Finger vein recognition based on local graph structural coding and CNN

In recent years, deep learning has received an excellent performance in the tasks of image feature extraction and image classification. Besides, the coding-based methods have been widely focused on because of their outstanding local description. In this paper, we propose a novel method for finger-vein recognition, which combines local coding and convolution neural network (LC-CNN). Based on local graph structure (LGS), a weighted symmetrical LGS is firstly proposed to locally represent the gradient relationship among the surrounding pixels. Then, the traditional local coding methods are reconstructed with a set of fixed sparse predefined binary convolution filters. To address the over-fitting of the network, we use the local coding convolution to alter standard convolution in pre-trained CNN. Finally, the extracted feature vector are input into a support vector machine (SVM) for images classification. Experimental results show that the proposed approach achieves better performance than the traditional coding methods on finger vein recognition.