Research on Denoising of Finger Vein Image Based on Deep Convolutional Neural Network
Chaoping Zhu, Yang Yong-bin, Yun Jang · 2019
Because of finger vein collection device and using habit, the collected finger vein images contain noise, the extraction of finger vein features may cause errors. For the problem, a kind of finger vein image denoising method is proposed based on the deep convolution neural network, which is to collect the finger vein image as the input image of network, the non-linear mapping of the noise image to the denoising image is constructed through the hidden layer, a symmetric network structure is formed by the convolution subnet and the deconvolution subnet, convolution subnet learns finger vein image characteristics, the deconvolution subnet restores the original image according to the feature graph, and the modified linear unit is used to obtain more details of finger vein texture. The experiment was conducted as a training set in the finger vein database of the polytechnic university of Hong Kong, use Tensorflow model to train network model in GPU environment. The experimental results show that this method can remove the noise in the finger veins, the higher peak signal-to-noise ratio is obtained, which lays the foundation for the subsequent feature extraction.