprint images are preprocessed using the improved fuzzy enhancement algorithm, then using the AlexNet with eight layers network structure for palmprint feature extraction. Finally match the feature with the hausdorff distance. The results of the experiments on three public available databases in different circumstances indicate that the proposed method achieves the best equal error rate (EER) 0.044% with respect to some typical methods of the state of the art.