Finger vein segmentation method and device based on neural network and probability graph model

Abstract

The invention discloses a finger vein segmentation method and device based on a neural network and a probability graph model. The method comprises the following steps: generating a preprocessed imagebeneficial to neural network training and testing; generating a gold standard beneficial to neural network training; taking the dense conditional random field of the Gaussian pairwise potential as a recurrent neural network to perform average field approximate reasoning; and fusing the dense conditional random field into a neural network, and performing training processing of the neural network. According to the method, a dense conditional random field (CRF) with Gaussian pairwise potential can be used as a recurrent neural network (RNN) to establish average field approximate reasoning; coarseoutput of a traditional neural network (CNN) can be refined in the forward transmission process, meanwhile, errors are transmitted back to the neural network (CNN) in the training process, and aftertraining for a certain period of time, due to the fact that the advantages of a dense conditional random field (CRF) are fully utilized, finer vein segmentation can be generated.

Description

technical field [0001] The invention relates to the technical field of neural networks, in particular to a method and device for segmenting finger veins based on neural networks and probability graph models. Background technique [0002] In recent years, as people have higher and higher requirements for the security and accuracy of biometric systems, biometric technology has received more and more attention. As one of many biometric identification technologies, finger vein recognition has become a current research hotspot due to its advantages of non-contact acquisition, live detection, difficult forgery, and low cost. The segmentation of blood vessels in finger vein images is a key step in vein recognition technology, and the quality of the segmentation effect directly affects the precision and accuracy of subsequent recognition. [0003] The traditional CNN has a convolution filter with a large receptive field, and generating pixel-level labels after reconstruction will p...

AI Technical Summary

Problems solved by technology

For example, in semantic segmentation tasks, this can lead to unclear boundaries and blob-like shapes

Second, CNN lacks smoothness constraints that encourage label consistency between similar pixels, as well as spatial and appearance consistency of label outputs. Lack of such smoothness constraints results in poor object contours and small pseudo-regions in the segmentation output.

Images