Research On Recognition Method Of Finger Biometrics Based On GCN

Finger biometric has always occupied a large proportion in the field of biometric recognition by the advantage of its high portability.Both single-model recognition and multi-model fusion are highly attractive for the science research.However,there are often some problems such as the low recognition efficiency,the excessive fusion dimensions,and the differences in feature expressions bothering researchers a lot.Taking finger vein,fingerprint,and finger knuckle print as research objects,this paper proposes a method of finger biometric recognition based on graph convolutional network.Firstly,a single-model recognition method based on graph convolutional network is proposed.The images are mapped into graph and the features of these three models are expressed by the k-NN weighted graph.Then a recognition method based on GCN is built.The graph convolution kernel based on Chebyshev polynomial is used and the fast graph pooling algorithm is added in order to reduce the dimension of the feature.Secondly,two finger multi-model fusion recognition methods are proposed.One is the fusion of the nodes' features when the image is mapped into the graph and the fusion graph structure is established based on the distance of the nodes' feature.The second method is the fusion of decision level in the GCN.We get the fusion probability matrix from three models to complete the recognition.Aiming at the above methods,the experiments about single-model recognition are taken firstly.The results prove that the recognition method based on GCN can get much higher recognition accuracy and efficiency than the traditional algorithm.Then the multi-model fusion methods are explored and compared.The experimental results show that the proposed fusion methods have a great advantage in recognition efficiency.Among them,the fusion method based on nodes' features has higher recognition accuracy.