Knuckle Pattern Recognition Method Based on Improved U-net

The information society has increasingly higher requirements for the accuracy and diversity of identity authentication. Biometric identification is a reliable method of identification, which has been widely used in real life. As a new biometric identification method, the knuckle pattern has the advantages of rich texture and high stability and bright prospects in the research of identity recognition. The traditional segmentation method has shortcomings such as breakpoints and segmentation of non-main texture features for the segmentation of complex texture information, although it is effective and simple. A knuckle pattern segmentation method based on deep learning is proposed, which is based on improved U-net. The U-net model is improved in two ways, including adding an attention mechanism to the U-net network and changing the traditional convolution into a deep separable convolution in the non-additional attention mechanism module in the model. Finally, combine the Surf algorithm to identify and match. So improved the accuracy of recognition and reduced the amount of calculation. The experimental results show that the improved U-net segmentation method used in this paper has a further improvement in the segmentation effect compared with the traditional segmentation method, and the final recognition rate is also significantly improved. The recognition rate reached 97.43%, and the single image matching time was 0.12s.